AMDGPU: Stop reserving 36-bytes before kernel arguments for amdpal

[arsenm]

Summary

This was inheriting the mesa behavior, and as far as I know nobody is
using opencl kernels with amdpal. The isMesaKernel check was
irrelevant because this property needs to be held for all functions.

[sebastian-ne]

Under the assumption that opencl or hip on pal do not use this, it looks fine to me.

[arsenm]

89c447e4e6b564636bbf32b15d67e40cf6c60387

llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h

public:
 
   Align getAlignmentForImplicitArgPtr() const {
     return isAmdHsaOS() ? Align(8) : Align(4);
   }
 
   /// Returns the offset in bytes from the start of the input buffer
   ///        of the first explicit kernel argument.
   unsigned getExplicitKernelArgOffset(const Function &F) const {
-    return isAmdHsaOrMesa(F) ? 0 : 36;
+    switch (TargetTriple.getOS()) {
+    case Triple::AMDHSA:
+    case Triple::AMDPAL:
+    case Triple::Mesa3D:
+      return 0;
+    case Triple::UnknownOS:
+    default:
+      // For legacy reasons unknown/other is treated as a different version of
+      // mesa.
+      return 36;
+    }
+
+    llvm_unreachable("invalid triple OS");
   }
 
   /// \returns Maximum number of work groups per compute unit supported by the
   /// subtarget and limited by given \p FlatWorkGroupSize.
   virtual unsigned getMaxWorkGroupsPerCU(unsigned FlatWorkGroupSize) const = 0;
 
   /// \returns Minimum flat work group size supported by the subtarget.
   virtual unsigned getMinFlatWorkGroupSize() const = 0;

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.div.fmas.ll

; GFX10_W64-NEXT:    ; return to shader part epilog
   %vcc = icmp eq i32 %d, 0
   %result = call double @llvm.amdgcn.div.fmas.f64(double %a, double %b, double %c, i1 %vcc)
   ret double %result
 }
 
 define amdgpu_kernel void @test_div_fmas_f32(float addrspace(1)* %out, [8 x i32], float %a, [8 x i32], float %b, [8 x i32], float %c, [8 x i32], i1 %d) {
 ; GFX7-LABEL: test_div_fmas_f32:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s2, s[0:1], 0x13
-; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x1c
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x25
-; GFX7-NEXT:    s_load_dword s5, s[0:1], 0x2e
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dword s2, s[0:1], 0xa
+; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x13
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x1c
+; GFX7-NEXT:    s_load_dword s5, s[0:1], 0x25
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX7-NEXT:    s_and_b32 s2, 1, s5
 ; GFX7-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    s_nop 1
 ; GFX7-NEXT:    v_div_fmas_f32 v0, v0, v1, v2
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x4c
-; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x70
-; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x94
-; GFX8-NEXT:    s_load_dword s5, s[0:1], 0xb8
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x28
+; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x4c
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x70
+; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x94
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX8-NEXT:    s_and_b32 s2, 1, s5
 ; GFX8-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX8-NEXT:    s_nop 3
 ; GFX8-NEXT:    v_div_fmas_f32 v2, v0, v1, v2
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32:
 ; GFX10_W32:       ; %bb.0:
 ; GFX10_W32-NEXT:    s_clause 0x4
-; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0xb8
-; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x70
-; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x94
-; GFX10_W32-NEXT:    s_load_dword s7, s[0:1], 0x4c
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x4c
+; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x70
+; GFX10_W32-NEXT:    s_load_dword s7, s[0:1], 0x28
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    s_and_b32 s0, 1, s4
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W32-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, s7, v0, v1
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32:
 ; GFX10_W64:       ; %bb.0:
 ; GFX10_W64-NEXT:    s_clause 0x4
-; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0xb8
-; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x70
-; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x94
-; GFX10_W64-NEXT:    s_load_dword s7, s[0:1], 0x4c
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x4c
+; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x70
+; GFX10_W64-NEXT:    s_load_dword s7, s[0:1], 0x28
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    s_and_b32 s0, 1, s4
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W64-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX10_W64-NEXT:    v_div_fmas_f32 v0, s7, v0, v1
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W64-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W64-NEXT:    s_endpgm
   %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 %d)
   store float %result, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f32_inline_imm_0(float addrspace(1)* %out, [8 x i32], float %a, [8 x i32], float %b, [8 x i32], float %c, [8 x i32], i1 %d) {
 ; GFX7-LABEL: test_div_fmas_f32_inline_imm_0:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s2, s[0:1], 0x1c
-; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x25
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x2e
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dword s2, s[0:1], 0x13
+; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x1c
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x25
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    s_and_b32 s2, 1, s4
 ; GFX7-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    s_nop 1
 ; GFX7-NEXT:    v_div_fmas_f32 v0, 1.0, v0, v1
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32_inline_imm_0:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x70
-; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x94
-; GFX8-NEXT:    s_load_dword s4, s[0:1], 0xb8
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x4c
+; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x70
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX8-NEXT:    s_and_b32 s2, 1, s4
 ; GFX8-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX8-NEXT:    s_nop 3
 ; GFX8-NEXT:    v_div_fmas_f32 v2, 1.0, v0, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32_inline_imm_0:
 ; GFX10_W32:       ; %bb.0:
 ; GFX10_W32-NEXT:    s_clause 0x3
-; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0xb8
-; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x94
-; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x70
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x70
+; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x4c
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    s_and_b32 s0, 1, s4
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W32-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, 1.0, s6, v0
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32_inline_imm_0:
 ; GFX10_W64:       ; %bb.0:
 ; GFX10_W64-NEXT:    s_clause 0x3
-; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0xb8
-; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x94
-; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x70
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x70
+; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x4c
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    s_and_b32 s0, 1, s4
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W64-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX10_W64-NEXT:    v_div_fmas_f32 v0, 1.0, s6, v0
 ; GFX10_W64-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W64-NEXT:    s_endpgm
   %result = call float @llvm.amdgcn.div.fmas.f32(float 1.0, float %b, float %c, i1 %d)
   store float %result, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f32_inline_imm_1(float addrspace(1)* %out, float %a, float %b, float %c, [8 x i32], i1 %d) {
 ; GFX7-LABEL: test_div_fmas_f32_inline_imm_1:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s2, s[0:1], 0xb
-; GFX7-NEXT:    s_load_dword s3, s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x16
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dword s2, s[0:1], 0x2
+; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0xd
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    s_and_b32 s2, 1, s4
 ; GFX7-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    s_nop 1
 ; GFX7-NEXT:    v_div_fmas_f32 v0, v0, 1.0, v1
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32_inline_imm_1:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x2c
-; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x34
-; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x58
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x8
+; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x10
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x34
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX8-NEXT:    s_and_b32 s2, 1, s4
 ; GFX8-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX8-NEXT:    s_nop 3
 ; GFX8-NEXT:    v_div_fmas_f32 v2, v0, 1.0, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32_inline_imm_1:
 ; GFX10_W32:       ; %bb.0:
 ; GFX10_W32-NEXT:    s_clause 0x3
-; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x58
-; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x34
-; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x2c
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x34
+; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x10
+; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x8
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    s_and_b32 s0, 1, s4
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W32-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, s6, 1.0, v0
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32_inline_imm_1:
 ; GFX10_W64:       ; %bb.0:
 ; GFX10_W64-NEXT:    s_clause 0x3
-; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x58
-; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x34
-; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x2c
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x34
+; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x10
+; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x8
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    s_and_b32 s0, 1, s4
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W64-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX10_W64-NEXT:    v_div_fmas_f32 v0, s6, 1.0, v0
 ; GFX10_W64-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W64-NEXT:    s_endpgm
   %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float 1.0, float %c, i1 %d)
   store float %result, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f32_inline_imm_2(float addrspace(1)* %out, [8 x i32], float %a, [8 x i32], float %b, [8 x i32], float %c, [8 x i32], i1 %d) {
 ; GFX7-LABEL: test_div_fmas_f32_inline_imm_2:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s2, s[0:1], 0x13
-; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x1c
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x2e
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dword s2, s[0:1], 0xa
+; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x13
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x25
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    s_and_b32 s2, 1, s4
 ; GFX7-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    s_nop 1
 ; GFX7-NEXT:    v_div_fmas_f32 v0, v0, v1, 1.0
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32_inline_imm_2:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x4c
-; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x70
-; GFX8-NEXT:    s_load_dword s4, s[0:1], 0xb8
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x28
+; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x4c
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX8-NEXT:    s_and_b32 s2, 1, s4
 ; GFX8-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX8-NEXT:    s_nop 3
 ; GFX8-NEXT:    v_div_fmas_f32 v2, v0, v1, 1.0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32_inline_imm_2:
 ; GFX10_W32:       ; %bb.0:
 ; GFX10_W32-NEXT:    s_clause 0x3
-; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0xb8
-; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x70
-; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x4c
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x4c
+; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x28
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    s_and_b32 s0, 1, s4
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W32-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, s6, v0, 1.0
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32_inline_imm_2:
 ; GFX10_W64:       ; %bb.0:
 ; GFX10_W64-NEXT:    s_clause 0x3
-; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0xb8
-; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x70
-; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x4c
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x4c
+; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x28
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    s_and_b32 s0, 1, s4
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W64-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX10_W64-NEXT:    v_div_fmas_f32 v0, s6, v0, 1.0
 ; GFX10_W64-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W64-NEXT:    s_endpgm
   %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float 1.0, i1 %d)
   store float %result, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f64(double addrspace(1)* %out, double %a, double %b, double %c, i1 %d) {
 ; GFX7-LABEL: test_div_fmas_f64:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x9
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x11
+; GFX7-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x0
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x8
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s8
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s10
 ; GFX7-NEXT:    s_and_b32 s0, 1, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s9
 ; GFX7-NEXT:    v_mov_b32_e32 v5, s11
 ; GFX7-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX7-NEXT:    s_nop 3
 ; GFX7-NEXT:    v_div_fmas_f64 v[0:1], v[0:1], v[2:3], v[4:5]
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f64:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x24
-; GFX8-NEXT:    s_load_dword s0, s[0:1], 0x44
+; GFX8-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x0
+; GFX8-NEXT:    s_load_dword s0, s[0:1], 0x20
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s8
 ; GFX8-NEXT:    v_mov_b32_e32 v4, s10
 ; GFX8-NEXT:    s_and_b32 s0, 1, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX8-NEXT:    v_mov_b32_e32 v3, s9
 ; GFX8-NEXT:    v_mov_b32_e32 v5, s11
 ; GFX8-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX8-NEXT:    s_nop 3
 ; GFX8-NEXT:    v_div_fmas_f64 v[0:1], v[0:1], v[2:3], v[4:5]
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX8-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f64:
 ; GFX10_W32:       ; %bb.0:
 ; GFX10_W32-NEXT:    s_clause 0x1
-; GFX10_W32-NEXT:    s_load_dword s2, s[0:1], 0x44
-; GFX10_W32-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dword s2, s[0:1], 0x20
+; GFX10_W32-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    s_and_b32 s0, 1, s2
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v0, s8
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v2, s10
 ; GFX10_W32-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, s9
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v3, s11
 ; GFX10_W32-NEXT:    v_div_fmas_f64 v[0:1], s[6:7], v[0:1], v[2:3]
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10_W32-NEXT:    global_store_dwordx2 v2, v[0:1], s[4:5]
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f64:
 ; GFX10_W64:       ; %bb.0:
 ; GFX10_W64-NEXT:    s_clause 0x1
-; GFX10_W64-NEXT:    s_load_dword s2, s[0:1], 0x44
-; GFX10_W64-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dword s2, s[0:1], 0x20
+; GFX10_W64-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    s_and_b32 s0, 1, s2
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v0, s8
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v2, s10
 ; GFX10_W64-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, s9
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v3, s11
 ; GFX10_W64-NEXT:    v_div_fmas_f64 v[0:1], s[6:7], v[0:1], v[2:3]
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10_W64-NEXT:    global_store_dwordx2 v2, v[0:1], s[4:5]
 ; GFX10_W64-NEXT:    s_endpgm
   %result = call double @llvm.amdgcn.div.fmas.f64(double %a, double %b, double %c, i1 %d)
   store double %result, double addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f32_cond_to_vcc(float addrspace(1)* %out, float %a, float %b, float %c, i32 %i) {
 ; GFX7-LABEL: test_div_fmas_f32_cond_to_vcc:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_cmp_eq_u32 s7, 0
 ; GFX7-NEXT:    s_cselect_b32 s2, 1, 0
 ; GFX7-NEXT:    s_and_b32 s2, 1, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX7-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_nop 2
 ; GFX7-NEXT:    v_div_fmas_f32 v0, v0, v1, v2
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32_cond_to_vcc:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x8
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_cmp_eq_u32 s7, 0
 ; GFX8-NEXT:    s_cselect_b32 s2, 1, 0
 ; GFX8-NEXT:    s_and_b32 s2, 1, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX8-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX8-NEXT:    s_nop 3
 ; GFX8-NEXT:    v_div_fmas_f32 v2, v0, v1, v2
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32_cond_to_vcc:
 ; GFX10_W32:       ; %bb.0:
 ; GFX10_W32-NEXT:    s_clause 0x1
-; GFX10_W32-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x8
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    s_cmp_eq_u32 s7, 0
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W32-NEXT:    s_cselect_b32 s0, 1, 0
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX10_W32-NEXT:    s_and_b32 s0, 1, s0
 ; GFX10_W32-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, s4, v0, v1
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32_cond_to_vcc:
 ; GFX10_W64:       ; %bb.0:
 ; GFX10_W64-NEXT:    s_clause 0x1
-; GFX10_W64-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x8
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    s_cmp_eq_u32 s7, 0
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX10_W64-NEXT:    s_cselect_b32 s0, 1, 0
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX10_W64-NEXT:    s_and_b32 s0, 1, s0
 ; GFX10_W64-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX10_W64-NEXT:    v_div_fmas_f32 v0, s4, v0, v1
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W64-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W64-NEXT:    s_endpgm
   %cmp = icmp eq i32 %i, 0
   %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 %cmp)
   store float %result, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f32_imm_false_cond_to_vcc(float addrspace(1)* %out, [8 x i32], float %a, [8 x i32], float %b, [8 x i32], float %c) {
 ; GFX7-LABEL: test_div_fmas_f32_imm_false_cond_to_vcc:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s2, s[0:1], 0x13
-; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x1c
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x25
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dword s2, s[0:1], 0xa
+; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x13
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x1c
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b64 vcc, 0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX7-NEXT:    v_div_fmas_f32 v0, v0, v1, v2
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32_imm_false_cond_to_vcc:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x4c
-; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x70
-; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x28
+; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x4c
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x70
 ; GFX8-NEXT:    s_mov_b64 vcc, 0
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX8-NEXT:    v_div_fmas_f32 v2, v0, v1, v2
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32_imm_false_cond_to_vcc:
 ; GFX10_W32:       ; %bb.0:
 ; GFX10_W32-NEXT:    s_clause 0x3
-; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x70
-; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x94
-; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x4c
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x4c
+; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x70
+; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x28
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W32-NEXT:    s_mov_b32 vcc_lo, 0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, s6, v0, v1
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32_imm_false_cond_to_vcc:
 ; GFX10_W64:       ; %bb.0:
 ; GFX10_W64-NEXT:    s_clause 0x3
-; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x70
-; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x94
-; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x4c
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x4c
+; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x70
+; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x28
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W64-NEXT:    s_mov_b64 vcc, 0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10_W64-NEXT:    v_div_fmas_f32 v0, s6, v0, v1
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W64-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W64-NEXT:    s_endpgm
   %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 false)
   store float %result, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f32_imm_true_cond_to_vcc(float addrspace(1)* %out, [8 x i32], float %a, [8 x i32], float %b, [8 x i32], float %c) {
 ; GFX7-LABEL: test_div_fmas_f32_imm_true_cond_to_vcc:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s2, s[0:1], 0x13
-; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x1c
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x25
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dword s2, s[0:1], 0xa
+; GFX7-NEXT:    s_load_dword s3, s[0:1], 0x13
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x1c
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b64 vcc, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX7-NEXT:    v_div_fmas_f32 v0, v0, v1, v2
 ; GFX7-NEXT:    s_mov_b32 s2, -1
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32_imm_true_cond_to_vcc:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x4c
-; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x70
-; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x94
+; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x28
+; GFX8-NEXT:    s_load_dword s3, s[0:1], 0x4c
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x70
 ; GFX8-NEXT:    s_mov_b64 vcc, -1
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX8-NEXT:    v_div_fmas_f32 v2, v0, v1, v2
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32_imm_true_cond_to_vcc:
 ; GFX10_W32:       ; %bb.0:
 ; GFX10_W32-NEXT:    s_clause 0x3
-; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x70
-; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x94
-; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x4c
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dword s4, s[0:1], 0x4c
+; GFX10_W32-NEXT:    s_load_dword s5, s[0:1], 0x70
+; GFX10_W32-NEXT:    s_load_dword s6, s[0:1], 0x28
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W32-NEXT:    s_mov_b32 vcc_lo, -1
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, s6, v0, v1
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32_imm_true_cond_to_vcc:
 ; GFX10_W64:       ; %bb.0:
 ; GFX10_W64-NEXT:    s_clause 0x3
-; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x70
-; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x94
-; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x4c
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dword s4, s[0:1], 0x4c
+; GFX10_W64-NEXT:    s_load_dword s5, s[0:1], 0x70
+; GFX10_W64-NEXT:    s_load_dword s6, s[0:1], 0x28
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W64-NEXT:    s_mov_b64 vcc, -1
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10_W64-NEXT:    v_div_fmas_f32 v0, s6, v0, v1
 ; GFX10_W64-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W64-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10_W64-NEXT:    s_endpgm
   %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 true)
   store float %result, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f32_logical_cond_to_vcc(float addrspace(1)* %out, float addrspace(1)* %in, [8 x i32], i32 %d) {
 ; GFX7-LABEL: test_div_fmas_f32_logical_cond_to_vcc:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX7-NEXT:    s_load_dword s8, s[0:1], 0x15
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x0
+; GFX7-NEXT:    s_load_dword s8, s[0:1], 0xc
 ; GFX7-NEXT:    s_mov_b32 s2, 0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_mov_b64 s[0:1], s[6:7]
 ; GFX7-NEXT:    buffer_load_dword v3, v[1:2], s[0:3], 0 addr64 glc
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)

 ; GFX7-NEXT:    s_and_b64 vcc, vcc, s[0:1]
 ; GFX7-NEXT:    s_mov_b64 s[6:7], s[2:3]
 ; GFX7-NEXT:    v_div_fmas_f32 v0, v3, v4, v1
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[4:7], 0 offset:8
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32_logical_cond_to_vcc:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x54
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[0:1], 0x30
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 2, v0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s7
 ; GFX8-NEXT:    v_add_u32_e32 v1, vcc, v1, v3
 ; GFX8-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; GFX8-NEXT:    v_add_u32_e32 v3, vcc, 4, v1
 ; GFX8-NEXT:    v_addc_u32_e32 v4, vcc, 0, v2, vcc

 ; GFX8-NEXT:    v_div_fmas_f32 v2, v1, v2, v3
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32_logical_cond_to_vcc:
 ; GFX10_W32:       ; %bb.0:
-; GFX10_W32-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x0
 ; GFX10_W32-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
-; GFX10_W32-NEXT:    s_load_dword s0, s[0:1], 0x54
+; GFX10_W32-NEXT:    s_load_dword s0, s[0:1], 0x30
 ; GFX10_W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    global_load_dword v2, v1, s[6:7] glc dlc
 ; GFX10_W32-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10_W32-NEXT:    global_load_dword v3, v1, s[6:7] offset:4 glc dlc
 ; GFX10_W32-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10_W32-NEXT:    global_load_dword v4, v1, s[6:7] offset:8 glc dlc
 ; GFX10_W32-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    s_cmp_lg_u32 s0, 0
 ; GFX10_W32-NEXT:    s_cselect_b32 s0, 1, 0
 ; GFX10_W32-NEXT:    s_and_b32 s0, 1, s0
 ; GFX10_W32-NEXT:    v_cmp_ne_u32_e64 s0, 0, s0
 ; GFX10_W32-NEXT:    s_and_b32 vcc_lo, vcc_lo, s0
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, v2, v3, v4
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[4:5] offset:8
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32_logical_cond_to_vcc:
 ; GFX10_W64:       ; %bb.0:
-; GFX10_W64-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x0
 ; GFX10_W64-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
-; GFX10_W64-NEXT:    s_load_dword s0, s[0:1], 0x54
+; GFX10_W64-NEXT:    s_load_dword s0, s[0:1], 0x30
 ; GFX10_W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    global_load_dword v2, v1, s[6:7] glc dlc
 ; GFX10_W64-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10_W64-NEXT:    global_load_dword v3, v1, s[6:7] offset:4 glc dlc
 ; GFX10_W64-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10_W64-NEXT:    global_load_dword v4, v1, s[6:7] offset:8 glc dlc
 ; GFX10_W64-NEXT:    s_waitcnt vmcnt(0)
; GFX10_W64-NEXT:    s_endpgm
   %result = call float @llvm.amdgcn.div.fmas.f32(float %a, float %b, float %c, i1 %and)
   store float %result, float addrspace(1)* %gep.out, align 4
   ret void
 }
 
 define amdgpu_kernel void @test_div_fmas_f32_i1_phi_vcc(float addrspace(1)* %out, [8 x i32], float addrspace(1)* %in, [8 x i32], i32 addrspace(1)* %dummy) {
 ; GFX7-LABEL: test_div_fmas_f32_i1_phi_vcc:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x13
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xa
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX7-NEXT:    s_mov_b32 s6, 0
 ; GFX7-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    buffer_load_dwordx3 v[1:3], v[1:2], s[4:7], 0 addr64
-; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 s6, 0
 ; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX7-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX7-NEXT:    s_cbranch_execz .LBB13_2
 ; GFX7-NEXT:  ; %bb.1: ; %bb
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x1d
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x14
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_cmp_lg_u32 s0, 0
 ; GFX7-NEXT:    s_cselect_b32 s6, 1, 0
 ; GFX7-NEXT:  .LBB13_2: ; %exit
 ; GFX7-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX7-NEXT:    s_and_b32 s0, 1, s6
 ; GFX7-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s0
 ; GFX7-NEXT:    s_mov_b32 s6, -1
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    s_nop 1
 ; GFX7-NEXT:    v_div_fmas_f32 v0, v1, v2, v3
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    buffer_store_dword v0, off, s[4:7], 0 offset:8
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_div_fmas_f32_i1_phi_vcc:
 ; GFX8:       ; %bb.0: ; %entry
-; GFX8-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
+; GFX8-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x28
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 2, v0
 ; GFX8-NEXT:    s_mov_b32 s6, 0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX8-NEXT:    v_add_u32_e32 v1, vcc, v1, v3
 ; GFX8-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; GFX8-NEXT:    flat_load_dwordx3 v[1:3], v[1:2]
-; GFX8-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX8-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX8-NEXT:    s_cbranch_execz .LBB13_2
 ; GFX8-NEXT:  ; %bb.1: ; %bb
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x74
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x50
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_cmp_lg_u32 s0, 0
 ; GFX8-NEXT:    s_cselect_b32 s6, 1, 0
 ; GFX8-NEXT:  .LBB13_2: ; %exit
 ; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_add_u32 s0, s2, 8
 ; GFX8-NEXT:    s_addc_u32 s1, s3, 0
 ; GFX8-NEXT:    s_and_b32 s2, 1, s6
 ; GFX8-NEXT:    v_cmp_ne_u32_e64 vcc, 0, s2
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    s_nop 2
 ; GFX8-NEXT:    v_div_fmas_f32 v2, v1, v2, v3
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX8-NEXT:    flat_store_dword v[0:1], v2
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX10_W32-LABEL: test_div_fmas_f32_i1_phi_vcc:
 ; GFX10_W32:       ; %bb.0: ; %entry
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x28
 ; GFX10_W32-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
 ; GFX10_W32-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
 ; GFX10_W32-NEXT:    s_mov_b32 s5, 0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    global_load_dwordx3 v[1:3], v1, s[2:3]
 ; GFX10_W32-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W32-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W32-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX10_W32-NEXT:    s_cbranch_execz .LBB13_2
 ; GFX10_W32-NEXT:  ; %bb.1: ; %bb
-; GFX10_W32-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x74
+; GFX10_W32-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x50
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    s_cmp_lg_u32 s0, 0
 ; GFX10_W32-NEXT:    s_cselect_b32 s5, 1, 0
 ; GFX10_W32-NEXT:  .LBB13_2: ; %exit
 ; GFX10_W32-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX10_W32-NEXT:    s_and_b32 s0, 1, s5
 ; GFX10_W32-NEXT:    v_cmp_ne_u32_e64 vcc_lo, 0, s0
 ; GFX10_W32-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10_W32-NEXT:    v_div_fmas_f32 v0, v1, v2, v3
 ; GFX10_W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10_W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W32-NEXT:    global_store_dword v1, v0, s[2:3] offset:8
 ; GFX10_W32-NEXT:    s_endpgm
 ;
 ; GFX10_W64-LABEL: test_div_fmas_f32_i1_phi_vcc:
 ; GFX10_W64:       ; %bb.0: ; %entry
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4c
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x28
 ; GFX10_W64-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
 ; GFX10_W64-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX10_W64-NEXT:    s_mov_b32 s6, 0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    global_load_dwordx3 v[1:3], v1, s[2:3]
 ; GFX10_W64-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX10_W64-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX10_W64-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX10_W64-NEXT:    s_cbranch_execz .LBB13_2
 ; GFX10_W64-NEXT:  ; %bb.1: ; %bb
-; GFX10_W64-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x74
+; GFX10_W64-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x50
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10_W64-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10_W64-NEXT:    s_cmp_lg_u32 s0, 0
 ; GFX10_W64-NEXT:    s_cselect_b32 s6, 1, 0
 ; GFX10_W64-NEXT:  .LBB13_2: ; %exit
 ; GFX10_W64-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX10_W64-NEXT:    s_and_b32 s0, 1, s6

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.sbfe.ll

 ; GFX6-NEXT:    ; return to shader part epilog
   %bfe_i32 = call i64 @llvm.amdgcn.sbfe.i64(i64 %src0, i32 %src1, i32 %src2)
   ret i64 %bfe_i32
 }
 
 define amdgpu_kernel void @bfe_i32_arg_arg_imm(i32 addrspace(1)* %out, i32 %src0, i32 %src1) #0 {
 ; GFX6-LABEL: bfe_i32_arg_arg_imm:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_and_b32 s3, s3, 63
 ; GFX6-NEXT:    s_or_b32 s3, s3, 0x7b0000
 ; GFX6-NEXT:    s_bfe_i32 s3, s4, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 %src0, i32 %src1, i32 123)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_arg_imm_arg(i32 addrspace(1)* %out, i32 %src0, i32 %src2) #0 {
 ; GFX6-LABEL: bfe_i32_arg_imm_arg:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 16
 ; GFX6-NEXT:    s_or_b32 s3, 59, s3
 ; GFX6-NEXT:    s_bfe_i32 s3, s4, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 %src0, i32 123, i32 %src2)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_imm_arg_arg(i32 addrspace(1)* %out, i32 %src1, i32 %src2) #0 {
 ; GFX6-LABEL: bfe_i32_imm_arg_arg:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_and_b32 s3, s3, 63
 ; GFX6-NEXT:    s_lshl_b32 s4, s4, 16
 ; GFX6-NEXT:    s_or_b32 s3, s3, s4
 ; GFX6-NEXT:    s_bfe_i32 s3, 0x7b, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 123, i32 %src1, i32 %src2)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_bfe_print_arg(i32 addrspace(1)* %out, i32 addrspace(1)* %src0) #0 {
 ; GFX6-LABEL: v_bfe_print_arg:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x80002
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %src0, align 4
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 %load, i32 2, i32 8)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_arg_0_width_reg_offset(i32 addrspace(1)* %out, i32 %src0, i32 %src1) #0 {
 ; GFX6-LABEL: bfe_i32_arg_0_width_reg_offset:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_and_b32 s3, s3, 63
 ; GFX6-NEXT:    s_bfe_i32 s3, s4, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.sbfe.i32(i32 %src0, i32 %src1, i32 0)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_arg_0_width_imm_offset(i32 addrspace(1)* %out, i32 %src0, i32 %src1) #0 {
 ; GFX6-LABEL: bfe_i32_arg_0_width_imm_offset:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 8
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.sbfe.i32(i32 %src0, i32 8, i32 0)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_6(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_6:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x1f0001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %shl, i32 1, i32 31)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_7(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_7:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x1f0000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %shl, i32 0, i32 31)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_8(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %shl, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_9(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_9:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %x, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_10(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_10:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x1f0001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %x, i32 1, i32 31)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_11(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_11:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x180008
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %x, i32 8, i32 24)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_12(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_12:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x80018
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %x, i32 24, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_13(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_13:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_ashr_i32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = ashr i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %shl, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4 ret void
 }
 
 define amdgpu_kernel void @bfe_i32_test_14(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_i32_test_14:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshr_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = lshr i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %shl, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4 ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_0(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_0:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_i32 s2, 0, 0
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 0, i32 0, i32 0)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_1(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_1:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_i32 s2, 0x302e, 0
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 12334, i32 0, i32 0)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_2(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_2:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_i32 s2, 0, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 0, i32 0, i32 1)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_3(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_3:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_i32 s2, 1, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 1, i32 0, i32 1)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_4(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_4:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_i32 s2, -1, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 4294967295, i32 0, i32 1)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_5(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_5:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x10007
 ; GFX6-NEXT:    s_bfe_i32 s2, 0x80, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 128, i32 7, i32 1)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_6(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_6:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x80000
 ; GFX6-NEXT:    s_bfe_i32 s2, 0x80, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 128, i32 0, i32 8)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_7(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_7:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x80000
 ; GFX6-NEXT:    s_bfe_i32 s2, 0x7f, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 127, i32 0, i32 8)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_8(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x80006
 ; GFX6-NEXT:    s_bfe_i32 s2, 0x7f, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 127, i32 6, i32 8)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_9(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_9:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x80010
 ; GFX6-NEXT:    s_bfe_i32 s2, 0x10000, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 65536, i32 16, i32 8)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_10(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_10:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x100010
 ; GFX6-NEXT:    s_bfe_i32 s2, 0xffff, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 65535, i32 16, i32 16)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_11(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_11:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x40004
 ; GFX6-NEXT:    s_bfe_i32 s2, 0xa0, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 160, i32 4, i32 4)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_12(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_12:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1001f
 ; GFX6-NEXT:    s_bfe_i32 s2, 0xa0, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 160, i32 31, i32 1)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_13(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_13:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x100010
 ; GFX6-NEXT:    s_bfe_i32 s2, 0x1fffe, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 131070, i32 16, i32 16)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_14(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_14:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1e0002
 ; GFX6-NEXT:    s_bfe_i32 s2, 0xa0, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 160, i32 2, i32 30)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_15(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_15:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1c0004
 ; GFX6-NEXT:    s_bfe_i32 s2, 0xa0, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 160, i32 4, i32 28)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_16(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_16:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_i32 s2, -1, 0x70001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 4294967295, i32 1, i32 7)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_17(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_17:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1f0001
 ; GFX6-NEXT:    s_bfe_i32 s2, 0xff, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 255, i32 1, i32 31)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_i32_constant_fold_test_18(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_i32_constant_fold_test_18:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1001f
 ; GFX6-NEXT:    s_bfe_i32 s2, 0xff, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_i32 = call i32 @llvm.amdgcn.sbfe.i32(i32 255, i32 31, i32 1)
   store i32 %bfe_i32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_sext_in_reg_i24(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_sext_in_reg_i24:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x180000
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x180000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %x, i32 0, i32 24)
   %shl = shl i32 %bfe, 8
   %ashr = ashr i32 %shl, 8
   store i32 %ashr, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @simplify_demanded_bfe_sdiv(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: simplify_demanded_bfe_sdiv:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, 2.0
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s6, -1
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s2, s[2:3], 0x0
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    v_mul_lo_u32 v1, -2, v0
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
; GFX6-NEXT:    s_endpgm
   %div = sdiv i32 %bfe, 2
   store i32 %div, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_0_width(i32 addrspace(1)* %out, i32 addrspace(1)* %ptr) #0 {
 ; GFX6-LABEL: bfe_0_width:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 8
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %ptr, align 4
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %load, i32 8, i32 0)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_8_bfe_8(i32 addrspace(1)* %out, i32 addrspace(1)* %ptr) #0 {
 ; GFX6-LABEL: bfe_8_bfe_8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_mov_b32 s4, 0x80000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, s4
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %ptr, align 4
   %bfe0 = call i32 @llvm.amdgcn.sbfe.i32(i32 %load, i32 0, i32 8)
   %bfe1 = call i32 @llvm.amdgcn.sbfe.i32(i32 %bfe0, i32 0, i32 8)
   store i32 %bfe1, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_8_bfe_16(i32 addrspace(1)* %out, i32 addrspace(1)* %ptr) #0 {
 ; GFX6-LABEL: bfe_8_bfe_16:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x80000
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x100000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %ptr, align 4
   %bfe0 = call i32 @llvm.amdgcn.sbfe.i32(i32 %load, i32 0, i32 8)
   %bfe1 = call i32 @llvm.amdgcn.sbfe.i32(i32 %bfe0, i32 0, i32 16)
   store i32 %bfe1, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; This really should be folded into 1
 define amdgpu_kernel void @bfe_16_bfe_8(i32 addrspace(1)* %out, i32 addrspace(1)* %ptr) #0 {
 ; GFX6-LABEL: bfe_16_bfe_8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x100000
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x80000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %ptr, align 4
   %bfe0 = call i32 @llvm.amdgcn.sbfe.i32(i32 %load, i32 0, i32 16)
   %bfe1 = call i32 @llvm.amdgcn.sbfe.i32(i32 %bfe0, i32 0, i32 8)
   store i32 %bfe1, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; Make sure there isn't a redundant BFE
 define amdgpu_kernel void @sext_in_reg_i8_to_i32_bfe(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
 ; GFX6-LABEL: sext_in_reg_i8_to_i32_bfe:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_add_i32 s3, s3, s4
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x80000
 ; GFX6-NEXT:    s_sext_i32_i8 s3, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %c = add i32 %a, %b ; add to prevent folding into extload
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %c, i32 0, i32 8)
   %shl = shl i32 %bfe, 24
   %ashr = ashr i32 %shl, 24
   store i32 %ashr, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @sext_in_reg_i8_to_i32_bfe_wrong(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
 ; GFX6-LABEL: sext_in_reg_i8_to_i32_bfe_wrong:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_add_i32 s3, s3, s4
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 8
 ; GFX6-NEXT:    s_sext_i32_i8 s3, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %c = add i32 %a, %b ; add to prevent folding into extload
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %c, i32 8, i32 0)
   %shl = shl i32 %bfe, 24
   %ashr = ashr i32 %shl, 24
   store i32 %ashr, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @sextload_i8_to_i32_bfe(i32 addrspace(1)* %out, i8 addrspace(1)* %ptr) #0 {
 ; GFX6-LABEL: sextload_i8_to_i32_bfe:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
 ; GFX6-NEXT:    s_mov_b32 s6, -1
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_load_sbyte v0, off, s[4:7], 0
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_bfe_i32 v0, v0, 0, 8
 ; GFX6-NEXT:    v_bfe_i32 v0, v0, 0, 8
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i8, i8 addrspace(1)* %ptr, align 1
   %sext = sext i8 %load to i32
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %sext, i32 0, i32 8)
   %shl = shl i32 %bfe, 24
   %ashr = ashr i32 %shl, 24
   store i32 %ashr, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @sextload_i8_to_i32_bfe_0(i32 addrspace(1)* %out, i8 addrspace(1)* %ptr) #0 {
 ; GFX6-LABEL: sextload_i8_to_i32_bfe_0:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
 ; GFX6-NEXT:    s_mov_b32 s6, -1
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_load_sbyte v0, off, s[4:7], 0
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_bfe_i32 v0, v0, 8, 0
 ; GFX6-NEXT:    v_bfe_i32 v0, v0, 0, 8
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i8, i8 addrspace(1)* %ptr, align 1
   %sext = sext i8 %load to i32
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %sext, i32 8, i32 0)
   %shl = shl i32 %bfe, 24
   %ashr = ashr i32 %shl, 24
   store i32 %ashr, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @sext_in_reg_i1_bfe_offset_0(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: sext_in_reg_i1_bfe_offset_0:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x10000
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %shr = ashr i32 %shl, 31
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %shr, i32 0, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @sext_in_reg_i1_bfe_offset_1(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: sext_in_reg_i1_bfe_offset_1:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x20000
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x10001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 30
   %shr = ashr i32 %shl, 30
   %bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %shr, i32 1, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @sext_in_reg_i2_bfe_offset_1(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: sext_in_reg_i2_bfe_offset_1:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x20000
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x20001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.ubfe.ll

 ; GFX6-NEXT:    ; return to shader part epilog
   %bfe_i32 = call i64 @llvm.amdgcn.ubfe.i64(i64 %src0, i32 %src1, i32 %src2)
   ret i64 %bfe_i32
 }
 
 define amdgpu_kernel void @bfe_u32_arg_arg_arg(i32 addrspace(1)* %out, i32 %src0, i32 %src1, i32 %src2) #0 {
 ; GFX6-LABEL: bfe_u32_arg_arg_arg:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s2, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
-; GFX6-NEXT:    s_load_dword s0, s[0:1], 0xb
+; GFX6-NEXT:    s_load_dword s2, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; GFX6-NEXT:    s_mov_b32 s6, -1
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_and_b32 s1, s2, 63
 ; GFX6-NEXT:    s_lshl_b32 s2, s2, 16
 ; GFX6-NEXT:    s_or_b32 s1, s1, s2
 ; GFX6-NEXT:    s_bfe_u32 s0, s0, s1
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 %src0, i32 %src1, i32 %src1)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_arg_arg_imm(i32 addrspace(1)* %out, i32 %src0, i32 %src1) #0 {
 ; GFX6-LABEL: bfe_u32_arg_arg_imm:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_and_b32 s3, s3, 63
 ; GFX6-NEXT:    s_or_b32 s3, s3, 0x7b0000
 ; GFX6-NEXT:    s_bfe_u32 s3, s4, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 %src0, i32 %src1, i32 123)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_arg_imm_arg(i32 addrspace(1)* %out, i32 %src0, i32 %src2) #0 {
 ; GFX6-LABEL: bfe_u32_arg_imm_arg:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 16
 ; GFX6-NEXT:    s_or_b32 s3, 59, s3
 ; GFX6-NEXT:    s_bfe_u32 s3, s4, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 %src0, i32 123, i32 %src2)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_imm_arg_arg(i32 addrspace(1)* %out, i32 %src1, i32 %src2) #0 {
 ; GFX6-LABEL: bfe_u32_imm_arg_arg:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_and_b32 s3, s3, 63
 ; GFX6-NEXT:    s_lshl_b32 s4, s4, 16
 ; GFX6-NEXT:    s_or_b32 s3, s3, s4
 ; GFX6-NEXT:    s_bfe_u32 s3, 0x7b, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 123, i32 %src1, i32 %src2)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_arg_0_width_reg_offset(i32 addrspace(1)* %out, i32 %src0, i32 %src1) #0 {
 ; GFX6-LABEL: bfe_u32_arg_0_width_reg_offset:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_and_b32 s3, s3, 63
 ; GFX6-NEXT:    s_bfe_u32 s3, s4, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 %src0, i32 %src1, i32 0)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_arg_0_width_imm_offset(i32 addrspace(1)* %out, i32 %src0, i32 %src1) #0 {
 ; GFX6-LABEL: bfe_u32_arg_0_width_imm_offset:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 8
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 %src0, i32 8, i32 0)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_zextload_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_zextload_i8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
 ; GFX6-NEXT:    s_mov_b32 s6, -1
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
 ; GFX6-NEXT:    s_waitcnt vmcnt(0)
 ; GFX6-NEXT:    v_bfe_u32 v0, v0, 0, 8
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i8, i8 addrspace(1)* %in
   %ext = zext i8 %load to i32
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %ext, i32 0, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; FIXME: Should be using s_add_i32
 define amdgpu_kernel void @bfe_u32_zext_in_reg_i8(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_zext_in_reg_i8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_add_i32 s3, s3, 1
 ; GFX6-NEXT:    s_and_b32 s3, s3, 0xff
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x80000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %in, align 4
   %add = add i32 %load, 1
   %ext = and i32 %add, 255
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %ext, i32 0, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_zext_in_reg_i16(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_zext_in_reg_i16:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_add_i32 s3, s3, 1
 ; GFX6-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x100000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %in, align 4
   %add = add i32 %load, 1
   %ext = and i32 %add, 65535
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %ext, i32 0, i32 16)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_zext_in_reg_i8_offset_1(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_zext_in_reg_i8_offset_1:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_add_i32 s3, s3, 1
 ; GFX6-NEXT:    s_and_b32 s3, s3, 0xff
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x80001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %in, align 4
   %add = add i32 %load, 1
   %ext = and i32 %add, 255
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %ext, i32 1, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_zext_in_reg_i8_offset_3(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_zext_in_reg_i8_offset_3:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_add_i32 s3, s3, 1
 ; GFX6-NEXT:    s_and_b32 s3, s3, 0xff
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x80003
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %in, align 4
   %add = add i32 %load, 1
   %ext = and i32 %add, 255
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %ext, i32 3, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_zext_in_reg_i8_offset_7(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_zext_in_reg_i8_offset_7:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_add_i32 s3, s3, 1
 ; GFX6-NEXT:    s_and_b32 s3, s3, 0xff
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x80007
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %in, align 4
   %add = add i32 %load, 1
   %ext = and i32 %add, 255
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %ext, i32 7, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_zext_in_reg_i16_offset_8(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_zext_in_reg_i16_offset_8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_add_i32 s3, s3, 1
 ; GFX6-NEXT:    s_and_b32 s3, s3, 0xffff
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x80008
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %load = load i32, i32 addrspace(1)* %in, align 4
   %add = add i32 %load, 1
   %ext = and i32 %add, 65535
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %ext, i32 8, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_1(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_1:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %x, i32 0, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_2(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_2:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x80000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shl, i32 0, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_3(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_3:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shl, i32 0, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_4(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_4:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x10000
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %shr = lshr i32 %shl, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shr, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_5(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_5:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s3, s3, 0x10000
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %shr = ashr i32 %shl, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shr, i32 0, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_6(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_6:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x1f0001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shl, i32 1, i32 31)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_7(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_7:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x1f0000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shl, i32 0, i32 31)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_8(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = shl i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shl, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_9(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_9:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %x, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_10(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_10:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x1f0001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %x, i32 1, i32 31)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_11(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_11:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x180008
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %x, i32 8, i32 24)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_12(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_12:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x80018
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %x, i32 24, i32 8)
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; V_ASHRREV_U32_e32 {{v[0-9]+}}, 31, {{v[0-9]+}}
 define amdgpu_kernel void @bfe_u32_test_13(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_13:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_ashr_i32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = ashr i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shl, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4 ret void
 }
 
 define amdgpu_kernel void @bfe_u32_test_14(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 {
 ; GFX6-LABEL: bfe_u32_test_14:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s3, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshr_b32 s3, s3, 31
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x1001f
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %x = load i32, i32 addrspace(1)* %in, align 4
   %shl = lshr i32 %x, 31
   %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 %shl, i32 31, i32 1)
   store i32 %bfe, i32 addrspace(1)* %out, align 4 ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_0(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_0:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_u32 s2, 0, 0
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 0, i32 0, i32 0)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_1(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_1:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_u32 s2, 0x302e, 0
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 12334, i32 0, i32 0)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_2(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_2:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_u32 s2, 0, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 0, i32 0, i32 1)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_3(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_3:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_u32 s2, 1, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 1, i32 0, i32 1)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_4(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_4:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_u32 s2, -1, 0x10000
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 4294967295, i32 0, i32 1)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_5(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_5:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x10007
 ; GFX6-NEXT:    s_bfe_u32 s2, 0x80, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 128, i32 7, i32 1)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_6(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_6:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x80000
 ; GFX6-NEXT:    s_bfe_u32 s2, 0x80, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 128, i32 0, i32 8)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_7(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_7:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x80000
 ; GFX6-NEXT:    s_bfe_u32 s2, 0x7f, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 127, i32 0, i32 8)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_8(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x80006
 ; GFX6-NEXT:    s_bfe_u32 s2, 0x7f, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 127, i32 6, i32 8)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_9(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_9:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x80010
 ; GFX6-NEXT:    s_bfe_u32 s2, 0x10000, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 65536, i32 16, i32 8)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_10(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_10:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x100010
 ; GFX6-NEXT:    s_bfe_u32 s2, 0xffff, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 65535, i32 16, i32 16)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_11(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_11:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x40004
 ; GFX6-NEXT:    s_bfe_u32 s2, 0xa0, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 160, i32 4, i32 4)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_12(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_12:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1001f
 ; GFX6-NEXT:    s_bfe_u32 s2, 0xa0, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 160, i32 31, i32 1)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_13(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_13:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x100010
 ; GFX6-NEXT:    s_bfe_u32 s2, 0x1fffe, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 131070, i32 16, i32 16)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_14(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_14:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1e0002
 ; GFX6-NEXT:    s_bfe_u32 s2, 0xa0, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 160, i32 2, i32 30)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_15(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_15:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1c0004
 ; GFX6-NEXT:    s_bfe_u32 s2, 0xa0, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 160, i32 4, i32 28)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_16(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_16:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_bfe_u32 s2, -1, 0x70001
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 4294967295, i32 1, i32 7)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_17(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_17:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1f0001
 ; GFX6-NEXT:    s_bfe_u32 s2, 0xff, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 255, i32 1, i32 31)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @bfe_u32_constant_fold_test_18(i32 addrspace(1)* %out) #0 {
 ; GFX6-LABEL: bfe_u32_constant_fold_test_18:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, 0x1001f
 ; GFX6-NEXT:    s_bfe_u32 s2, 0xff, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 255, i32 31, i32 1)
   store i32 %bfe_u32, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; Make sure that SimplifyDemandedBits doesn't cause the and to be
 ; reduced to the bits demanded by the bfe.
 
 ; XXX: The operand to v_bfe_u32 could also just directly be the load register.
 define amdgpu_kernel void @simplify_bfe_u32_multi_use_arg(i32 addrspace(1)* %out0,
 ; GFX6-LABEL: simplify_bfe_u32_multi_use_arg:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 s6, -1
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_load_dword s8, s[2:3], 0x0
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_and_b32 s8, s8, 63
 ; GFX6-NEXT:    s_bfe_u32 s9, s8, 0x20002
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s9
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s8
 ; GFX6-NEXT:    buffer_store_dword v1, off, s[4:7], 0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
                                             i32 addrspace(1)* %out1,
                                             i32 addrspace(1)* %in) #0 {
   %src = load i32, i32 addrspace(1)* %in, align 4
   %and = and i32 %src, 63
   %bfe_u32 = call i32 @llvm.amdgcn.ubfe.i32(i32 %and, i32 2, i32 2)
   store i32 %bfe_u32, i32 addrspace(1)* %out0, align 4
   store i32 %and, i32 addrspace(1)* %out1, align 4
   ret void
 }
 
 define amdgpu_kernel void @lshr_and(i32 addrspace(1)* %out, i32 %a) #0 {
 ; GFX6-LABEL: lshr_and:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x30006
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %b = lshr i32 %a, 6
   %c = and i32 %b, 7
   store i32 %c, i32 addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @v_lshr_and(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 {
 ; GFX6-LABEL: v_lshr_and:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xc
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x3
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshr_b32 s3, s3, s4
 ; GFX6-NEXT:    s_and_b32 s3, s3, 7
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %c = lshr i32 %a, %b
   %d = and i32 %c, 7
   store i32 %d, i32 addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @and_lshr(i32 addrspace(1)* %out, i32 %a) #0 {
 ; GFX6-LABEL: and_lshr:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x30006
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %b = and i32 %a, 448
   %c = lshr i32 %b, 6
   store i32 %c, i32 addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @and_lshr2(i32 addrspace(1)* %out, i32 %a) #0 {
 ; GFX6-LABEL: and_lshr2:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x30006
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %b = and i32 %a, 511
   %c = lshr i32 %b, 6
   store i32 %c, i32 addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @shl_lshr(i32 addrspace(1)* %out, i32 %a) #0 {
 ; GFX6-LABEL: shl_lshr:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x2
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_u32 s3, s3, 0x150002
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
   %b = shl i32 %a, 9

llvm/test/CodeGen/AMDGPU/GlobalISel/shl-ext-reduce.ll

; GFX10-NEXT:    s_setpc_b64 s[30:31]
   %ext = sext i32 %and to i64
   %shl = shl i64 %ext, 2
   ret i64 %shl
 }
 
 define amdgpu_kernel void @mulu24_shl64(i32 addrspace(1)* nocapture %arg) {
 ; GFX7-LABEL: mulu24_shl64:
 ; GFX7:       ; %bb.0: ; %bb
-; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX7-NEXT:    v_and_b32_e32 v0, 6, v0
 ; GFX7-NEXT:    v_mul_u32_u24_e32 v0, 7, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX7-NEXT:    v_lshl_b64 v[2:3], v[0:1], 2
 ; GFX7-NEXT:    s_mov_b32 s2, 0
 ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    buffer_store_dword v1, v[2:3], s[0:3], 0 addr64
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: mulu24_shl64:
 ; GFX8:       ; %bb.0: ; %bb
-; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX8-NEXT:    v_and_b32_e32 v0, 6, v0
 ; GFX8-NEXT:    v_mul_u32_u24_e32 v0, 7, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX8-NEXT:    v_lshlrev_b64 v[2:3], 2, v[0:1]
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX8-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v4, v2
 ; GFX8-NEXT:    v_addc_u32_e32 v3, vcc, v5, v3, vcc
 ; GFX8-NEXT:    flat_store_dword v[2:3], v1
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: mulu24_shl64:
 ; GFX9:       ; %bb.0: ; %bb
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    v_and_b32_e32 v0, 6, v0
 ; GFX9-NEXT:    v_mul_u32_u24_e32 v0, 7, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    v_lshlrev_b64 v[2:3], 2, v[0:1]
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v4, v2
 ; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v5, v3, vcc
 ; GFX9-NEXT:    global_store_dword v[2:3], v1, off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: mulu24_shl64:
 ; GFX10:       ; %bb.0: ; %bb
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v0, 6, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    v_mul_u32_u24_e32 v0, 7, v0
 ; GFX10-NEXT:    v_lshlrev_b64 v[2:3], 2, v[0:1]
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX10-NEXT:    v_add_co_u32 v2, vcc_lo, v4, v2
 ; GFX10-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v5, v3, vcc_lo
 ; GFX10-NEXT:    global_store_dword v[2:3], v1, off
 ; GFX10-NEXT:    s_endpgm
 bb:
   %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
   %tmp1 = and i32 %tmp, 6
   %mulconv = mul nuw nsw i32 %tmp1, 7
   %tmp2 = zext i32 %mulconv to i64
   %tmp3 = getelementptr inbounds i32, i32 addrspace(1)* %arg, i64 %tmp2
   store i32 0, i32 addrspace(1)* %tmp3, align 4
   ret void
 }
 
 define amdgpu_kernel void @muli24_shl64(i64 addrspace(1)* nocapture %arg, i32 addrspace(1)* nocapture readonly %arg1) {
 ; GFX7-LABEL: muli24_shl64:
 ; GFX7:       ; %bb.0: ; %bb
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX7-NEXT:    s_mov_b32 s6, 0
 ; GFX7-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; GFX7-NEXT:    buffer_load_dword v1, v[1:2], s[4:7], 0 addr64
 ; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 3, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_or_b32_e32 v0, 0xff800000, v1
 ; GFX7-NEXT:    v_mul_i32_i24_e32 v1, -7, v0
 ; GFX7-NEXT:    v_lshl_b64 v[0:1], v[1:2], 3
 ; GFX7-NEXT:    v_add_i32_e32 v2, vcc, v3, v5
 ; GFX7-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: muli24_shl64:
 ; GFX8:       ; %bb.0: ; %bb
-; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 2, v0
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v5, 3, v0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX8-NEXT:    v_add_u32_e32 v1, vcc, v1, v3
 ; GFX8-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; GFX8-NEXT:    flat_load_dword v4, v[1:2]
 ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
 ; GFX8-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v5
 ; GFX8-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_or_b32_e32 v0, 0xff800000, v4
 ; GFX8-NEXT:    v_mul_i32_i24_e32 v0, -7, v0
 ; GFX8-NEXT:    v_lshlrev_b64 v[0:1], 3, v[0:1]
 ; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GFX8-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: muli24_shl64:
 ; GFX9:       ; %bb.0: ; %bb
-; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v1, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    v_or_b32_e32 v1, 0xff800000, v1
 ; GFX9-NEXT:    v_mul_i32_i24_e32 v1, -7, v1
 ; GFX9-NEXT:    v_lshlrev_b64 v[1:2], 3, v[1:2]
 ; GFX9-NEXT:    global_store_dwordx2 v0, v[1:2], s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: muli24_shl64:
 ; GFX10:       ; %bb.0: ; %bb
-; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v1, v1, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_or_b32_e32 v1, 0xff800000, v1
 ; GFX10-NEXT:    v_mul_i32_i24_e32 v1, -7, v1

llvm/test/CodeGen/AMDGPU/GlobalISel/store-local.128.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX9 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=hawaii -verify-machineinstrs < %s | FileCheck --check-prefix=GFX7 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10 %s
 
 ; FIXME:
 ; XUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=tahiti -verify-machineinstrs < %s | FileCheck --check-prefix=GFX6 %s
 
 define amdgpu_kernel void @store_lds_v4i32(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX9-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX7-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX10-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align1(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align1:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_bfe_u32 s0, 8, 0x100000
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_bfe_u32 s3, s4, 0x100000
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    s_lshr_b32 s3, s3, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    s_lshr_b32 s1, s4, 16
 ; GFX9-NEXT:    ds_write_b8 v1, v0

 ; GFX9-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX9-NEXT:    ds_write_b8 v1, v0 offset:14
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    ds_write_b8 v1, v0 offset:15
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align1:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 s1, 0x80008
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_bfe_u32 s3, s4, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    s_lshr_b32 s2, s4, 16
 ; GFX7-NEXT:    ds_write_b8 v1, v0

 ; GFX7-NEXT:    ds_write_b8 v1, v0 offset:14
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX7-NEXT:    ds_write_b8 v1, v0 offset:15
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align1:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_bfe_u32 s0, 8, 0x100000
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_bfe_u32 s3, s4, 0x100000
 ; GFX10-NEXT:    s_lshr_b32 s1, s4, 16
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-NEXT:    s_lshr_b32 s2, s5, 16
 ; GFX10-NEXT:    s_bfe_u32 s4, s5, 0x100000

 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 1
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align2(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align2:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_lshr_b32 s0, s4, 16
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    ds_write_b16 v1, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    ds_write_b16 v1, v0 offset:2
 ; GFX9-NEXT:    s_lshr_b32 s0, s5, 16

 ; GFX9-NEXT:    v_mov_b32_e32 v0, s7
 ; GFX9-NEXT:    ds_write_b16 v1, v0 offset:12
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    ds_write_b16 v1, v0 offset:14
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align2:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_lshr_b32 s1, s4, 16
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    ds_write_b16 v1, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX7-NEXT:    ds_write_b16 v1, v0 offset:2

 ; GFX7-NEXT:    ds_write_b16 v1, v0 offset:12
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    ds_write_b16 v1, v0 offset:14
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align2:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    s_lshr_b32 s0, s4, 16
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX10-NEXT:    s_lshr_b32 s1, s5, 16
 ; GFX10-NEXT:    s_lshr_b32 s2, s6, 16

 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 2
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align4(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align4:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX9-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s7
 ; GFX9-NEXT:    ds_write2_b32 v1, v3, v0 offset0:2 offset1:3
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align4:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX7-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s7
 ; GFX7-NEXT:    ds_write2_b32 v1, v0, v2 offset0:2 offset1:3
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align4:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s7
 ; GFX10-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX10-NEXT:    ds_write2_b32 v1, v3, v4 offset0:2 offset1:3
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align8(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align8:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX9-NEXT:    ds_write2_b64 v4, v[0:1], v[2:3] offset1:1
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align8:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX7-NEXT:    ds_write2_b64 v4, v[0:1], v[2:3] offset1:1
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align8:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX10-NEXT:    ds_write2_b64 v4, v[0:1], v[2:3] offset1:1
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align16(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align16:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX9-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align16:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX7-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align16:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX10-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 16
   ret void
 }

llvm/test/CodeGen/AMDGPU/GlobalISel/store-local.96.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX9 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=hawaii -verify-machineinstrs < %s | FileCheck --check-prefix=GFX7 %s
 ; RUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10 %s
 
 ; FIXME:
 ; XUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=tahiti -verify-machineinstrs < %s | FileCheck --check-prefix=GFX6 %s
 
 define amdgpu_kernel void @store_lds_v3i32(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s12
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s13
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s14
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX9-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s12
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s13
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s14
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX7-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s12
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s13
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s14
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX10-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align1(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align1:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_bfe_u32 s0, 8, 0x100000
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_bfe_u32 s3, s4, 0x100000
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    s_lshr_b32 s3, s3, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    s_lshr_b32 s1, s4, 16
 ; GFX9-NEXT:    ds_write_b8 v1, v0

 ; GFX9-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX9-NEXT:    ds_write_b8 v1, v0 offset:10
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    ds_write_b8 v1, v0 offset:11
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align1:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 s1, 0x80008
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_bfe_u32 s3, s4, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    s_lshr_b32 s2, s4, 16
 ; GFX7-NEXT:    ds_write_b8 v1, v0

 ; GFX7-NEXT:    ds_write_b8 v1, v0 offset:10
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX7-NEXT:    ds_write_b8 v1, v0 offset:11
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align1:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_bfe_u32 s0, 8, 0x100000
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_lshr_b32 s1, s4, 16
 ; GFX10-NEXT:    s_bfe_u32 s3, s4, 0x100000
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-NEXT:    s_lshr_b32 s2, s5, 16
 ; GFX10-NEXT:    s_bfe_u32 s4, s5, 0x100000

 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 1
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align2(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align2:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_lshr_b32 s0, s4, 16
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    ds_write_b16 v1, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    ds_write_b16 v1, v0 offset:2
 ; GFX9-NEXT:    s_lshr_b32 s0, s5, 16
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX9-NEXT:    ds_write_b16 v1, v0 offset:4
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    ds_write_b16 v1, v0 offset:6
 ; GFX9-NEXT:    s_lshr_b32 s0, s6, 16
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX9-NEXT:    ds_write_b16 v1, v0 offset:8
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    ds_write_b16 v1, v0 offset:10
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align2:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_lshr_b32 s1, s4, 16
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    ds_write_b16 v1, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX7-NEXT:    ds_write_b16 v1, v0 offset:2
 ; GFX7-NEXT:    s_lshr_b32 s0, s5, 16
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX7-NEXT:    ds_write_b16 v1, v0 offset:4
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    ds_write_b16 v1, v0 offset:6
 ; GFX7-NEXT:    s_lshr_b32 s0, s6, 16
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX7-NEXT:    ds_write_b16 v1, v0 offset:8
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    ds_write_b16 v1, v0 offset:10
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align2:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-NEXT:    s_lshr_b32 s0, s4, 16
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    s_lshr_b32 s1, s5, 16
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX10-NEXT:    s_lshr_b32 s2, s6, 16

 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 2
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align4(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align4:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX9-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX9-NEXT:    ds_write_b32 v1, v3 offset:8
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align4:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX7-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX7-NEXT:    ds_write_b32 v1, v0 offset:8
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align4:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX10-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX10-NEXT:    ds_write_b32 v1, v3 offset:8
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align8(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align8:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX9-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX9-NEXT:    ds_write_b32 v1, v3 offset:8
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align8:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX7-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX7-NEXT:    ds_write_b32 v1, v0 offset:8
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align8:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX10-NEXT:    ds_write2_b32 v1, v0, v2 offset1:1
 ; GFX10-NEXT:    ds_write_b32 v1, v3 offset:8
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align16(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align16:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s12
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s13
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s14
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX9-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align16:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s12
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s13
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s14
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX7-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align16:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s12
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s13
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s14
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX10-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 16
   ret void
 }

llvm/test/CodeGen/AMDGPU/ds-sub-offset.ll

entry:
   %arrayidx = getelementptr inbounds i32, i32 addrspace(3)* %tmp0, i32 3
   store i32 123, i32 addrspace(3)* %arrayidx
   ret void
 }
 
 define amdgpu_kernel void @write_ds_sub0_offset0_global_clamp_bit(float %dummy.val) #0 {
 ; CI-LABEL: write_ds_sub0_offset0_global_clamp_bit:
 ; CI:       ; %bb.0: ; %entry
-; CI-NEXT:    s_load_dword s0, s[0:1], 0x9
+; CI-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_sub_i32_e32 v0, vcc, 0, v0
 ; CI-NEXT:    s_mov_b64 vcc, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v1, s0
 ; CI-NEXT:    s_mov_b32 s0, 0
 ; CI-NEXT:    v_div_fmas_f32 v1, v1, v1, v1
 ; CI-NEXT:    v_mov_b32_e32 v2, 0x7b
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    s_mov_b32 s1, s0
 ; CI-NEXT:    ds_write_b32 v0, v2 offset:12
 ; CI-NEXT:    buffer_store_dword v1, off, s[0:3], 0
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: write_ds_sub0_offset0_global_clamp_bit:
 ; GFX9:       ; %bb.0: ; %entry
-; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-NEXT:    s_mov_b64 vcc, 0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    v_sub_u32_e32 v3, 0, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0x7b
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX9-NEXT:    v_div_fmas_f32 v2, v1, v1, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    ds_write_b32 v3, v4 offset:12
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: write_ds_sub0_offset0_global_clamp_bit:
 ; GFX10:       ; %bb.0: ; %entry
-; GFX10-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-NEXT:    s_mov_b32 vcc_lo, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v3, 0x7b
 ; GFX10-NEXT:    v_sub_nc_u32_e32 v2, 0, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    ds_write_b32 v2, v3 offset:12
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    s_endpgm
   %ptr = inttoptr i32 %add to i64 addrspace(3)*
   store i64 123, i64 addrspace(3)* %ptr, align 4
   ret void
 }
 
 define amdgpu_kernel void @add_x_shl_neg_to_sub_misaligned_i64_max_offset_clamp_bit(float %dummy.val) #1 {
 ; CI-LABEL: add_x_shl_neg_to_sub_misaligned_i64_max_offset_clamp_bit:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dword s0, s[0:1], 0x9
+; CI-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_sub_i32_e32 v0, vcc, 0x3fb, v0
 ; CI-NEXT:    s_mov_b64 vcc, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v1, s0
 ; CI-NEXT:    s_mov_b32 s0, 0
 ; CI-NEXT:    v_div_fmas_f32 v1, v1, v1, v1
 ; CI-NEXT:    v_mov_b32_e32 v2, 0x7b
 ; CI-NEXT:    v_mov_b32_e32 v3, 0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    s_mov_b32 s1, s0
 ; CI-NEXT:    ds_write2_b32 v0, v2, v3 offset1:1
 ; CI-NEXT:    buffer_store_dword v1, off, s[0:3], 0
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: add_x_shl_neg_to_sub_misaligned_i64_max_offset_clamp_bit:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-NEXT:    s_mov_b64 vcc, 0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    v_sub_u32_e32 v3, 0x3fb, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0x7b
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX9-NEXT:    v_div_fmas_f32 v2, v1, v1, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    ds_write2_b32 v3, v4, v5 offset1:1
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: add_x_shl_neg_to_sub_misaligned_i64_max_offset_clamp_bit:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-NEXT:    s_mov_b32 vcc_lo, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v3, 0x7b
 ; GFX10-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX10-NEXT:    v_sub_nc_u32_e32 v2, 0x3fb, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    ds_write2_b32 v2, v3, v4 offset1:1

llvm/test/CodeGen/AMDGPU/ds_read2.ll

 @lds.f64 = addrspace(3) global [512 x double] undef, align 8
 
 define amdgpu_kernel void @simple_read2_f32(float addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f32:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b32 v[1:2], v0 offset1:8
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v2 offset1:8
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   %val0 = load float, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   %val1 = load float, float addrspace(3)* %arrayidx1, align 4
   %sum = fadd float %val0, %val1
   %out.gep = getelementptr inbounds float, float addrspace(1)* %out, i32 %x.i
   store float %sum, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_read2_f32_max_offset(float addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f32_max_offset:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b32 v[1:2], v0 offset1:255
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f32_max_offset:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v2 offset1:255
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   %val0 = load float, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 255
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   %val1 = load float, float addrspace(3)* %arrayidx1, align 4
   %sum = fadd float %val0, %val1
   %out.gep = getelementptr inbounds float, float addrspace(1)* %out, i32 %x.i
   store float %sum, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_read2_f32_too_far(float addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f32_too_far:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read_b32 v1, v0
 ; CI-NEXT:    ds_read_b32 v2, v0 offset:1028
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f32_too_far:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    ds_read_b32 v1, v0
 ; GFX9-NEXT:    ds_read_b32 v2, v0 offset:1028
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v1, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   %val0 = load float, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 257
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   %val1 = load float, float addrspace(3)* %arrayidx1, align 4
   %sum = fadd float %val0, %val1
   %out.gep = getelementptr inbounds float, float addrspace(1)* %out, i32 %x.i
   store float %sum, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_read2_f32_x2(float addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f32_x2:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b32 v[1:2], v0 offset1:8
 ; CI-NEXT:    ds_read2_b32 v[3:4], v0 offset0:11 offset1:27
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v1, v1, v2
 ; CI-NEXT:    v_add_f32_e32 v2, v3, v4
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f32_x2:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 2, v0
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v4 offset1:8
 ; GFX9-NEXT:    ds_read2_b32 v[2:3], v4 offset0:11 offset1:27
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    v_add_f32_e32 v1, v2, v3
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    global_store_dword v4, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 0

 ; CI-LABEL: simple_read2_f32_x2_barrier:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b32 v[1:2], v0 offset1:8
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_barrier
 ; CI-NEXT:    ds_read2_b32 v[3:4], v0 offset0:11 offset1:27
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    v_add_f32_e32 v1, v1, v2
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v3, v4
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f32_x2_barrier:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 2, v0
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v4 offset1:8
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_barrier
 ; GFX9-NEXT:    ds_read2_b32 v[2:3], v4 offset0:11 offset1:27
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v1, v2, v3
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    global_store_dword v4, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 0

 ; element results in only folding the inner pair.
 define amdgpu_kernel void @simple_read2_f32_x2_nonzero_base(float addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f32_x2_nonzero_base:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b32 v[1:2], v0 offset0:2 offset1:8
 ; CI-NEXT:    ds_read2_b32 v[3:4], v0 offset0:11 offset1:27
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v1, v1, v2
 ; CI-NEXT:    v_add_f32_e32 v2, v3, v4
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 offset:8
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f32_x2_nonzero_base:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 2, v0
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v4 offset0:2 offset1:8
 ; GFX9-NEXT:    ds_read2_b32 v[2:3], v4 offset0:11 offset1:27
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    v_add_f32_e32 v1, v2, v3
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    global_store_dword v4, v0, s[0:1] offset:8
 ; GFX9-NEXT:    s_endpgm
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2

 ; Be careful of vectors of pointers. We don't know if the 2 pointers
 ; in the vectors are really the same base, so this is not safe to
 ; merge.
 ; Base pointers come from different subregister of same super
 ; register. We can't safely merge this.
 define amdgpu_kernel void @read2_ptr_is_subreg_arg_f32(float addrspace(1)* %out, <2 x float addrspace(3)*> %lds.ptr) #0 {
 ; CI-LABEL: read2_ptr_is_subreg_arg_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 m0, -1
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v1, s2
 ; CI-NEXT:    v_mov_b32_e32 v2, s3
 ; CI-NEXT:    ds_read_b32 v1, v1 offset:32
 ; CI-NEXT:    ds_read_b32 v2, v2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: read2_ptr_is_subreg_arg_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX9-NEXT:    ds_read_b32 v1, v1 offset:32
 ; GFX9-NEXT:    ds_read_b32 v2, v2
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v1, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]

 
 ; Apply a constant scalar offset after the pointer vector extract.  We
 ; are rejecting merges that have the same, constant 0 offset, so make
 ; sure we are really rejecting it because of the different
 ; subregisters.
 define amdgpu_kernel void @read2_ptr_is_subreg_arg_offset_f32(float addrspace(1)* %out, <2 x float addrspace(3)*> %lds.ptr) #0 {
 ; CI-LABEL: read2_ptr_is_subreg_arg_offset_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 m0, -1
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v1, s2
 ; CI-NEXT:    v_mov_b32_e32 v2, s3
 ; CI-NEXT:    ds_read_b32 v1, v1 offset:32
 ; CI-NEXT:    ds_read_b32 v2, v2 offset:32
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: read2_ptr_is_subreg_arg_offset_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX9-NEXT:    ds_read_b32 v1, v1 offset:32
 ; GFX9-NEXT:    ds_read_b32 v2, v2 offset:32
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v1, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]

 }
 
 define amdgpu_kernel void @read2_ptr_is_subreg_f32(float addrspace(1)* %out) #0 {
 ; CI-LABEL: read2_ptr_is_subreg_f32:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b32 v[1:2], v0 offset1:8
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: read2_ptr_is_subreg_f32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v2 offset1:8
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %ptr.0 = insertelement <2 x [512 x float] addrspace(3)*> undef, [512 x float] addrspace(3)* @lds, i32 0
   %ptr.1 = insertelement <2 x [512 x float] addrspace(3)*> %ptr.0, [512 x float] addrspace(3)* @lds, i32 1
   %x.i.v.0 = insertelement <2 x i32> undef, i32 %x.i, i32 0

 
 define amdgpu_kernel void @simple_read2_f32_volatile_0(float addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f32_volatile_0:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read_b32 v1, v0
 ; CI-NEXT:    ds_read_b32 v2, v0 offset:32
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f32_volatile_0:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    ds_read_b32 v1, v0
 ; GFX9-NEXT:    ds_read_b32 v2, v0 offset:32
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v1, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   %val0 = load volatile float, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   %val1 = load float, float addrspace(3)* %arrayidx1, align 4
   %sum = fadd float %val0, %val1
   %out.gep = getelementptr inbounds float, float addrspace(1)* %out, i32 %x.i
   store float %sum, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_read2_f32_volatile_1(float addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f32_volatile_1:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read_b32 v1, v0
 ; CI-NEXT:    ds_read_b32 v2, v0 offset:32
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f32_e32 v2, v1, v2
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f32_volatile_1:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    ds_read_b32 v1, v0
 ; GFX9-NEXT:    ds_read_b32 v2, v0 offset:32
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v1, v1, v2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   %val0 = load float, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   %val1 = load volatile float, float addrspace(3)* %arrayidx1, align 4
   %sum = fadd float %val0, %val1
   %out.gep = getelementptr inbounds float, float addrspace(1)* %out, i32 %x.i
   store float %sum, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 ; Can't fold since not correctly aligned.
 define amdgpu_kernel void @unaligned_read2_f32(float addrspace(1)* %out, float addrspace(3)* %lds) #0 {
 ; CI-LABEL: unaligned_read2_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dword s2, s[0:1], 0xb
+; CI-NEXT:    s_load_dword s2, s[0:1], 0x2
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v1, vcc, s2, v0
 ; CI-NEXT:    ds_read_u8 v2, v1 offset:34
 ; CI-NEXT:    ds_read_u8 v3, v1 offset:32
 ; CI-NEXT:    ds_read_u8 v4, v1 offset:3
 ; CI-NEXT:    ds_read_u8 v5, v1 offset:2
 ; CI-NEXT:    ds_read_u8 v6, v1 offset:1

 ; CI-NEXT:    v_add_f32_e32 v2, v4, v1
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-ALIGNED-LABEL: unaligned_read2_f32:
 ; GFX9-ALIGNED:       ; %bb.0:
-; GFX9-ALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x2c
-; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX9-ALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x8
+; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX9-ALIGNED-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-ALIGNED-NEXT:    v_add_u32_e32 v1, s4, v0
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v2, v1
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v3, v1 offset:1
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v4, v1 offset:2
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v5, v1 offset:3
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v6, v1 offset:32

 ; GFX9-ALIGNED-NEXT:    v_lshl_or_b32 v1, v1, 8, v8
 ; GFX9-ALIGNED-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
 ; GFX9-ALIGNED-NEXT:    v_add_f32_e32 v1, v2, v1
 ; GFX9-ALIGNED-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX9-ALIGNED-NEXT:    s_endpgm
 ;
 ; GFX9-UNALIGNED-LABEL: unaligned_read2_f32:
 ; GFX9-UNALIGNED:       ; %bb.0:
-; GFX9-UNALIGNED-NEXT:    s_load_dword s2, s[0:1], 0x2c
+; GFX9-UNALIGNED-NEXT:    s_load_dword s2, s[0:1], 0x8
 ; GFX9-UNALIGNED-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    v_add_u32_e32 v0, s2, v2
 ; GFX9-UNALIGNED-NEXT:    ds_read2_b32 v[0:1], v0 offset1:8
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-UNALIGNED-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX9-UNALIGNED-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds float, float addrspace(3)* %lds, i32 %x.i
   %val0 = load float, float addrspace(3)* %arrayidx0, align 1
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds float, float addrspace(3)* %lds, i32 %add.x
   %val1 = load float, float addrspace(3)* %arrayidx1, align 1
   %sum = fadd float %val0, %val1
   %out.gep = getelementptr inbounds float, float addrspace(1)* %out, i32 %x.i
   store float %sum, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 define amdgpu_kernel void @unaligned_offset_read2_f32(float addrspace(1)* %out, float addrspace(3)* %lds) #0 {
 ; CI-LABEL: unaligned_offset_read2_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dword s2, s[0:1], 0xb
+; CI-NEXT:    s_load_dword s2, s[0:1], 0x2
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v1, vcc, s2, v0
 ; CI-NEXT:    ds_read_u8 v2, v1 offset:11
 ; CI-NEXT:    ds_read_u8 v3, v1 offset:9
 ; CI-NEXT:    ds_read_u8 v4, v1 offset:8
 ; CI-NEXT:    ds_read_u8 v5, v1 offset:7
 ; CI-NEXT:    ds_read_u8 v6, v1 offset:6

 ; CI-NEXT:    v_add_f32_e32 v2, v4, v1
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-ALIGNED-LABEL: unaligned_offset_read2_f32:
 ; GFX9-ALIGNED:       ; %bb.0:
-; GFX9-ALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x2c
-; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX9-ALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x8
+; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX9-ALIGNED-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-ALIGNED-NEXT:    v_add_u32_e32 v1, s4, v0
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v2, v1 offset:5
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v3, v1 offset:6
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v4, v1 offset:7
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v5, v1 offset:8
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v6, v1 offset:9

 ; GFX9-ALIGNED-NEXT:    v_lshl_or_b32 v1, v1, 8, v8
 ; GFX9-ALIGNED-NEXT:    v_lshl_or_b32 v1, v1, 16, v3
 ; GFX9-ALIGNED-NEXT:    v_add_f32_e32 v1, v2, v1
 ; GFX9-ALIGNED-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX9-ALIGNED-NEXT:    s_endpgm
 ;
 ; GFX9-UNALIGNED-LABEL: unaligned_offset_read2_f32:
 ; GFX9-UNALIGNED:       ; %bb.0:
-; GFX9-UNALIGNED-NEXT:    s_load_dword s2, s[0:1], 0x2c
+; GFX9-UNALIGNED-NEXT:    s_load_dword s2, s[0:1], 0x8
 ; GFX9-UNALIGNED-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    v_add3_u32 v0, s2, v2, 5
 ; GFX9-UNALIGNED-NEXT:    ds_read2_b32 v[0:1], v0 offset1:1
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-UNALIGNED-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX9-UNALIGNED-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
; GFX9-UNALIGNED-NEXT:    s_endpgm
   %out.gep = getelementptr inbounds float, float addrspace(1)* %out, i32 %x.i
   store float %sum, float addrspace(1)* %out.gep, align 4
   ret void
 }
 
 define amdgpu_kernel void @misaligned_2_simple_read2_f32(float addrspace(1)* %out, float addrspace(3)* %lds) #0 {
 ; CI-LABEL: misaligned_2_simple_read2_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dword s2, s[0:1], 0xb
+; CI-NEXT:    s_load_dword s2, s[0:1], 0x2
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v1, vcc, s2, v0
 ; CI-NEXT:    ds_read_u16 v2, v1 offset:32
 ; CI-NEXT:    ds_read_u16 v3, v1 offset:2
 ; CI-NEXT:    ds_read_u16 v4, v1
 ; CI-NEXT:    ds_read_u16 v1, v1 offset:34
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(2)
 ; CI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; CI-NEXT:    s_waitcnt lgkmcnt(1)
 ; CI-NEXT:    v_or_b32_e32 v3, v3, v4
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; CI-NEXT:    v_or_b32_e32 v1, v1, v2
 ; CI-NEXT:    v_add_f32_e32 v2, v3, v1
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-ALIGNED-LABEL: misaligned_2_simple_read2_f32:
 ; GFX9-ALIGNED:       ; %bb.0:
-; GFX9-ALIGNED-NEXT:    s_load_dword s2, s[0:1], 0x2c
+; GFX9-ALIGNED-NEXT:    s_load_dword s2, s[0:1], 0x8
 ; GFX9-ALIGNED-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-ALIGNED-NEXT:    v_add_u32_e32 v1, s2, v0
 ; GFX9-ALIGNED-NEXT:    ds_read_u16 v2, v1
 ; GFX9-ALIGNED-NEXT:    ds_read_u16 v3, v1 offset:2
 ; GFX9-ALIGNED-NEXT:    ds_read_u16 v4, v1 offset:32
 ; GFX9-ALIGNED-NEXT:    ds_read_u16 v1, v1 offset:34
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(2)
 ; GFX9-ALIGNED-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-ALIGNED-NEXT:    v_lshl_or_b32 v1, v1, 16, v4
 ; GFX9-ALIGNED-NEXT:    v_add_f32_e32 v1, v2, v1
 ; GFX9-ALIGNED-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-ALIGNED-NEXT:    s_endpgm
 ;
 ; GFX9-UNALIGNED-LABEL: misaligned_2_simple_read2_f32:
 ; GFX9-UNALIGNED:       ; %bb.0:
-; GFX9-UNALIGNED-NEXT:    s_load_dword s2, s[0:1], 0x2c
+; GFX9-UNALIGNED-NEXT:    s_load_dword s2, s[0:1], 0x8
 ; GFX9-UNALIGNED-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    v_add_u32_e32 v0, s2, v2
 ; GFX9-UNALIGNED-NEXT:    ds_read2_b32 v[0:1], v0 offset1:8
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-UNALIGNED-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX9-UNALIGNED-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1

 }
 
 define amdgpu_kernel void @simple_read2_f64(double addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b64 v[0:3], v4 offset1:8
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_mov_b32_e32 v5, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
 ; CI-NEXT:    buffer_store_dwordx2 v[0:1], v[4:5], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f64:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
 ; GFX9-NEXT:    ds_read2_b64 v[0:3], v4 offset1:8
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
 ; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %x.i
   %val0 = load double, double addrspace(3)* %arrayidx0, align 8
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %add.x
   %val1 = load double, double addrspace(3)* %arrayidx1, align 8
   %sum = fadd double %val0, %val1
   %out.gep = getelementptr inbounds double, double addrspace(1)* %out, i32 %x.i
   store double %sum, double addrspace(1)* %out.gep, align 8
   ret void
 }
 
 define amdgpu_kernel void @simple_read2_f64_max_offset(double addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f64_max_offset:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b64 v[0:3], v4 offset1:255
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_mov_b32_e32 v5, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
 ; CI-NEXT:    buffer_store_dwordx2 v[0:1], v[4:5], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f64_max_offset:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
 ; GFX9-NEXT:    ds_read2_b64 v[0:3], v4 offset1:255
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
 ; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %x.i
   %val0 = load double, double addrspace(3)* %arrayidx0, align 8
   %add.x = add nsw i32 %x.i, 255
   %arrayidx1 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %add.x
   %val1 = load double, double addrspace(3)* %arrayidx1, align 8
   %sum = fadd double %val0, %val1
   %out.gep = getelementptr inbounds double, double addrspace(1)* %out, i32 %x.i
   store double %sum, double addrspace(1)* %out.gep, align 8
   ret void
 }
 
 define amdgpu_kernel void @simple_read2_f64_too_far(double addrspace(1)* %out) #0 {
 ; CI-LABEL: simple_read2_f64_too_far:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read_b64 v[1:2], v0
 ; CI-NEXT:    ds_read_b64 v[3:4], v0 offset:2056
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f64 v[2:3], v[1:2], v[3:4]
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_read2_f64_too_far:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
 ; GFX9-NEXT:    ds_read_b64 v[0:1], v4
 ; GFX9-NEXT:    ds_read_b64 v[2:3], v4 offset:2056
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
 ; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %arrayidx0 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %x.i
   %val0 = load double, double addrspace(3)* %arrayidx0, align 8
   %add.x = add nsw i32 %x.i, 257
   %arrayidx1 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %add.x
   %val1 = load double, double addrspace(3)* %arrayidx1, align 8
   %sum = fadd double %val0, %val1
   %out.gep = getelementptr inbounds double, double addrspace(1)* %out, i32 %x.i
   store double %sum, double addrspace(1)* %out.gep, align 8
   ret void
 }
 
 ; Alignment only 4
 define amdgpu_kernel void @misaligned_read2_f64(double addrspace(1)* %out, double addrspace(3)* %lds) #0 {
 ; CI-LABEL: misaligned_read2_f64:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dword s2, s[0:1], 0xb
+; CI-NEXT:    s_load_dword s2, s[0:1], 0x2
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v3, vcc, s2, v0
 ; CI-NEXT:    ds_read2_b32 v[1:2], v3 offset1:1
 ; CI-NEXT:    ds_read2_b32 v[3:4], v3 offset0:14 offset1:15
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_f64 v[2:3], v[1:2], v[3:4]
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: misaligned_read2_f64:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_u32_e32 v2, s2, v4
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v2 offset1:1
 ; GFX9-NEXT:    ds_read2_b32 v[2:3], v2 offset0:14 offset1:15
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
 ; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm

 @foo = addrspace(3) global [4 x i32] undef, align 4
 
 define amdgpu_kernel void @load_constant_adjacent_offsets(i32 addrspace(1)* %out) {
 ; CI-LABEL: load_constant_adjacent_offsets:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_mov_b32_e32 v0, 0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read_b64 v[0:1], v0
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: load_constant_adjacent_offsets:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    ds_read_b64 v[0:1], v2
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_u32_e32 v0, v0, v1
 ; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %val0 = load i32, i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @foo, i32 0, i32 0), align 4
   %val1 = load i32, i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @foo, i32 0, i32 1), align 4
   %sum = add i32 %val0, %val1
   store i32 %sum, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @load_constant_disjoint_offsets(i32 addrspace(1)* %out) {
 ; CI-LABEL: load_constant_disjoint_offsets:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_mov_b32_e32 v0, 0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b32 v[0:1], v0 offset1:2
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: load_constant_disjoint_offsets:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v2 offset1:2
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_u32_e32 v0, v0, v1
 ; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %val0 = load i32, i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @foo, i32 0, i32 0), align 4
   %val1 = load i32, i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @foo, i32 0, i32 2), align 4
   %sum = add i32 %val0, %val1
   store i32 %sum, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 @bar = addrspace(3) global [4 x i64] undef, align 4
 
 define amdgpu_kernel void @load_misaligned64_constant_offsets(i64 addrspace(1)* %out) {
 ; CI-LABEL: load_misaligned64_constant_offsets:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_mov_b32_e32 v0, 0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read_b128 v[0:3], v0
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; CI-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; CI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: load_misaligned64_constant_offsets:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    ds_read_b128 v[0:3], v4
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %val0 = load i64, i64 addrspace(3)* getelementptr inbounds ([4 x i64], [4 x i64] addrspace(3)* @bar, i32 0, i32 0), align 4
   %val1 = load i64, i64 addrspace(3)* getelementptr inbounds ([4 x i64], [4 x i64] addrspace(3)* @bar, i32 0, i32 1), align 4
   %sum = add i64 %val0, %val1
   store i64 %sum, i64 addrspace(1)* %out, align 8
   ret void
 }
 
 @bar.large = addrspace(3) global [4096 x i64] undef, align 4
 
 define amdgpu_kernel void @load_misaligned64_constant_large_offsets(i64 addrspace(1)* %out) {
 ; CI-LABEL: load_misaligned64_constant_large_offsets:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    v_mov_b32_e32 v2, 0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read_b64 v[0:1], v2 offset:16384
 ; CI-NEXT:    ds_read_b64 v[2:3], v2 offset:32760
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; CI-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; CI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: load_misaligned64_constant_large_offsets:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    ds_read_b64 v[0:1], v4 offset:16384
 ; GFX9-NEXT:    ds_read_b64 v[2:3], v4 offset:32760
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
 ; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %val0 = load i64, i64 addrspace(3)* getelementptr inbounds ([4096 x i64], [4096 x i64] addrspace(3)* @bar.large, i32 0, i32 2048), align 4
   %val1 = load i64, i64 addrspace(3)* getelementptr inbounds ([4096 x i64], [4096 x i64] addrspace(3)* @bar.large, i32 0, i32 4095), align 4
   %sum = add i64 %val0, %val1
   store i64 %sum, i64 addrspace(1)* %out, align 8
   ret void
 }
 
 @sgemm.lA = internal unnamed_addr addrspace(3) global [264 x float] undef, align 4
 @sgemm.lB = internal unnamed_addr addrspace(3) global [776 x float] undef, align 4
 
 define amdgpu_kernel void @sgemm_inner_loop_read2_sequence(float addrspace(1)* %C, i32 %lda, i32 %ldb) #0 {
 ; CI-LABEL: sgemm_inner_loop_read2_sequence:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
 ; CI-NEXT:    s_lshl_b32 s0, s2, 2
 ; CI-NEXT:    s_add_i32 s1, s0, 0xc20
 ; CI-NEXT:    s_addk_i32 s0, 0xc60
 ; CI-NEXT:    v_mov_b32_e32 v0, s1
 ; CI-NEXT:    v_mov_b32_e32 v2, s0
 ; CI-NEXT:    v_lshlrev_b32_e32 v8, 2, v1
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_read2_b32 v[0:1], v0 offset1:1

 ; GFX9-NEXT:    ds_read2_b32 v[8:9], v8 offset0:64 offset1:65
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(4)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(3)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v2
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v3
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(2)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v4
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v5
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v6
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v7
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v8
 ; GFX9-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX9-NEXT:    v_add_f32_e32 v0, v0, v9
 ; GFX9-NEXT:    global_store_dword v10, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
   %sum.8 = fadd float %sum.7, %tmp25
   store float %sum.8, float addrspace(1)* %C, align 4
   ret void
 }
 
 define amdgpu_kernel void @misaligned_read2_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(3)* %in) #0 {
 ; CI-LABEL: misaligned_read2_v2i32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dword s2, s[0:1], 0xb
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dword s2, s[0:1], 0x2
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v0, s2
 ; CI-NEXT:    ds_read2_b32 v[0:1], v0 offset1:1
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: misaligned_read2_v2i32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x8
+; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v0 offset1:1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
 ; GFX9-NEXT:    s_endpgm
   %load = load <2 x i32>, <2 x i32> addrspace(3)* %in, align 4
   store <2 x i32> %load, <2 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @misaligned_read2_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %in) #0 {
 ; CI-LABEL: misaligned_read2_i64:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dword s2, s[0:1], 0xb
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dword s2, s[0:1], 0x2
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v0, s2
 ; CI-NEXT:    ds_read2_b32 v[0:1], v0 offset1:1
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: misaligned_read2_i64:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x8
+; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v0 offset1:1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
 ; GFX9-NEXT:    s_endpgm
   %load = load i64, i64 addrspace(3)* %in, align 4
   store i64 %load, i64 addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @ds_read_diff_base_interleaving(
 ; CI-LABEL: ds_read_diff_base_interleaving:
 ; CI:       ; %bb.0: ; %bb
-; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
 ; CI-NEXT:    v_lshlrev_b32_e32 v4, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v2, vcc, s4, v1
 ; CI-NEXT:    v_add_i32_e32 v3, vcc, s5, v4
 ; CI-NEXT:    v_add_i32_e32 v5, vcc, s6, v1
 ; CI-NEXT:    ds_read2_b32 v[0:1], v2 offset1:1

 ; CI-NEXT:    v_mul_f32_e32 v1, v5, v7
 ; CI-NEXT:    s_mov_b32 s2, -1
 ; CI-NEXT:    v_sub_f32_e32 v0, v0, v1
 ; CI-NEXT:    buffer_store_dword v0, off, s[0:3], 0 offset:40
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: ds_read_diff_base_interleaving:
 ; GFX9:       ; %bb.0: ; %bb
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_u32_e32 v2, s4, v1
 ; GFX9-NEXT:    v_add_u32_e32 v3, s5, v0
 ; GFX9-NEXT:    v_add_u32_e32 v4, s6, v1
 ; GFX9-NEXT:    v_add_u32_e32 v6, s7, v0
 ; GFX9-NEXT:    ds_read2_b32 v[0:1], v2 offset1:1
 ; GFX9-NEXT:    ds_read2_b32 v[2:3], v3 offset1:4

 ; CI-NEXT:    s_load_dwordx4 s[40:43], s[40:41], 0x0
 ; CI-NEXT:    s_mov_b32 s14, s10
 ; CI-NEXT:    s_mov_b32 s12, s8
 ; CI-NEXT:    s_mov_b32 s13, s9
 ; CI-NEXT:    v_lshlrev_b32_e32 v3, 2, v0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_add_u32 s40, s40, s11
 ; CI-NEXT:    s_mov_b64 s[10:11], s[6:7]
-; CI-NEXT:    s_load_dwordx2 s[36:37], s[4:5], 0x9
-; CI-NEXT:    s_load_dword s6, s[4:5], 0xb
+; CI-NEXT:    s_load_dwordx2 s[36:37], s[4:5], 0x0
+; CI-NEXT:    s_load_dword s6, s[4:5], 0x2
 ; CI-NEXT:    s_addc_u32 s41, s41, 0
-; CI-NEXT:    s_add_u32 s8, s4, 48
+; CI-NEXT:    s_add_u32 s8, s4, 12
 ; CI-NEXT:    s_addc_u32 s9, s5, 0
 ; CI-NEXT:    s_getpc_b64 s[4:5]
 ; CI-NEXT:    s_add_u32 s4, s4, void_func_void@gotpcrel32@lo+4
 ; CI-NEXT:    s_addc_u32 s5, s5, void_func_void@gotpcrel32@hi+12
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_add_i32_e32 v40, vcc, s6, v3
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_load_dwordx2 s[16:17], s[4:5], 0x0

 ; GFX9-NEXT:    s_mov_b32 s14, s10
 ; GFX9-NEXT:    s_mov_b32 s12, s8
 ; GFX9-NEXT:    s_mov_b32 s13, s9
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_add_u32 s36, s36, s11
 ; GFX9-NEXT:    s_addc_u32 s37, s37, 0
 ; GFX9-NEXT:    s_mov_b64 s[10:11], s[6:7]
-; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x2c
-; GFX9-NEXT:    s_load_dwordx2 s[34:35], s[4:5], 0x24
-; GFX9-NEXT:    s_add_u32 s8, s4, 48
+; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x8
+; GFX9-NEXT:    s_load_dwordx2 s[34:35], s[4:5], 0x0
+; GFX9-NEXT:    s_add_u32 s8, s4, 12
 ; GFX9-NEXT:    s_addc_u32 s9, s5, 0
 ; GFX9-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-NEXT:    s_add_u32 s4, s4, void_func_void@gotpcrel32@lo+4
 ; GFX9-NEXT:    s_addc_u32 s5, s5, void_func_void@gotpcrel32@hi+12
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_lshl_add_u32 v41, v0, 2, s6
 ; GFX9-NEXT:    s_load_dwordx2 s[16:17], s[4:5], 0x0
 ; GFX9-NEXT:    ds_read_b32 v42, v41

 ; CI-NEXT:    s_waitcnt lgkmcnt(2)
 ; CI-NEXT:    v_or_b32_e32 v2, v2, v3
 ; CI-NEXT:    s_waitcnt lgkmcnt(1)
 ; CI-NEXT:    v_or_b32_e32 v1, v1, v4
 ; CI-NEXT:    ds_read_u8 v4, v0 offset:67
 ; CI-NEXT:    ds_read_u8 v6, v0 offset:66
 ; CI-NEXT:    ds_read_u8 v0, v0 offset:65
 ; CI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; CI-NEXT:    v_or_b32_e32 v1, v2, v1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_lshlrev_b32_e32 v2, 8, v6
 ; CI-NEXT:    v_or_b32_e32 v0, v2, v0
 ; CI-NEXT:    v_lshlrev_b32_e32 v2, 8, v5
 ; CI-NEXT:    v_or_b32_e32 v2, v2, v4
 ; CI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000

 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v4, v2 offset:66
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v5, v2 offset:67
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v6, v2 offset:68
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v1, v2 offset:69
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v7, v2 offset:72
 ; GFX9-ALIGNED-NEXT:    ds_read_u8 v8, v2 offset:71
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(7)
 ; GFX9-ALIGNED-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
-; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-ALIGNED-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX9-ALIGNED-NEXT:    v_lshlrev_b32_e32 v1, 8, v7
 ; GFX9-ALIGNED-NEXT:    v_or_b32_sdwa v1, v1, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; GFX9-ALIGNED-NEXT:    v_or_b32_e32 v1, v1, v0
 ; GFX9-ALIGNED-NEXT:    v_lshlrev_b32_e32 v0, 8, v4
 ; GFX9-ALIGNED-NEXT:    v_or_b32_e32 v0, v0, v3
 ; GFX9-ALIGNED-NEXT:    v_lshlrev_b32_e32 v3, 8, v6
 ; GFX9-ALIGNED-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; GFX9-ALIGNED-NEXT:    v_or_b32_e32 v0, v3, v0
 ; GFX9-ALIGNED-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-ALIGNED-NEXT:    s_endpgm
 ;
 ; GFX9-UNALIGNED-LABEL: read2_v2i32_align1_odd_offset:
 ; GFX9-UNALIGNED:       ; %bb.0: ; %entry
 ; GFX9-UNALIGNED-NEXT:    v_mov_b32_e32 v0, 0x41
-; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; GFX9-UNALIGNED-NEXT:    ds_read2_b32 v[0:1], v0 offset1:1
 ; GFX9-UNALIGNED-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-UNALIGNED-NEXT:    s_endpgm
 entry:
   %load = load <2 x i32>, <2 x i32> addrspace(3)* bitcast (i8 addrspace(3)* getelementptr (i8, i8 addrspace(3)* bitcast ([100 x <2 x i32>] addrspace(3)* @v2i32_align1 to i8 addrspace(3)*), i32 65) to <2 x i32> addrspace(3)*), align 1
   store <2 x i32> %load, <2 x i32> addrspace(1)* %out

llvm/test/CodeGen/AMDGPU/ds_write2.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --force-update
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=bonaire -verify-machineinstrs -mattr=+load-store-opt < %s | FileCheck -enable-var-scope --check-prefix=CI %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs -mattr=+load-store-opt,-unaligned-access-mode < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-ALIGNED %s
 ; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs -mattr=+load-store-opt,+unaligned-access-mode < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,GFX9-UNALIGNED %s
 
 @lds = addrspace(3) global [512 x float] undef, align 4
 @lds.f64 = addrspace(3) global [512 x double] undef, align 8
 
 define amdgpu_kernel void @simple_write2_one_val_f32(float addrspace(1)* %C, float addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_one_val_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dword v1, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write2_b32 v0, v1, v1 offset1:8
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_one_val_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v1 offset1:8
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %in.gep = getelementptr float, float addrspace(1)* %in, i32 %x.i
   %val = load float, float addrspace(1)* %in.gep, align 4
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   store float %val, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   store float %val, float addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_f32(float addrspace(1)* %C, float addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_two_val_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dword v2, v[0:1], s[0:3], 0 addr64 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    buffer_load_dword v1, v[0:1], s[0:3], 0 addr64 offset:4 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_write2_b32 v0, v2, v1 offset1:8
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[0:1] offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset1:8
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %in.gep.0 = getelementptr float, float addrspace(1)* %in, i32 %x.i
   %in.gep.1 = getelementptr float, float addrspace(1)* %in.gep.0, i32 1
   %val0 = load volatile float, float addrspace(1)* %in.gep.0, align 4
   %val1 = load volatile float, float addrspace(1)* %in.gep.1, align 4
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   store float %val0, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   store float %val1, float addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_f32_volatile_0(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
 ; CI-LABEL: simple_write2_two_val_f32_volatile_0:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s7, 0xf000
 ; CI-NEXT:    s_mov_b32 s6, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_mov_b64 s[4:5], s[0:1]
 ; CI-NEXT:    s_mov_b64 s[0:1], s[2:3]
 ; CI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; CI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    buffer_load_dword v1, v[0:1], s[0:3], 0 addr64 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_write_b32 v0, v2
 ; CI-NEXT:    ds_write_b32 v0, v1 offset:32
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_f32_volatile_0:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write_b32 v0, v1
 ; GFX9-NEXT:    ds_write_b32 v0, v2 offset:32
; GFX9-NEXT:    s_endpgm
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   store float %val1, float addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_f32_volatile_1(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
 ; CI-LABEL: simple_write2_two_val_f32_volatile_1:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s7, 0xf000
 ; CI-NEXT:    s_mov_b32 s6, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_mov_b64 s[4:5], s[0:1]
 ; CI-NEXT:    s_mov_b64 s[0:1], s[2:3]
 ; CI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; CI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    buffer_load_dword v1, v[0:1], s[0:3], 0 addr64 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_write_b32 v0, v2
 ; CI-NEXT:    ds_write_b32 v0, v1 offset:32
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_f32_volatile_1:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write_b32 v0, v1
 ; GFX9-NEXT:    ds_write_b32 v0, v2 offset:32

 ; 2 data subregisters from different super registers.
 ; TODO: GFX9 has v_mov_b32_e32 v2, lds@abs32@lo
 ;       This should be an s_mov_b32. The v_mov_b32 gets introduced by an
 ;       early legalization of the constant bus constraint on the v_lshl_add_u32,
 ;       and then SIFoldOperands folds in an unlucky order.
 define amdgpu_kernel void @simple_write2_two_val_subreg2_mixed_f32(float addrspace(1)* %C, <2 x float> addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_two_val_subreg2_mixed_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
 ; CI-NEXT:    v_mov_b32_e32 v2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx2 v[3:4], v[1:2], s[0:3], 0 addr64 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx2 v[1:2], v[1:2], s[0:3], 0 addr64 offset:8 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_write2_b32 v0, v3, v2 offset1:8
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_subreg2_mixed_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    ; kill: killed $vgpr4
 ; GFX9-NEXT:    ; kill: killed $sgpr0_sgpr1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[1:2], v4, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[2:3], v4, s[0:1] offset:8 glc
; GFX9-NEXT:    s_endpgm
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   store float %val1.1, float addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_subreg2_f32(float addrspace(1)* %C, <2 x float> addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_two_val_subreg2_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
 ; CI-NEXT:    v_mov_b32_e32 v2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx2 v[1:2], v[1:2], s[0:3], 0 addr64
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write2_b32 v0, v1, v2 offset1:8
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_subreg2_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[1:2], v1, s[0:1]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset1:8
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %in.gep = getelementptr <2 x float>, <2 x float> addrspace(1)* %in, i32 %x.i
   %val = load <2 x float>, <2 x float> addrspace(1)* %in.gep, align 8
   %val0 = extractelement <2 x float> %val, i32 0
   %val1 = extractelement <2 x float> %val, i32 1
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   store float %val0, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   store float %val1, float addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_subreg4_f32(float addrspace(1)* %C, <4 x float> addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_two_val_subreg4_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
 ; CI-NEXT:    v_mov_b32_e32 v2, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx4 v[1:4], v[1:2], s[0:3], 0 addr64
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write2_b32 v0, v1, v4 offset1:8
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_subreg4_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx4 v[1:4], v1, s[0:1]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v4 offset1:8
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %in.gep = getelementptr <4 x float>, <4 x float> addrspace(1)* %in, i32 %x.i
   %val = load <4 x float>, <4 x float> addrspace(1)* %in.gep, align 16
   %val0 = extractelement <4 x float> %val, i32 0
   %val1 = extractelement <4 x float> %val, i32 3
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   store float %val0, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   store float %val1, float addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_max_offset_f32(float addrspace(1)* %C, float addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_two_val_max_offset_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dword v2, v[0:1], s[0:3], 0 addr64 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    buffer_load_dword v1, v[0:1], s[0:3], 0 addr64 offset:4 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_write2_b32 v0, v2, v1 offset1:255
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_max_offset_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dword v2, v0, s[0:1] offset:4 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset1:255
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %in.gep.0 = getelementptr float, float addrspace(1)* %in, i32 %x.i
   %in.gep.1 = getelementptr float, float addrspace(1)* %in.gep.0, i32 1
   %val0 = load volatile float, float addrspace(1)* %in.gep.0, align 4
   %val1 = load volatile float, float addrspace(1)* %in.gep.1, align 4
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
   store float %val0, float addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 255
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   store float %val1, float addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_too_far_f32(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
 ; CI-LABEL: simple_write2_two_val_too_far_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s7, 0xf000
 ; CI-NEXT:    s_mov_b32 s6, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_mov_b64 s[4:5], s[0:1]
 ; CI-NEXT:    s_mov_b64 s[0:1], s[2:3]
 ; CI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; CI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
 ; CI-NEXT:    buffer_load_dword v1, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt vmcnt(1)
 ; CI-NEXT:    ds_write_b32 v0, v2
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write_b32 v0, v1 offset:1028
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_too_far_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
 ; GFX9-NEXT:    ds_write_b32 v0, v1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write_b32 v0, v2 offset:1028
; GFX9-NEXT:    s_endpgm
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
   store float %val1, float addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_f32_x2(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
 ; CI-LABEL: simple_write2_two_val_f32_x2:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s7, 0xf000
 ; CI-NEXT:    s_mov_b32 s6, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_mov_b64 s[4:5], s[0:1]
 ; CI-NEXT:    s_mov_b64 s[0:1], s[2:3]
 ; CI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; CI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
 ; CI-NEXT:    buffer_load_dword v1, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write2_b32 v0, v2, v1 offset1:8
 ; CI-NEXT:    ds_write2_b32 v0, v2, v1 offset0:11 offset1:27
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_f32_x2:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset1:8
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset0:11 offset1:27
 ; GFX9-NEXT:    s_endpgm
; GFX9-NEXT:    s_endpgm
   store float %val1, float addrspace(3)* %arrayidx3, align 4
 
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_f32_x2_nonzero_base(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
 ; CI-LABEL: simple_write2_two_val_f32_x2_nonzero_base:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s7, 0xf000
 ; CI-NEXT:    s_mov_b32 s6, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_mov_b64 s[4:5], s[0:1]
 ; CI-NEXT:    s_mov_b64 s[0:1], s[2:3]
 ; CI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; CI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
 ; CI-NEXT:    buffer_load_dword v1, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write2_b32 v0, v2, v1 offset0:3 offset1:8
 ; CI-NEXT:    ds_write2_b32 v0, v2, v1 offset0:11 offset1:27
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_f32_x2_nonzero_base:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    global_load_dword v2, v0, s[2:3]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset0:3 offset1:8
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset0:11 offset1:27
 ; GFX9-NEXT:    s_endpgm
; GFX9-NEXT:    s_endpgm
   store float %val1, float addrspace(3)* %arrayidx3, align 4
 
   ret void
 }
 
 define amdgpu_kernel void @write2_ptr_subreg_arg_two_val_f32(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1, <2 x float addrspace(3)*> %lds.ptr) #0 {
 ; CI-LABEL: write2_ptr_subreg_arg_two_val_f32:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
-; CI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xf
+; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2
+; CI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x6
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_mov_b64 s[0:1], s[4:5]
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_mov_b64 s[4:5], s[6:7]
 ; CI-NEXT:    s_mov_b64 s[6:7], s[2:3]
 ; CI-NEXT:    buffer_load_dword v2, v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
 ; CI-NEXT:    v_mov_b32_e32 v1, s8
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    v_mov_b32_e32 v3, s9
 ; CI-NEXT:    s_waitcnt vmcnt(1)
 ; CI-NEXT:    ds_write_b32 v1, v2 offset:32
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write_b32 v3, v0 offset:32
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: write2_ptr_subreg_arg_two_val_f32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x3c
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x8
+; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x18
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[4:5]
 ; GFX9-NEXT:    global_load_dword v2, v0, s[6:7]
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
 ; GFX9-NEXT:    ds_write_b32 v0, v1 offset:32
; GFX9-NEXT:    s_endpgm
   %add.x = add nsw i32 %x.i, 8
   store float %val1, float addrspace(3)* %gep.1.offset, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_one_val_f64(double addrspace(1)* %C, double addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_one_val_f64:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx2 v[1:2], v[0:1], s[0:3], 0 addr64
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write2_b64 v0, v[1:2], v[1:2] offset1:8
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_one_val_f64:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v2, s[0:1]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b64 v2, v[0:1], v[0:1] offset1:8
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %in.gep = getelementptr double, double addrspace(1)* %in, i32 %x.i
   %val = load double, double addrspace(1)* %in.gep, align 8
   %arrayidx0 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %x.i
   store double %val, double addrspace(3)* %arrayidx0, align 8
   %add.x = add nsw i32 %x.i, 8
   %arrayidx1 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %add.x
   store double %val, double addrspace(3)* %arrayidx1, align 8
   ret void
 }
 
 define amdgpu_kernel void @misaligned_simple_write2_one_val_f64(double addrspace(1)* %C, double addrspace(1)* %in, double addrspace(3)* %lds) #0 {
 ; CI-LABEL: misaligned_simple_write2_one_val_f64:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; CI-NEXT:    s_load_dword s0, s[0:1], 0xd
+; CI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; CI-NEXT:    s_load_dword s0, s[0:1], 0x4
 ; CI-NEXT:    s_mov_b32 s7, 0xf000
 ; CI-NEXT:    s_mov_b32 s6, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx2 v[1:2], v[0:1], s[4:7], 0 addr64
 ; CI-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; CI-NEXT:    ds_write2_b32 v0, v1, v2 offset0:14 offset1:15
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: misaligned_simple_write2_one_val_f64:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x34
+; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x10
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-NEXT:    v_add_u32_e32 v2, s4, v2
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b32 v2, v0, v1 offset1:1
 ; GFX9-NEXT:    ds_write2_b32 v2, v0, v1 offset0:14 offset1:15
 ; GFX9-NEXT:    s_endpgm
   %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %in.gep = getelementptr double, double addrspace(1)* %in, i32 %x.i
   %val = load double, double addrspace(1)* %in.gep, align 8
   %arrayidx0 = getelementptr inbounds double, double addrspace(3)* %lds, i32 %x.i
   store double %val, double addrspace(3)* %arrayidx0, align 4
   %add.x = add nsw i32 %x.i, 7
   %arrayidx1 = getelementptr inbounds double, double addrspace(3)* %lds, i32 %add.x
   store double %val, double addrspace(3)* %arrayidx1, align 4
   ret void
 }
 
 define amdgpu_kernel void @unaligned_offset_simple_write2_one_val_f64(double addrspace(1)* %C, double addrspace(1)* %in, double addrspace(3)* %lds) #0 {
 ; CI-LABEL: unaligned_offset_simple_write2_one_val_f64:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; CI-NEXT:    s_load_dword s0, s[0:1], 0xd
+; CI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; CI-NEXT:    s_load_dword s0, s[0:1], 0x4
 ; CI-NEXT:    s_mov_b32 s7, 0xf000
 ; CI-NEXT:    s_mov_b32 s6, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx2 v[1:2], v[0:1], s[4:7], 0 addr64
 ; CI-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
 ; CI-NEXT:    s_mov_b32 m0, -1

 ; CI-NEXT:    ds_write_b8 v0, v2 offset:14
 ; CI-NEXT:    ds_write_b8 v0, v3 offset:12
 ; CI-NEXT:    ds_write_b8 v0, v4 offset:11
 ; CI-NEXT:    ds_write_b8 v0, v5 offset:10
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-ALIGNED-LABEL: unaligned_offset_simple_write2_one_val_f64:
 ; GFX9-ALIGNED:       ; %bb.0:
-; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX9-ALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x34
+; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX9-ALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x10
 ; GFX9-ALIGNED-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-ALIGNED-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-ALIGNED-NEXT:    v_add_u32_e32 v2, s4, v2
 ; GFX9-ALIGNED-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-ALIGNED-NEXT:    ds_write_b8_d16_hi v2, v0 offset:7
 ; GFX9-ALIGNED-NEXT:    ds_write_b8 v2, v0 offset:5
 ; GFX9-ALIGNED-NEXT:    v_lshrrev_b32_e32 v3, 24, v0

 ; GFX9-ALIGNED-NEXT:    ds_write_b8 v2, v0 offset:16
 ; GFX9-ALIGNED-NEXT:    ds_write_b8 v2, v1 offset:14
 ; GFX9-ALIGNED-NEXT:    ds_write_b8 v2, v3 offset:12
 ; GFX9-ALIGNED-NEXT:    ds_write_b8 v2, v4 offset:10
 ; GFX9-ALIGNED-NEXT:    s_endpgm
 ;
 ; GFX9-UNALIGNED-LABEL: unaligned_offset_simple_write2_one_val_f64:
 ; GFX9-UNALIGNED:       ; %bb.0:
-; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX9-UNALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x34
+; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX9-UNALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x10
 ; GFX9-UNALIGNED-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-UNALIGNED-NEXT:    v_add_u32_e32 v2, s4, v2
 ; GFX9-UNALIGNED-NEXT:    v_add_u32_e32 v3, 5, v2
 ; GFX9-UNALIGNED-NEXT:    v_add_u32_e32 v2, 9, v2
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    ds_write2_b32 v3, v0, v1 offset1:1
; GFX9-UNALIGNED-NEXT:    s_endpgm
   %addr1 = bitcast i8 addrspace(3)* %addr1.i8 to double addrspace(3)*
   store double %val, double addrspace(3)* %addr1, align 1
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_two_val_f64(double addrspace(1)* %C, double addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_two_val_f64:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; CI-NEXT:    s_mov_b32 s3, 0xf000
 ; CI-NEXT:    s_mov_b32 s2, 0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; CI-NEXT:    v_mov_b32_e32 v1, 0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    buffer_load_dwordx2 v[4:5], v[0:1], s[0:3], 0 addr64 offset:8 glc
 ; CI-NEXT:    s_waitcnt vmcnt(0)
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    ds_write2_b64 v0, v[2:3], v[4:5] offset1:8
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: simple_write2_two_val_f64:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v4, s[0:1] glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[2:3], v4, s[0:1] offset:8 glc
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    ds_write2_b64 v4, v[0:1], v[2:3] offset1:8
 ; GFX9-NEXT:    s_endpgm

 }
 
 @sgemm.lA = internal unnamed_addr addrspace(3) global [264 x float] undef, align 4
 @sgemm.lB = internal unnamed_addr addrspace(3) global [776 x float] undef, align 4
 
 define amdgpu_kernel void @write2_sgemm_sequence(float addrspace(1)* %C, i32 %lda, i32 %ldb, float addrspace(1)* %in) #0 {
 ; CI-LABEL: write2_sgemm_sequence:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x4
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; CI-NEXT:    s_lshl_b32 s1, s2, 2
 ; CI-NEXT:    s_add_i32 s2, s1, 0xc20
 ; CI-NEXT:    s_addk_i32 s1, 0xc60
 ; CI-NEXT:    v_mov_b32_e32 v0, s2
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v2, s0
 ; CI-NEXT:    v_mov_b32_e32 v3, s0
 ; CI-NEXT:    ds_write2_b32 v0, v2, v3 offset1:1
 ; CI-NEXT:    v_mov_b32_e32 v0, s1
 ; CI-NEXT:    ds_write2_b32 v0, v2, v3 offset1:1
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 2, v1
 ; CI-NEXT:    ds_write2_b32 v0, v2, v3 offset1:1
 ; CI-NEXT:    ds_write2_b32 v0, v2, v3 offset0:32 offset1:33
 ; CI-NEXT:    ds_write2_b32 v0, v2, v3 offset0:64 offset1:65
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: write2_sgemm_sequence:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x10
 ; GFX9-NEXT:    s_lshl_b32 s2, s2, 2
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-NEXT:    s_add_i32 s1, s2, 0xc20
 ; GFX9-NEXT:    s_addk_i32 s2, 0xc60
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_endpgm
   %arrayidx80 = getelementptr inbounds [776 x float], [776 x float] addrspace(3)* @sgemm.lB, i32 0, i32 %add79
   store float %val, float addrspace(3)* %arrayidx80, align 4
   ret void
 }
 
 define amdgpu_kernel void @simple_write2_v4f32_superreg_align4(<4 x float> addrspace(3)* %out, <4 x float> addrspace(1)* %in) #0 {
 ; CI-LABEL: simple_write2_v4f32_superreg_align4:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; CI-NEXT:    s_load_dword s4, s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; CI-NEXT:    s_load_dword s4, s[0:1], 0x0
 ; CI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; CI-NEXT:    s_mov_b32 m0, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
 ; CI-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
 ; CI-NEXT:    v_mov_b32_e32 v1, s0
 ; CI-NEXT:    v_mov_b32_e32 v2, s1
 ; CI-NEXT:    v_mov_b32_e32 v3, s2
 ; CI-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; CI-NEXT:    v_mov_b32_e32 v1, s3
 ; CI-NEXT:    ds_write2_b32 v0, v3, v1 offset0:2 offset1:3
 ; CI-NEXT:    s_endpgm
 ;
 ; GFX9-ALIGNED-LABEL: simple_write2_v4f32_superreg_align4:
 ; GFX9-ALIGNED:       ; %bb.0:
-; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX9-ALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x24
+; GFX9-ALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX9-ALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x0
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-ALIGNED-NEXT:    v_lshl_add_u32 v0, v0, 4, s4
 ; GFX9-ALIGNED-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
 ; GFX9-ALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-ALIGNED-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX9-ALIGNED-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX9-ALIGNED-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX9-ALIGNED-NEXT:    v_mov_b32_e32 v4, s3
 ; GFX9-ALIGNED-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; GFX9-ALIGNED-NEXT:    ds_write2_b32 v0, v3, v4 offset0:2 offset1:3
 ; GFX9-ALIGNED-NEXT:    s_endpgm
 ;
 ; GFX9-UNALIGNED-LABEL: simple_write2_v4f32_superreg_align4:
 ; GFX9-UNALIGNED:       ; %bb.0:
-; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX9-UNALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x24
+; GFX9-UNALIGNED-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x8
+; GFX9-UNALIGNED-NEXT:    s_load_dword s4, s[0:1], 0x0
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    v_lshl_add_u32 v4, v0, 4, s4
 ; GFX9-UNALIGNED-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
 ; GFX9-UNALIGNED-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-UNALIGNED-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-UNALIGNED-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-UNALIGNED-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-UNALIGNED-NEXT:    v_mov_b32_e32 v3, s3

llvm/test/CodeGen/AMDGPU/flat-scratch.ll

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: store_load_sindex_kernel:
 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v0, 15
-; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s4, s3
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s5, 0
 ; GFX9-PAL-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-PAL-NEXT:    s_and_b32 s0, s0, 15
 ; GFX9-PAL-NEXT:    s_add_i32 s1, s1, 4
 ; GFX9-PAL-NEXT:    s_lshl_b32 s0, s0, 2

 ; GFX10-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX10-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX10-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX10-PAL-NEXT:    s_add_u32 s4, s4, s3
 ; GFX10-PAL-NEXT:    s_addc_u32 s5, s5, 0
 ; GFX10-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
 ; GFX10-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
-; GFX10-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX10-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX10-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX10-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX10-PAL-NEXT:    s_lshl_b32 s1, s1, 2
 ; GFX10-PAL-NEXT:    s_add_i32 s0, s0, 4
 ; GFX10-PAL-NEXT:    s_add_i32 s1, s1, 4
 ; GFX10-PAL-NEXT:    scratch_store_dword off, v0, s0

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: store_load_sindex_small_offset_kernel:
 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX9-PAL-NEXT:    s_mov_b32 vcc_hi, 0
-; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s4, s3
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s5, 0
 ; GFX9-PAL-NEXT:    scratch_load_dword v0, off, vcc_hi offset:4 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-PAL-NEXT:    s_and_b32 s0, s0, 15

 ; GFX1010-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX1010-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX1010-PAL-NEXT:    s_add_u32 s4, s4, s3
 ; GFX1010-PAL-NEXT:    s_addc_u32 s5, s5, 0
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
-; GFX1010-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX1010-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX1010-PAL-NEXT:    s_mov_b32 vcc_lo, 0
 ; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, vcc_lo offset:4 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s1, s1, 2

 ; GFX1030-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX1030-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX1030-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX1030-PAL-NEXT:    s_add_u32 s4, s4, s3
 ; GFX1030-PAL-NEXT:    s_addc_u32 s5, s5, 0
 ; GFX1030-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
 ; GFX1030-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
-; GFX1030-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX1030-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX1030-PAL-NEXT:    scratch_load_dword v0, off, off offset:4 glc dlc
 ; GFX1030-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1030-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX1030-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX1030-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX1030-PAL-NEXT:    s_lshl_b32 s1, s1, 2
 ; GFX1030-PAL-NEXT:    s_addk_i32 s0, 0x104

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: store_load_sindex_large_offset_kernel:
 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX9-PAL-NEXT:    s_mov_b32 vcc_hi, 0
-; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s4, s3
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s5, 0
 ; GFX9-PAL-NEXT:    scratch_load_dword v0, off, vcc_hi offset:4 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_lshl_b32 s1, s0, 2
 ; GFX9-PAL-NEXT:    s_and_b32 s0, s0, 15

 ; GFX1010-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX1010-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX1010-PAL-NEXT:    s_add_u32 s4, s4, s3
 ; GFX1010-PAL-NEXT:    s_addc_u32 s5, s5, 0
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
 ; GFX1010-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
-; GFX1010-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX1010-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX1010-PAL-NEXT:    s_mov_b32 vcc_lo, 0
 ; GFX1010-PAL-NEXT:    scratch_load_dword v0, off, vcc_lo offset:4 glc dlc
 ; GFX1010-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1010-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX1010-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1010-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX1010-PAL-NEXT:    s_lshl_b32 s1, s1, 2

 ; GFX1030-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX1030-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX1030-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX1030-PAL-NEXT:    s_add_u32 s4, s4, s3
 ; GFX1030-PAL-NEXT:    s_addc_u32 s5, s5, 0
 ; GFX1030-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
 ; GFX1030-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
-; GFX1030-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX1030-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX1030-PAL-NEXT:    scratch_load_dword v0, off, off offset:4 glc dlc
 ; GFX1030-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1030-PAL-NEXT:    v_mov_b32_e32 v0, 15
 ; GFX1030-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-PAL-NEXT:    s_and_b32 s1, s0, 15
 ; GFX1030-PAL-NEXT:    s_lshl_b32 s0, s0, 2
 ; GFX1030-PAL-NEXT:    s_lshl_b32 s1, s1, 2
 ; GFX1030-PAL-NEXT:    s_addk_i32 s0, 0x4004

 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX9-PAL-LABEL: store_load_vidx_sidx_offset:
 ; GFX9-PAL:       ; %bb.0: ; %bb
 ; GFX9-PAL-NEXT:    s_getpc_b64 s[4:5]
 ; GFX9-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX9-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, 4
-; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX9-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX9-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX9-PAL-NEXT:    s_add_u32 flat_scratch_lo, s4, s3
 ; GFX9-PAL-NEXT:    v_add_u32_e32 v0, s0, v0
 ; GFX9-PAL-NEXT:    s_addc_u32 flat_scratch_hi, s5, 0
 ; GFX9-PAL-NEXT:    v_lshl_add_u32 v0, v0, 2, v1
 ; GFX9-PAL-NEXT:    v_mov_b32_e32 v1, 15
 ; GFX9-PAL-NEXT:    scratch_store_dword v0, v1, off offset:1024
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    scratch_load_dword v0, v0, off offset:1024 glc
 ; GFX9-PAL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-PAL-NEXT:    s_endpgm
 ;
 ; GFX10-PAL-LABEL: store_load_vidx_sidx_offset:
 ; GFX10-PAL:       ; %bb.0: ; %bb
 ; GFX10-PAL-NEXT:    s_getpc_b64 s[4:5]
 ; GFX10-PAL-NEXT:    s_mov_b32 s4, s0
 ; GFX10-PAL-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; GFX10-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-PAL-NEXT:    s_and_b32 s5, s5, 0xffff
 ; GFX10-PAL-NEXT:    s_add_u32 s4, s4, s3
 ; GFX10-PAL-NEXT:    s_addc_u32 s5, s5, 0
 ; GFX10-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
 ; GFX10-PAL-NEXT:    s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
-; GFX10-PAL-NEXT:    s_load_dword s0, s[0:1], 0x24
+; GFX10-PAL-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-PAL-NEXT:    v_mov_b32_e32 v1, 15
 ; GFX10-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-PAL-NEXT:    v_add_nc_u32_e32 v0, s0, v0
 ; GFX10-PAL-NEXT:    v_lshl_add_u32 v0, v0, 2, 4
 ; GFX10-PAL-NEXT:    scratch_store_dword v0, v1, off offset:1024
 ; GFX10-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-PAL-NEXT:    scratch_load_dword v0, v0, off offset:1024 glc dlc
 ; GFX10-PAL-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-agent.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_agent_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_agent_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_agent_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_agent_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_agent_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_agent_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_agent_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_nontemporal_load_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc slc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_nontemporal_load_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
 ; SKIP-CACHE-INV-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc slc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2 glc slc
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_nontemporal_store_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2 glc slc

 ; GFX10-CU-NEXT:    flat_load_dword v2, v[1:2]
 ; GFX10-CU-NEXT:    v_add_co_ci_u32_e64 v1, s0, s3, 0, s0
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2 glc slc
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_nontemporal_store_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[1:2]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
 ; SKIP-CACHE-INV-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-singlethread.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_singlethread_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_singlethread_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-system.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_system_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_system_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_system_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_system_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_system_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_system_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_system_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-volatile.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_nontemporal_load_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_nontemporal_load_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
 ; SKIP-CACHE-INV-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1] glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_nontemporal_store_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_add_co_ci_u32_e64 v1, s0, s3, 0, s0
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_nontemporal_store_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[1:2]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
 ; SKIP-CACHE-INV-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_volatile_workgroup_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_volatile_workgroup_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
    i32 %in, i32* %out) {
 entry:
   store atomic volatile i32 %in, i32* %out syncscope("workgroup") release, align 4
   ret void
 }
 
 declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-wavefront.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_wavefront_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_acq_relc_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-workgroup.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    flat_load_dword v2, v[0:1]
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: flat_workgroup_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_swap v2, v[0:1], v2 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    flat_store_dword v[0:1], v2
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX10-CU-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s2, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s3, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s3
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v[0:1], v[2:3]

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonicmonotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    flat_store_dword v[0:1], v2
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: flat_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s2, 16
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s3, 0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v3, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s0
 ; SKIP-CACHE-INV-NEXT:    flat_atomic_cmpswap v2, v[0:1], v[2:3] glc

llvm/test/CodeGen/AMDGPU/memory-legalizer-global-agent.ll

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_agent_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_agent_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_agent_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_agent_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_agent_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_agent_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_agent_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll

 ; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_nontemporal_load_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v0, v0, s[0:1] slc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_nontemporal_load_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, 0

 ; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3] glc slc
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_nontemporal_store_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3] glc slc
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_nontemporal_store_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, 0
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

llvm/test/CodeGen/AMDGPU/memory-legalizer-global-singlethread.ll

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_singlethread_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_singlethread_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

llvm/test/CodeGen/AMDGPU/memory-legalizer-global-system.ll

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_system_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_system_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_system_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_relese_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_system_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_system_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_system_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    buffer_gl0_inv
 ; GFX10-CU-NEXT:    buffer_gl1_inv
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_system_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/memory-legalizer-global-volatile.ll

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_volatile_load_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v0, v0, s[0:1] glc dlc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_volatile_load_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_volatile_store_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_volatile_store_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, 0
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_volatile_workgroup_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_volatile_workgroup_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
    i32 %in, i32 addrspace(1)* %out) {
 entry:
   store atomic volatile i32 %in, i32 addrspace(1)* %out syncscope("workgroup") release, align 4
   ret void
 }
 
 declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-global-wavefront.ll

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_wavefront_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

llvm/test/CodeGen/AMDGPU/memory-legalizer-global-workgroup.ll

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_monotonic_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s7, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s6, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, off, s[4:7], 0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s5, s3

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_unordered_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_release_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: global_workgroup_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    global_atomic_swap v1, v0, v1, s[0:1] glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dword s4, s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_swap v0, off, s[0:3], 0 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v2, v[0:1], s[2:3] offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    global_atomic_cmpswap v0, v2, v[0:1], s[2:3] offset:16 glc
 ; GFX10-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: global_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s4
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s5
 ; SKIP-CACHE-INV-NEXT:    buffer_atomic_cmpswap v[0:1], off, s[0:3], 0 offset:16 glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

llvm/test/CodeGen/AMDGPU/memory-legalizer-local-agent.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_release_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_agent_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_agent_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

llvm/test/CodeGen/AMDGPU/memory-legalizer-local-nontemporal.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_nontemporal_load_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_nontemporal_load_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_nontemporal_store_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_nontemporal_store_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1

llvm/test/CodeGen/AMDGPU/memory-legalizer-local-singlethread.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_monotonic_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_release_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_seq_cst_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_release_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_release_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_singlethread_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_singlethread_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

llvm/test/CodeGen/AMDGPU/memory-legalizer-local-system.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_release_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_system_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_system_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

llvm/test/CodeGen/AMDGPU/memory-legalizer-local-volatile.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_volatile_load_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX10-CU-NEXT:    v_lshl_add_u32 v0, v0, 2, s2
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_volatile_load_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_volatile_store_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_volatile_store_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_volatile_workgroup_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_volatile_workgroup_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
    i32 %in, i32 addrspace(3)* %out) {
 entry:
   store atomic volatile i32 %in, i32 addrspace(3)* %out syncscope("workgroup") release, align 4
   ret void
 }
 
 declare i32 @llvm.amdgcn.workitem.id.x()

llvm/test/CodeGen/AMDGPU/memory-legalizer-local-wavefront.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_monotonic_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_release_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_seq_cst_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_release_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_release_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_wavefront_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

llvm/test/CodeGen/AMDGPU/memory-legalizer-local-workgroup.ll

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_store:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_unordered_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_read_b32 v0, v0
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v1, v0
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_load:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    ds_read_b32 v0, v0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v1, v0
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_unordered_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_unordered_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s1
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_store:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_store:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_atomicrmw:

 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v0, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_atomicrmw:

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-CU-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_ret_atomicrmw:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    ds_wrxchg_rtn_b32 v1, v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_b32 v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX90A-NOTTGSPLIT-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NOTTGSPLIT-NEXT:    v_mov_b32_e32 v0, s0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_monotonic_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_acquire_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_monotonic_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acquire_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_release_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_acq_rel_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

 ; GFX10-CU-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX10-CU-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; GFX10-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-CU-NEXT:    ds_write_b32 v0, v1
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 m0, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
-; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s3
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s0
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s2
+; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v2, s1
 ; SKIP-CACHE-INV-NEXT:    ds_cmpst_rtn_b32 v1, v0, v1, v2 offset:16
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    ds_write_b32 v0, v1
 ; SKIP-CACHE-INV-NEXT:    s_endpgm
 ;
 ; GFX90A-NOTTGSPLIT-LABEL: local_workgroup_one_as_seq_cst_seq_cst_ret_cmpxchg:
 ; GFX90A-NOTTGSPLIT:       ; %bb.0: ; %entry
 ; GFX90A-NOTTGSPLIT-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0

llvm/test/CodeGen/AMDGPU/memory-legalizer-private-nontemporal.ll

 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: private_nontemporal_load_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
-; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen glc slc
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: private_nontemporal_load_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
-; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen glc slc
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1

 ;
 ; SKIP-CACHE-INV-LABEL: private_nontemporal_store_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, v1, s[4:7], 0 offen glc slc
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

 ; SKIP-CACHE-INV-LABEL: private_nontemporal_store_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v1, v0, s[4:7], 0 offen glc slc
 ; SKIP-CACHE-INV-NEXT:    s_endpgm

llvm/test/CodeGen/AMDGPU/memory-legalizer-private-volatile.ll

 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: private_volatile_load_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
-; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s2
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s2, -1

 ; GFX10-CU-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-CU-NEXT:    s_endpgm
 ;
 ; SKIP-CACHE-INV-LABEL: private_volatile_load_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
-; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dword s2, s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s2, v0
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
 ; SKIP-CACHE-INV-NEXT:    buffer_load_dword v0, v0, s[4:7], 0 offen glc
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s3, 0xf000

 ;
 ; SKIP-CACHE-INV-LABEL: private_volatile_store_0:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v0, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v0, v1, s[4:7], 0 offen
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

 ; SKIP-CACHE-INV-LABEL: private_volatile_store_1:
 ; SKIP-CACHE-INV:       ; %bb.0: ; %entry
 ; SKIP-CACHE-INV-NEXT:    s_getpc_b64 s[4:5]
 ; SKIP-CACHE-INV-NEXT:    s_mov_b32 s4, s0
 ; SKIP-CACHE-INV-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_add_u32 s4, s4, s3
-; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x9
-; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0xb
+; SKIP-CACHE-INV-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
+; SKIP-CACHE-INV-NEXT:    s_load_dword s0, s[0:1], 0x2
 ; SKIP-CACHE-INV-NEXT:    s_addc_u32 s5, s5, 0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    s_load_dword s1, s[2:3], 0x0
 ; SKIP-CACHE-INV-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt lgkmcnt(0)
 ; SKIP-CACHE-INV-NEXT:    v_mov_b32_e32 v1, s1
 ; SKIP-CACHE-INV-NEXT:    buffer_store_dword v1, v0, s[4:7], 0 offen
 ; SKIP-CACHE-INV-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/store-local.128.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX9 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=hawaii -verify-machineinstrs < %s | FileCheck --check-prefix=GFX7 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=tahiti -verify-machineinstrs < %s | FileCheck --check-prefix=GFX6 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10 %s
 
 define amdgpu_kernel void @store_lds_v4i32(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX9-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v4i32:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX6-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX6-NEXT:    ds_write2_b64 v4, v[2:3], v[0:1] offset1:1
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX10-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align1(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align1:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    ds_write_b8 v0, v1 offset:12
 ; GFX9-NEXT:    ds_write_b8_d16_hi v0, v1 offset:14
 ; GFX9-NEXT:    ds_write_b8 v0, v2 offset:8
 ; GFX9-NEXT:    ds_write_b8_d16_hi v0, v2 offset:10

 ; GFX9-NEXT:    s_lshr_b32 s0, s4, 24
 ; GFX9-NEXT:    ds_write_b8 v0, v1 offset:1
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX9-NEXT:    ds_write_b8 v0, v1 offset:3
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align1:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    ds_write_b8 v0, v1 offset:12
 ; GFX7-NEXT:    ds_write_b8 v0, v2 offset:8
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1

 ; GFX7-NEXT:    s_lshr_b32 s0, s0, 16
 ; GFX7-NEXT:    ds_write_b8 v0, v1 offset:3
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    ds_write_b8 v0, v1 offset:2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v4i32_align1:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX6-NEXT:    ds_write_b8 v0, v1 offset:12
 ; GFX6-NEXT:    ds_write_b8 v0, v2 offset:8
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s1

 ; GFX6-NEXT:    ds_write_b8 v0, v1 offset:3
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX6-NEXT:    ds_write_b8 v0, v1 offset:2
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align1:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX10-NEXT:    s_lshr_b32 s3, s6, 24
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    s_lshr_b32 s0, s7, 8
 ; GFX10-NEXT:    s_lshr_b32 s2, s6, 8
 ; GFX10-NEXT:    s_lshr_b32 s6, s5, 8

 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 1
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align2(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align2:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    ds_write_b16 v0, v1 offset:12
 ; GFX9-NEXT:    ds_write_b16_d16_hi v0, v1 offset:14
 ; GFX9-NEXT:    ds_write_b16 v0, v2 offset:8
 ; GFX9-NEXT:    ds_write_b16_d16_hi v0, v2 offset:10
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    ds_write_b16 v0, v1 offset:4
 ; GFX9-NEXT:    ds_write_b16_d16_hi v0, v1 offset:6
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX9-NEXT:    ds_write_b16 v0, v1
 ; GFX9-NEXT:    ds_write_b16_d16_hi v0, v1 offset:2
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align2:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    ds_write_b16 v0, v1 offset:12
 ; GFX7-NEXT:    ds_write_b16 v0, v2 offset:8
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1

 ; GFX7-NEXT:    s_lshr_b32 s0, s0, 16
 ; GFX7-NEXT:    ds_write_b16 v0, v1 offset:6
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    ds_write_b16 v0, v1 offset:2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v4i32_align2:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX6-NEXT:    ds_write_b16 v0, v1 offset:12
 ; GFX6-NEXT:    ds_write_b16 v0, v2 offset:8
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s1

 ; GFX6-NEXT:    ds_write_b16 v0, v1 offset:6
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX6-NEXT:    ds_write_b16 v0, v1 offset:2
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align2:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX10-NEXT:    ds_write_b16 v0, v1 offset:12
 ; GFX10-NEXT:    ds_write_b16_d16_hi v0, v1 offset:14
 ; GFX10-NEXT:    ds_write_b16 v0, v2 offset:8
 ; GFX10-NEXT:    ds_write_b16_d16_hi v0, v2 offset:10
 ; GFX10-NEXT:    ds_write_b16 v0, v3 offset:4
 ; GFX10-NEXT:    ds_write_b16_d16_hi v0, v3 offset:6
 ; GFX10-NEXT:    ds_write_b16 v0, v4
 ; GFX10-NEXT:    ds_write_b16_d16_hi v0, v4 offset:2
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 2
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align4(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align4:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX9-NEXT:    ds_write2_b32 v0, v3, v1 offset0:2 offset1:3
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align4:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX7-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX7-NEXT:    ds_write2_b32 v0, v1, v2 offset0:2 offset1:3
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v4i32_align4:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX6-NEXT:    ds_write2_b32 v0, v2, v1 offset1:1
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX6-NEXT:    ds_write2_b32 v0, v2, v1 offset0:2 offset1:3
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align4:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s7
 ; GFX10-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; GFX10-NEXT:    ds_write2_b32 v0, v3, v4 offset0:2 offset1:3
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align8(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align8:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX9-NEXT:    ds_write2_b64 v4, v[0:1], v[2:3] offset1:1
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align8:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    ds_write2_b64 v4, v[0:1], v[2:3] offset1:1
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v4i32_align8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX6-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX6-NEXT:    ds_write2_b64 v4, v[2:3], v[0:1] offset1:1
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align8:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX10-NEXT:    ds_write2_b64 v4, v[0:1], v[2:3] offset1:1
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v4i32_align16(<4 x i32> addrspace(3)* %out, <4 x i32> %x) {
 ; GFX9-LABEL: store_lds_v4i32_align16:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX9-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v4i32_align16:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX7-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v4i32_align16:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX6-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s7
 ; GFX6-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX6-NEXT:    ds_write2_b64 v4, v[2:3], v[0:1] offset1:1
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v4i32_align16:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s7
 ; GFX10-NEXT:    ds_write_b128 v4, v[0:3]
 ; GFX10-NEXT:    s_endpgm
   store <4 x i32> %x, <4 x i32> addrspace(3)* %out, align 16
   ret void
 }

llvm/test/CodeGen/AMDGPU/store-local.96.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX9 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=hawaii -verify-machineinstrs < %s | FileCheck --check-prefix=GFX7 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=tahiti -verify-machineinstrs < %s | FileCheck --check-prefix=GFX6 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10 %s
 
 define amdgpu_kernel void @store_lds_v3i32(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX9-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX7-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v3i32:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX6-NEXT:    ds_write_b32 v2, v1 offset:8
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX6-NEXT:    ds_write_b64 v2, v[0:1]
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX10-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align1(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align1:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX9-NEXT:    ds_write_b8 v0, v1 offset:8
 ; GFX9-NEXT:    ds_write_b8_d16_hi v0, v1 offset:10
 ; GFX9-NEXT:    ds_write_b8 v0, v2 offset:4
 ; GFX9-NEXT:    ds_write_b8_d16_hi v0, v2 offset:6

 ; GFX9-NEXT:    s_lshr_b32 s0, s4, 24
 ; GFX9-NEXT:    ds_write_b8 v0, v1 offset:1
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX9-NEXT:    ds_write_b8 v0, v1 offset:3
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align1:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX7-NEXT:    ds_write_b8 v0, v1 offset:8
 ; GFX7-NEXT:    ds_write_b8 v0, v2 offset:4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0

 ; GFX7-NEXT:    s_lshr_b32 s0, s0, 16
 ; GFX7-NEXT:    ds_write_b8 v0, v1 offset:3
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    ds_write_b8 v0, v1 offset:2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v3i32_align1:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX6-NEXT:    ds_write_b8 v0, v1 offset:8
 ; GFX6-NEXT:    ds_write_b8 v0, v2 offset:4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s0

 ; GFX6-NEXT:    ds_write_b8 v0, v1 offset:3
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX6-NEXT:    ds_write_b8 v0, v1 offset:2
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align1:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s4
 ; GFX10-NEXT:    s_lshr_b32 s0, s6, 8
 ; GFX10-NEXT:    s_lshr_b32 s1, s6, 24
 ; GFX10-NEXT:    s_lshr_b32 s2, s5, 8

 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 1
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align2(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align2:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX9-NEXT:    ds_write_b16 v0, v1 offset:8
 ; GFX9-NEXT:    ds_write_b16_d16_hi v0, v1 offset:10
 ; GFX9-NEXT:    ds_write_b16 v0, v2 offset:4
 ; GFX9-NEXT:    ds_write_b16_d16_hi v0, v2 offset:6
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX9-NEXT:    ds_write_b16 v0, v1
 ; GFX9-NEXT:    ds_write_b16_d16_hi v0, v1 offset:2
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align2:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX7-NEXT:    ds_write_b16 v0, v1 offset:8
 ; GFX7-NEXT:    ds_write_b16 v0, v2 offset:4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    s_lshr_b32 s2, s2, 16
 ; GFX7-NEXT:    ds_write_b16 v0, v1
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX7-NEXT:    s_lshr_b32 s1, s1, 16
 ; GFX7-NEXT:    ds_write_b16 v0, v1 offset:10
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    s_lshr_b32 s0, s0, 16
 ; GFX7-NEXT:    ds_write_b16 v0, v1 offset:6
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    ds_write_b16 v0, v1 offset:2
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v3i32_align2:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX6-NEXT:    ds_write_b16 v0, v1 offset:8
 ; GFX6-NEXT:    ds_write_b16 v0, v2 offset:4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX6-NEXT:    s_lshr_b32 s2, s2, 16
 ; GFX6-NEXT:    ds_write_b16 v0, v1
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX6-NEXT:    s_lshr_b32 s1, s1, 16
 ; GFX6-NEXT:    ds_write_b16 v0, v1 offset:10
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX6-NEXT:    s_lshr_b32 s0, s0, 16
 ; GFX6-NEXT:    ds_write_b16 v0, v1 offset:6
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX6-NEXT:    ds_write_b16 v0, v1 offset:2
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align2:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s4
 ; GFX10-NEXT:    ds_write_b16 v0, v1 offset:8
 ; GFX10-NEXT:    ds_write_b16_d16_hi v0, v1 offset:10
 ; GFX10-NEXT:    ds_write_b16 v0, v2 offset:4
 ; GFX10-NEXT:    ds_write_b16_d16_hi v0, v2 offset:6
 ; GFX10-NEXT:    ds_write_b16 v0, v3
 ; GFX10-NEXT:    ds_write_b16_d16_hi v0, v3 offset:2
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 2
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align4(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align4:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX9-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; GFX9-NEXT:    ds_write_b32 v0, v3 offset:8
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align4:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s0
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX7-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX7-NEXT:    ds_write_b32 v0, v1 offset:8
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v3i32_align4:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX6-NEXT:    ds_write2_b32 v0, v2, v1 offset1:1
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX6-NEXT:    ds_write_b32 v0, v1 offset:8
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align4:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX10-NEXT:    ds_write2_b32 v0, v1, v2 offset1:1
 ; GFX10-NEXT:    ds_write_b32 v0, v3 offset:8
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align8(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align8:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    ds_write_b32 v2, v3 offset:8
 ; GFX9-NEXT:    ds_write_b64 v2, v[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align8:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX7-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX7-NEXT:    ds_write_b32 v2, v1 offset:8
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    ds_write_b64 v2, v[0:1]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v3i32_align8:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX6-NEXT:    ds_write_b32 v2, v1 offset:8
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX6-NEXT:    ds_write_b64 v2, v[0:1]
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align8:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    ds_write_b32 v2, v3 offset:8
 ; GFX10-NEXT:    ds_write_b64 v2, v[0:1]
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @store_lds_v3i32_align16(<3 x i32> addrspace(3)* %out, <3 x i32> %x) {
 ; GFX9-LABEL: store_lds_v3i32_align16:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX9-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX7-LABEL: store_lds_v3i32_align16:
 ; GFX7:       ; %bb.0:
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
-; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x4
+; GFX7-NEXT:    s_load_dword s0, s[0:1], 0x0
 ; GFX7-NEXT:    s_mov_b32 m0, -1
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX7-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX6-LABEL: store_lds_v3i32_align16:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x4
 ; GFX6-NEXT:    s_mov_b32 m0, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s4
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX6-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX6-NEXT:    ds_write_b32 v2, v1 offset:8
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX6-NEXT:    ds_write_b64 v2, v[0:1]
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: store_lds_v3i32_align16:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x10
+; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX10-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX10-NEXT:    ds_write_b96 v3, v[0:2]
 ; GFX10-NEXT:    s_endpgm
   store <3 x i32> %x, <3 x i32> addrspace(3)* %out, align 16
   ret void
 }