diff --git a/llvm/test/CodeGen/AMDGPU/fmul-2-combine-multi-use.ll b/llvm/test/CodeGen/AMDGPU/fmul-2-combine-multi-use.ll
--- a/llvm/test/CodeGen/AMDGPU/fmul-2-combine-multi-use.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmul-2-combine-multi-use.ll
@@ -1,10 +1,10 @@
-; XUN: llc -mtriple=amdgcn-amd-amdhsa -denormal-fp-math-f32=preserve-sign -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SI %s
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tonga -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,SIVI,VI-DENORM %s
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tonga -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,SIVI,VI-FLUSH %s
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_10,GFX10-DENORM %s
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_10,GFX10-FLUSH %s
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_10,GFX11-DENORM %s
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10,GFX8_10,GFX10-FLUSH %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tonga -denormal-fp-math-f32=preserve-sign  -verify-machineinstrs < %s | FileCheck -check-prefixes=VI,VI-DENORM %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tonga -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign  -verify-machineinstrs < %s | FileCheck -check-prefixes=VI,VI-FLUSH %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -denormal-fp-math-f32=preserve-sign  -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10,GFX10-DENORM %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign  -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10,GFX10-FLUSH %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -denormal-fp-math-f32=preserve-sign  -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-DENORM %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -denormal-fp-math=preserve-sign -denormal-fp-math-f32=preserve-sign  -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-FLUSH %s
 
 ; Make sure (fmul (fadd x, x), c) -> (fmul x, (fmul 2.0, c)) doesn't
 ; make add an instruction if the fadd has more than one use.
@@ -12,22 +12,56 @@
 declare half @llvm.fabs.f16(half) #1
 declare float @llvm.fabs.f32(float) #1
 
-; GCN-LABEL: {{^}}multiple_fadd_use_test_f32:
-; SI: v_max_legacy_f32_e64 [[A16:v[0-9]+]],
-; SI: v_add_f32_e32 [[A17:v[0-9]+]], [[A16]], [[A16]]
-; SI: v_mul_f32_e32 [[A18:v[0-9]+]], [[A17]], [[A17]]
-; SI: v_mad_f32 [[A20:v[0-9]+]], -[[A18]], [[A17]], 1.0
-; SI: buffer_store_dword [[A20]]
-
-; GFX8_10: v_add_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -1.0
-; GFX8_10: v_add_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -1.0
-; GFX8_10: v_cmp_gt_f32_e64 {{vcc|vcc_lo}}, |v{{[0-9]+}}|, |v{{[0-9]+}}|
-; GFX8_10: v_cndmask_b32_e32
-; GFX8_10: v_add_f32_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, |v{{[0-9]+}}|
-; GFX8_10: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
-; VI:      v_mad_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
-; GFX10:   v_fma_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
 define amdgpu_kernel void @multiple_fadd_use_test_f32(ptr addrspace(1) %out, float %x, float %y, float %z) #0 {
+; VI-LABEL: multiple_fadd_use_test_f32:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    v_add_f32_e64 v0, s3, -1.0
+; VI-NEXT:    v_add_f32_e64 v1, s2, -1.0
+; VI-NEXT:    v_cmp_gt_f32_e64 vcc, |v0|, |v1|
+; VI-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
+; VI-NEXT:    v_add_f32_e64 v0, |v0|, |v0|
+; VI-NEXT:    v_mul_f32_e32 v1, v0, v0
+; VI-NEXT:    v_mad_f32 v2, -v1, v0, 1.0
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    flat_store_dword v[0:1], v2
+; VI-NEXT:    s_endpgm
+;
+; GFX10-LABEL: multiple_fadd_use_test_f32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-NEXT:    v_mov_b32_e32 v2, 0
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_add_f32_e64 v0, s3, -1.0
+; GFX10-NEXT:    v_add_f32_e64 v1, s2, -1.0
+; GFX10-NEXT:    v_cmp_gt_f32_e64 vcc_lo, |v0|, |v1|
+; GFX10-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc_lo
+; GFX10-NEXT:    v_add_f32_e64 v0, |v0|, |v0|
+; GFX10-NEXT:    v_mul_f32_e32 v1, v0, v0
+; GFX10-NEXT:    v_fma_f32 v0, -v1, v0, 1.0
+; GFX10-NEXT:    global_store_dword v2, v0, s[0:1]
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: multiple_fadd_use_test_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX11-NEXT:    v_mov_b32_e32 v2, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_add_f32_e64 v0, s3, -1.0
+; GFX11-NEXT:    v_add_f32_e64 v1, s2, -1.0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX11-NEXT:    v_cmp_gt_f32_e64 vcc_lo, |v0|, |v1|
+; GFX11-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc_lo
+; GFX11-NEXT:    v_add_f32_e64 v0, |v0|, |v0|
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-NEXT:    v_mul_f32_e32 v1, v0, v0
+; GFX11-NEXT:    v_fma_f32 v0, -v1, v0, 1.0
+; GFX11-NEXT:    global_store_b32 v2, v0, s[0:1]
+; GFX11-NEXT:    s_nop 0
+; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT:    s_endpgm
   %a11 = fadd float %y, -1.0
   %a12 = call float @llvm.fabs.f32(float %a11)
   %a13 = fadd float %x, -1.0
@@ -42,14 +76,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}multiple_use_fadd_fmac_f32:
-; GCN-DAG:   v_add_f32_e64 [[MUL2:v[0-9]+]], [[X:s[0-9]+]], s{{[0-9]+}}
-; SIVI-DAG:  v_mac_f32_e64 [[MAD:v[0-9]+]], [[X]], 2.0
-; GFX10-DAG: v_fma_f32 [[MAD:v[0-9]+]], [[X]], 2.0, s{{[0-9]+}}
-; GCN-DAG:   buffer_store_{{dword|b32}} [[MUL2]]
-; GCN-DAG:   buffer_store_{{dword|b32}} [[MAD]]
-; GCN:       s_endpgm
 define amdgpu_kernel void @multiple_use_fadd_fmac_f32(ptr addrspace(1) %out, float %x, [8 x i32], float %y) #0 {
+; VI-LABEL: multiple_use_fadd_fmac_f32:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-NEXT:    s_load_dword s6, s[4:5], 0x8
+; VI-NEXT:    s_load_dword s3, s[4:5], 0x2c
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    s_add_u32 s2, s0, 4
+; VI-NEXT:    v_add_f32_e64 v2, s6, s6
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_mov_b32_e32 v3, s3
+; VI-NEXT:    s_addc_u32 s3, s1, 0
+; VI-NEXT:    flat_store_dword v[0:1], v2
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_mov_b32_e32 v0, s2
+; VI-NEXT:    v_mac_f32_e64 v3, s6, 2.0
+; VI-NEXT:    v_mov_b32_e32 v1, s3
+; VI-NEXT:    flat_store_dword v[0:1], v3
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    s_endpgm
+;
+; GFX10-LABEL: multiple_use_fadd_fmac_f32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_clause 0x2
+; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-NEXT:    s_load_dword s3, s[4:5], 0x2c
+; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_add_f32_e64 v1, s2, s2
+; GFX10-NEXT:    v_fma_f32 v2, s2, 2.0, s3
+; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_store_dword v0, v2, s[0:1] offset:4
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: multiple_use_fadd_fmac_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x2
+; GFX11-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-NEXT:    s_load_b32 s3, s[0:1], 0x2c
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_add_f32_e64 v1, s2, s2
+; GFX11-NEXT:    v_fma_f32 v2, s2, 2.0, s3
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_store_b32 v0, v2, s[0:1] offset:4 dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_nop 0
+; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT:    s_endpgm
   %out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
   %mul2 = fmul fast float %x, 2.0
   %mad = fadd fast float %mul2, %y
@@ -58,14 +139,53 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}multiple_use_fadd_fmad_f32:
-; GCN-DAG:   v_add_f32_e64 [[MUL2:v[0-9]+]], |[[X:s[0-9]+]]|, |s{{[0-9]+}}|
-; SIVI-DAG:  v_mad_f32 [[MAD:v[0-9]+]], |[[X]]|, 2.0, v{{[0-9]+}}
-; GFX10-DAG: v_fma_f32 [[MAD:v[0-9]+]], |[[X]]|, 2.0, s{{[0-9]+}}
-; GCN-DAG:   buffer_store_{{dword|b32}} [[MUL2]]
-; GCN-DAG:   buffer_store_{{dword|b32}} [[MAD]]
-; GCN:       s_endpgm
 define amdgpu_kernel void @multiple_use_fadd_fmad_f32(ptr addrspace(1) %out, float %x, float %y) #0 {
+; VI-LABEL: multiple_use_fadd_fmad_f32:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    s_add_u32 s4, s0, 4
+; VI-NEXT:    v_add_f32_e64 v2, |s2|, |s2|
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_mov_b32_e32 v3, s3
+; VI-NEXT:    s_addc_u32 s5, s1, 0
+; VI-NEXT:    flat_store_dword v[0:1], v2
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_mov_b32_e32 v0, s4
+; VI-NEXT:    v_mad_f32 v3, |s2|, 2.0, v3
+; VI-NEXT:    v_mov_b32_e32 v1, s5
+; VI-NEXT:    flat_store_dword v[0:1], v3
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    s_endpgm
+;
+; GFX10-LABEL: multiple_use_fadd_fmad_f32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_add_f32_e64 v1, |s2|, |s2|
+; GFX10-NEXT:    v_fma_f32 v2, |s2|, 2.0, s3
+; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_store_dword v0, v2, s[0:1] offset:4
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: multiple_use_fadd_fmad_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x0
+; GFX11-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_add_f32_e64 v1, |s2|, |s2|
+; GFX11-NEXT:    v_fma_f32 v2, |s2|, 2.0, s3
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_store_b32 v0, v2, s[0:1] offset:4 dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_nop 0
+; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT:    s_endpgm
   %out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
   %x.abs = call float @llvm.fabs.f32(float %x)
   %mul2 = fmul fast float %x.abs, 2.0
@@ -75,12 +195,59 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}multiple_use_fadd_multi_fmad_f32:
-; SIVI:  v_mad_f32 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, v{{[0-9]+}}
-; SIVI:  v_mad_f32 {{v[0-9]+}}, |[[X]]|, 2.0, v{{[0-9]+}}
-; GFX10: v_fma_f32 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, {{s[0-9]+}}
-; GFX10: v_fma_f32 {{v[0-9]+}}, |[[X]]|, 2.0, {{s[0-9]+}}
 define amdgpu_kernel void @multiple_use_fadd_multi_fmad_f32(ptr addrspace(1) %out, float %x, float %y, float %z) #0 {
+; VI-LABEL: multiple_use_fadd_multi_fmad_f32:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x0
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x8
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    s_add_u32 s4, s6, 4
+; VI-NEXT:    v_mov_b32_e32 v0, s1
+; VI-NEXT:    v_mov_b32_e32 v1, s2
+; VI-NEXT:    v_mad_f32 v2, |s0|, 2.0, v0
+; VI-NEXT:    v_mad_f32 v3, |s0|, 2.0, v1
+; VI-NEXT:    v_mov_b32_e32 v0, s6
+; VI-NEXT:    v_mov_b32_e32 v1, s7
+; VI-NEXT:    s_addc_u32 s5, s7, 0
+; VI-NEXT:    flat_store_dword v[0:1], v2
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_mov_b32_e32 v0, s4
+; VI-NEXT:    v_mov_b32_e32 v1, s5
+; VI-NEXT:    flat_store_dword v[0:1], v3
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    s_endpgm
+;
+; GFX10-LABEL: multiple_use_fadd_multi_fmad_f32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x8
+; GFX10-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x0
+; GFX10-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_fma_f32 v1, |s0|, 2.0, s1
+; GFX10-NEXT:    v_fma_f32 v2, |s0|, 2.0, s2
+; GFX10-NEXT:    global_store_dword v0, v1, s[6:7]
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    global_store_dword v0, v2, s[6:7] offset:4
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: multiple_use_fadd_multi_fmad_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b128 s[4:7], s[0:1], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_fma_f32 v1, |s4|, 2.0, s5
+; GFX11-NEXT:    v_fma_f32 v2, |s4|, 2.0, s6
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_store_b32 v0, v2, s[0:1] offset:4 dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_nop 0
+; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT:    s_endpgm
   %out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
   %x.abs = call float @llvm.fabs.f32(float %x)
   %mul2 = fmul fast float %x.abs, 2.0
@@ -91,11 +258,47 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}fmul_x2_xn2_f32:
-; GCN: v_mul_f32_e64 [[TMP0:v[0-9]+]], [[X:s[0-9]+]], -4.0
-; GCN: v_mul_f32_e32 [[RESULT:v[0-9]+]], [[X]], [[TMP0]]
-; GCN: buffer_store_{{dword|b32}} [[RESULT]]
 define amdgpu_kernel void @fmul_x2_xn2_f32(ptr addrspace(1) %out, float %x, float %y) #0 {
+; VI-LABEL: fmul_x2_xn2_f32:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_load_dword s2, s[4:5], 0x8
+; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    v_mul_f32_e64 v0, s2, -4.0
+; VI-NEXT:    v_mul_f32_e32 v2, s2, v0
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    flat_store_dword v[0:1], v2
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    s_endpgm
+;
+; GFX10-LABEL: fmul_x2_xn2_f32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mul_f32_e64 v0, s2, -4.0
+; GFX10-NEXT:    v_mul_f32_e32 v0, s2, v0
+; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fmul_x2_xn2_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_mul_f32_e64 v0, s2, -4.0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mul_f32 v0, s2, v0
+; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_nop 0
+; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT:    s_endpgm
   %out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
   %mul2 = fmul fast float %x, 2.0
   %muln2 = fmul fast float %x, -2.0
@@ -104,13 +307,48 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}fmul_x2_xn3_f32:
-; SIVI:  v_mov_b32_e32 [[K:v[0-9]+]], 0xc0c00000
-; SIVI:  v_mul_f32_e32 [[TMP0:v[0-9]+]], [[X:s[0-9]+]], [[K]]
-; GFX10: v_mul_f32_e64 [[TMP0:v[0-9]+]], 0xc0c00000, [[X:s[0-9]+]]
-; GCN:   v_mul_f32_e32 [[RESULT:v[0-9]+]], [[X]], [[TMP0]]
-; GCN:   buffer_store_{{dword|b32}} [[RESULT]]
 define amdgpu_kernel void @fmul_x2_xn3_f32(ptr addrspace(1) %out, float %x, float %y) #0 {
+; VI-LABEL: fmul_x2_xn3_f32:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_load_dword s2, s[4:5], 0x8
+; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-NEXT:    v_mov_b32_e32 v0, 0xc0c00000
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    v_mul_f32_e32 v0, s2, v0
+; VI-NEXT:    v_mul_f32_e32 v2, s2, v0
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    flat_store_dword v[0:1], v2
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    s_endpgm
+;
+; GFX10-LABEL: fmul_x2_xn3_f32:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mul_f32_e64 v0, 0xc0c00000, s2
+; GFX10-NEXT:    v_mul_f32_e32 v0, s2, v0
+; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fmul_x2_xn3_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_mul_f32_e64 v0, 0xc0c00000, s2
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mul_f32 v0, s2, v0
+; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_nop 0
+; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT:    s_endpgm
   %out.gep.1 = getelementptr float, ptr addrspace(1) %out, i32 1
   %mul2 = fmul fast float %x, 2.0
   %muln2 = fmul fast float %x, -3.0
@@ -119,19 +357,125 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}multiple_fadd_use_test_f16:
-; GFX8_10:      v_add_f16_e64 v{{[0-9]+}}, s{{[0-9]+}}, -1.0
-; GFX8_10:      v_add_f16_e64 v{{[0-9]+}}, s{{[0-9]+}}, -1.0
-; GFX8_10:      v_cmp_gt_f16_e64 {{vcc|vcc_lo}}, |v{{[0-9]+}}|, |v{{[0-9]+}}|
-; GFX8_10:      v_cndmask_b32_e32
-; GFX8_10:      v_add_f16_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, |v{{[0-9]+}}|
-; GFX8_10:      v_mul_f16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
-; VI-FLUSH:     v_mad_f16 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
-; VI-DENORM:    v_fma_f16 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
-; GFX10-DENORM: v_fma_f16 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
-; GFX11-DENORM: v_fma_f16 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, 1.0
-; GFX10-FLUSH:  v_sub_f16_e32 v{{[0-9]+}}, 1.0, v{{[0-9]+}}
 define amdgpu_kernel void @multiple_fadd_use_test_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg, i16 zeroext %z.arg) #0 {
+; VI-DENORM-LABEL: multiple_fadd_use_test_f16:
+; VI-DENORM:       ; %bb.0:
+; VI-DENORM-NEXT:    s_load_dword s2, s[4:5], 0x8
+; VI-DENORM-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-DENORM-NEXT:    s_lshr_b32 s3, s2, 16
+; VI-DENORM-NEXT:    v_add_f16_e64 v0, s2, -1.0
+; VI-DENORM-NEXT:    v_add_f16_e64 v1, s3, -1.0
+; VI-DENORM-NEXT:    v_cmp_gt_f16_e64 vcc, |v1|, |v0|
+; VI-DENORM-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; VI-DENORM-NEXT:    v_add_f16_e64 v0, |v0|, |v0|
+; VI-DENORM-NEXT:    v_mul_f16_e32 v1, v0, v0
+; VI-DENORM-NEXT:    v_fma_f16 v2, -v1, v0, 1.0
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s0
+; VI-DENORM-NEXT:    v_mov_b32_e32 v1, s1
+; VI-DENORM-NEXT:    flat_store_short v[0:1], v2
+; VI-DENORM-NEXT:    s_endpgm
+;
+; VI-FLUSH-LABEL: multiple_fadd_use_test_f16:
+; VI-FLUSH:       ; %bb.0:
+; VI-FLUSH-NEXT:    s_load_dword s2, s[4:5], 0x8
+; VI-FLUSH-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-FLUSH-NEXT:    s_lshr_b32 s3, s2, 16
+; VI-FLUSH-NEXT:    v_add_f16_e64 v0, s2, -1.0
+; VI-FLUSH-NEXT:    v_add_f16_e64 v1, s3, -1.0
+; VI-FLUSH-NEXT:    v_cmp_gt_f16_e64 vcc, |v1|, |v0|
+; VI-FLUSH-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; VI-FLUSH-NEXT:    v_add_f16_e64 v0, |v0|, |v0|
+; VI-FLUSH-NEXT:    v_mul_f16_e32 v1, v0, v0
+; VI-FLUSH-NEXT:    v_mad_f16 v2, -v1, v0, 1.0
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s0
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v1, s1
+; VI-FLUSH-NEXT:    flat_store_short v[0:1], v2
+; VI-FLUSH-NEXT:    s_endpgm
+;
+; GFX10-DENORM-LABEL: multiple_fadd_use_test_f16:
+; GFX10-DENORM:       ; %bb.0:
+; GFX10-DENORM-NEXT:    s_load_dword s0, s[4:5], 0x8
+; GFX10-DENORM-NEXT:    v_mov_b32_e32 v2, 0
+; GFX10-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_lshr_b32 s1, s0, 16
+; GFX10-DENORM-NEXT:    v_add_f16_e64 v0, s0, -1.0
+; GFX10-DENORM-NEXT:    v_add_f16_e64 v1, s1, -1.0
+; GFX10-DENORM-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-DENORM-NEXT:    v_cmp_gt_f16_e64 vcc_lo, |v1|, |v0|
+; GFX10-DENORM-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX10-DENORM-NEXT:    v_add_f16_e64 v0, |v0|, |v0|
+; GFX10-DENORM-NEXT:    v_mul_f16_e32 v1, v0, v0
+; GFX10-DENORM-NEXT:    v_fma_f16 v0, -v1, v0, 1.0
+; GFX10-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-DENORM-NEXT:    global_store_short v2, v0, s[0:1]
+; GFX10-DENORM-NEXT:    s_endpgm
+;
+; GFX10-FLUSH-LABEL: multiple_fadd_use_test_f16:
+; GFX10-FLUSH:       ; %bb.0:
+; GFX10-FLUSH-NEXT:    s_load_dword s0, s[4:5], 0x8
+; GFX10-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-FLUSH-NEXT:    s_lshr_b32 s1, s0, 16
+; GFX10-FLUSH-NEXT:    v_add_f16_e64 v0, s0, -1.0
+; GFX10-FLUSH-NEXT:    v_add_f16_e64 v1, s1, -1.0
+; GFX10-FLUSH-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-FLUSH-NEXT:    v_cmp_gt_f16_e64 vcc_lo, |v1|, |v0|
+; GFX10-FLUSH-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX10-FLUSH-NEXT:    v_add_f16_e64 v0, |v0|, |v0|
+; GFX10-FLUSH-NEXT:    v_mul_f16_e32 v1, v0, v0
+; GFX10-FLUSH-NEXT:    v_mul_f16_e32 v0, v1, v0
+; GFX10-FLUSH-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-FLUSH-NEXT:    v_sub_f16_e32 v0, 1.0, v0
+; GFX10-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-FLUSH-NEXT:    global_store_short v1, v0, s[0:1]
+; GFX10-FLUSH-NEXT:    s_endpgm
+;
+; GFX11-DENORM-LABEL: multiple_fadd_use_test_f16:
+; GFX11-DENORM:       ; %bb.0:
+; GFX11-DENORM-NEXT:    s_clause 0x1
+; GFX11-DENORM-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-DENORM-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-DENORM-NEXT:    v_mov_b32_e32 v2, 0
+; GFX11-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-DENORM-NEXT:    s_lshr_b32 s3, s2, 16
+; GFX11-DENORM-NEXT:    v_add_f16_e64 v0, s2, -1.0
+; GFX11-DENORM-NEXT:    v_add_f16_e64 v1, s3, -1.0
+; GFX11-DENORM-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX11-DENORM-NEXT:    v_cmp_gt_f16_e64 vcc_lo, |v1|, |v0|
+; GFX11-DENORM-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX11-DENORM-NEXT:    v_add_f16_e64 v0, |v0|, |v0|
+; GFX11-DENORM-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-DENORM-NEXT:    v_mul_f16_e32 v1, v0, v0
+; GFX11-DENORM-NEXT:    v_fma_f16 v0, -v1, v0, 1.0
+; GFX11-DENORM-NEXT:    global_store_b16 v2, v0, s[0:1]
+; GFX11-DENORM-NEXT:    s_nop 0
+; GFX11-DENORM-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-DENORM-NEXT:    s_endpgm
+;
+; GFX11-FLUSH-LABEL: multiple_fadd_use_test_f16:
+; GFX11-FLUSH:       ; %bb.0:
+; GFX11-FLUSH-NEXT:    s_clause 0x1
+; GFX11-FLUSH-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-FLUSH-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FLUSH-NEXT:    s_lshr_b32 s3, s2, 16
+; GFX11-FLUSH-NEXT:    v_add_f16_e64 v0, s2, -1.0
+; GFX11-FLUSH-NEXT:    v_add_f16_e64 v1, s3, -1.0
+; GFX11-FLUSH-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX11-FLUSH-NEXT:    v_cmp_gt_f16_e64 vcc_lo, |v1|, |v0|
+; GFX11-FLUSH-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc_lo
+; GFX11-FLUSH-NEXT:    v_add_f16_e64 v0, |v0|, |v0|
+; GFX11-FLUSH-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-FLUSH-NEXT:    v_mul_f16_e32 v1, v0, v0
+; GFX11-FLUSH-NEXT:    v_mul_f16_e32 v0, v1, v0
+; GFX11-FLUSH-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-FLUSH-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX11-FLUSH-NEXT:    v_sub_f16_e32 v0, 1.0, v0
+; GFX11-FLUSH-NEXT:    global_store_b16 v1, v0, s[0:1]
+; GFX11-FLUSH-NEXT:    s_nop 0
+; GFX11-FLUSH-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-FLUSH-NEXT:    s_endpgm
   %x = bitcast i16 %x.arg to half
   %y = bitcast i16 %y.arg to half
   %z = bitcast i16 %z.arg to half
@@ -149,19 +493,117 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}multiple_use_fadd_fmac_f16:
-; GCN-DAG: v_add_f16_e64 [[MUL2:v[0-9]+]], [[X:s[0-9]+]], s{{[0-9]+}}
-
-; VI-FLUSH-DAG:     v_mac_f16_e64 [[MAD:v[0-9]+]], [[X]], 2.0
-; VI-DENORM-DAG:    v_fma_f16 [[MAD:v[0-9]+]], [[X]], 2.0, v{{[0-9]+}}
-; GFX10-FLUSH-DAG:  v_add_f16_e32 [[MAD:v[0-9]+]], s{{[0-9]+}}, [[MUL2]]
-; GFX10-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], [[X]], 2.0, s{{[0-9]+}}
-; GFX11-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], [[X]], 2.0, s{{[0-9]+}}
-
-; GCN-DAG: buffer_store_{{short|b16}} [[MUL2]]
-; GCN-DAG: buffer_store_{{short|b16}} [[MAD]]
-; GCN: s_endpgm
 define amdgpu_kernel void @multiple_use_fadd_fmac_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg) #0 {
+; VI-DENORM-LABEL: multiple_use_fadd_fmac_f16:
+; VI-DENORM:       ; %bb.0:
+; VI-DENORM-NEXT:    s_load_dword s6, s[4:5], 0x8
+; VI-DENORM-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-DENORM-NEXT:    s_lshr_b32 s3, s6, 16
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s3
+; VI-DENORM-NEXT:    v_fma_f16 v3, s6, 2.0, v0
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s0
+; VI-DENORM-NEXT:    v_add_f16_e64 v2, s6, s6
+; VI-DENORM-NEXT:    s_add_u32 s2, s0, 2
+; VI-DENORM-NEXT:    v_mov_b32_e32 v1, s1
+; VI-DENORM-NEXT:    s_addc_u32 s3, s1, 0
+; VI-DENORM-NEXT:    flat_store_short v[0:1], v2
+; VI-DENORM-NEXT:    s_waitcnt vmcnt(0)
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s2
+; VI-DENORM-NEXT:    v_mov_b32_e32 v1, s3
+; VI-DENORM-NEXT:    flat_store_short v[0:1], v3
+; VI-DENORM-NEXT:    s_waitcnt vmcnt(0)
+; VI-DENORM-NEXT:    s_endpgm
+;
+; VI-FLUSH-LABEL: multiple_use_fadd_fmac_f16:
+; VI-FLUSH:       ; %bb.0:
+; VI-FLUSH-NEXT:    s_load_dword s6, s[4:5], 0x8
+; VI-FLUSH-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-FLUSH-NEXT:    s_lshr_b32 s3, s6, 16
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s0
+; VI-FLUSH-NEXT:    v_add_f16_e64 v2, s6, s6
+; VI-FLUSH-NEXT:    s_add_u32 s2, s0, 2
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v1, s1
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v3, s3
+; VI-FLUSH-NEXT:    s_addc_u32 s3, s1, 0
+; VI-FLUSH-NEXT:    flat_store_short v[0:1], v2
+; VI-FLUSH-NEXT:    s_waitcnt vmcnt(0)
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s2
+; VI-FLUSH-NEXT:    v_mac_f16_e64 v3, s6, 2.0
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v1, s3
+; VI-FLUSH-NEXT:    flat_store_short v[0:1], v3
+; VI-FLUSH-NEXT:    s_waitcnt vmcnt(0)
+; VI-FLUSH-NEXT:    s_endpgm
+;
+; GFX10-DENORM-LABEL: multiple_use_fadd_fmac_f16:
+; GFX10-DENORM:       ; %bb.0:
+; GFX10-DENORM-NEXT:    s_clause 0x1
+; GFX10-DENORM-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-DENORM-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-DENORM-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_lshr_b32 s3, s2, 16
+; GFX10-DENORM-NEXT:    v_add_f16_e64 v1, s2, s2
+; GFX10-DENORM-NEXT:    v_fma_f16 v2, s2, 2.0, s3
+; GFX10-DENORM-NEXT:    global_store_short v0, v1, s[0:1]
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    global_store_short v0, v2, s[0:1] offset:2
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    s_endpgm
+;
+; GFX10-FLUSH-LABEL: multiple_use_fadd_fmac_f16:
+; GFX10-FLUSH:       ; %bb.0:
+; GFX10-FLUSH-NEXT:    s_clause 0x1
+; GFX10-FLUSH-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-FLUSH-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-FLUSH-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-FLUSH-NEXT:    v_add_f16_e64 v0, s2, s2
+; GFX10-FLUSH-NEXT:    s_lshr_b32 s2, s2, 16
+; GFX10-FLUSH-NEXT:    v_add_f16_e32 v2, s2, v0
+; GFX10-FLUSH-NEXT:    global_store_short v1, v0, s[0:1]
+; GFX10-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-FLUSH-NEXT:    global_store_short v1, v2, s[0:1] offset:2
+; GFX10-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-FLUSH-NEXT:    s_endpgm
+;
+; GFX11-DENORM-LABEL: multiple_use_fadd_fmac_f16:
+; GFX11-DENORM:       ; %bb.0:
+; GFX11-DENORM-NEXT:    s_clause 0x1
+; GFX11-DENORM-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-DENORM-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-DENORM-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-DENORM-NEXT:    s_lshr_b32 s3, s2, 16
+; GFX11-DENORM-NEXT:    v_add_f16_e64 v1, s2, s2
+; GFX11-DENORM-NEXT:    v_fma_f16 v2, s2, 2.0, s3
+; GFX11-DENORM-NEXT:    global_store_b16 v0, v1, s[0:1] dlc
+; GFX11-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-DENORM-NEXT:    global_store_b16 v0, v2, s[0:1] offset:2 dlc
+; GFX11-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-DENORM-NEXT:    s_nop 0
+; GFX11-DENORM-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-DENORM-NEXT:    s_endpgm
+;
+; GFX11-FLUSH-LABEL: multiple_use_fadd_fmac_f16:
+; GFX11-FLUSH:       ; %bb.0:
+; GFX11-FLUSH-NEXT:    s_clause 0x1
+; GFX11-FLUSH-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-FLUSH-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-FLUSH-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FLUSH-NEXT:    v_add_f16_e64 v0, s2, s2
+; GFX11-FLUSH-NEXT:    s_lshr_b32 s2, s2, 16
+; GFX11-FLUSH-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-FLUSH-NEXT:    v_add_f16_e32 v2, s2, v0
+; GFX11-FLUSH-NEXT:    global_store_b16 v1, v0, s[0:1] dlc
+; GFX11-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-FLUSH-NEXT:    global_store_b16 v1, v2, s[0:1] offset:2 dlc
+; GFX11-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-FLUSH-NEXT:    s_nop 0
+; GFX11-FLUSH-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-FLUSH-NEXT:    s_endpgm
   %x = bitcast i16 %x.arg to half
   %y = bitcast i16 %y.arg to half
   %out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1
@@ -172,19 +614,117 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}multiple_use_fadd_fmad_f16:
-; GCN-DAG: v_add_f16_e64 [[MUL2:v[0-9]+]], |[[X:s[0-9]+]]|, |s{{[0-9]+}}|
-
-; VI-FLUSH-DAG:     v_mad_f16 [[MAD:v[0-9]+]], |[[X]]|, 2.0, v{{[0-9]+}}
-; VI-DENORM-DAG:    v_fma_f16 [[MAD:v[0-9]+]], |[[X]]|, 2.0, v{{[0-9]+}}
-; GFX10-FLUSH-DAG:  v_add_f16_e32 [[MAD:v[0-9]+]], s{{[0-9]+}}, [[MUL2]]
-; GFX10-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], |[[X]]|, 2.0, s{{[0-9]+}}
-; GFX11-DENORM-DAG: v_fma_f16 [[MAD:v[0-9]+]], |[[X]]|, 2.0, s{{[0-9]+}}
-
-; GCN-DAG: buffer_store_{{short|b16}} [[MUL2]]
-; GCN-DAG: buffer_store_{{short|b16}} [[MAD]]
-; GCN: s_endpgm
 define amdgpu_kernel void @multiple_use_fadd_fmad_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg) #0 {
+; VI-DENORM-LABEL: multiple_use_fadd_fmad_f16:
+; VI-DENORM:       ; %bb.0:
+; VI-DENORM-NEXT:    s_load_dword s6, s[4:5], 0x8
+; VI-DENORM-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-DENORM-NEXT:    s_lshr_b32 s3, s6, 16
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s3
+; VI-DENORM-NEXT:    v_fma_f16 v3, |s6|, 2.0, v0
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s0
+; VI-DENORM-NEXT:    v_add_f16_e64 v2, |s6|, |s6|
+; VI-DENORM-NEXT:    s_add_u32 s2, s0, 2
+; VI-DENORM-NEXT:    v_mov_b32_e32 v1, s1
+; VI-DENORM-NEXT:    s_addc_u32 s3, s1, 0
+; VI-DENORM-NEXT:    flat_store_short v[0:1], v2
+; VI-DENORM-NEXT:    s_waitcnt vmcnt(0)
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s2
+; VI-DENORM-NEXT:    v_mov_b32_e32 v1, s3
+; VI-DENORM-NEXT:    flat_store_short v[0:1], v3
+; VI-DENORM-NEXT:    s_waitcnt vmcnt(0)
+; VI-DENORM-NEXT:    s_endpgm
+;
+; VI-FLUSH-LABEL: multiple_use_fadd_fmad_f16:
+; VI-FLUSH:       ; %bb.0:
+; VI-FLUSH-NEXT:    s_load_dword s6, s[4:5], 0x8
+; VI-FLUSH-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-FLUSH-NEXT:    s_lshr_b32 s3, s6, 16
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s3
+; VI-FLUSH-NEXT:    v_mad_f16 v3, |s6|, 2.0, v0
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s0
+; VI-FLUSH-NEXT:    v_add_f16_e64 v2, |s6|, |s6|
+; VI-FLUSH-NEXT:    s_add_u32 s2, s0, 2
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v1, s1
+; VI-FLUSH-NEXT:    s_addc_u32 s3, s1, 0
+; VI-FLUSH-NEXT:    flat_store_short v[0:1], v2
+; VI-FLUSH-NEXT:    s_waitcnt vmcnt(0)
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s2
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v1, s3
+; VI-FLUSH-NEXT:    flat_store_short v[0:1], v3
+; VI-FLUSH-NEXT:    s_waitcnt vmcnt(0)
+; VI-FLUSH-NEXT:    s_endpgm
+;
+; GFX10-DENORM-LABEL: multiple_use_fadd_fmad_f16:
+; GFX10-DENORM:       ; %bb.0:
+; GFX10-DENORM-NEXT:    s_clause 0x1
+; GFX10-DENORM-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-DENORM-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-DENORM-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_lshr_b32 s3, s2, 16
+; GFX10-DENORM-NEXT:    v_add_f16_e64 v1, |s2|, |s2|
+; GFX10-DENORM-NEXT:    v_fma_f16 v2, |s2|, 2.0, s3
+; GFX10-DENORM-NEXT:    global_store_short v0, v1, s[0:1]
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    global_store_short v0, v2, s[0:1] offset:2
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    s_endpgm
+;
+; GFX10-FLUSH-LABEL: multiple_use_fadd_fmad_f16:
+; GFX10-FLUSH:       ; %bb.0:
+; GFX10-FLUSH-NEXT:    s_clause 0x1
+; GFX10-FLUSH-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-FLUSH-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-FLUSH-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-FLUSH-NEXT:    v_add_f16_e64 v0, |s2|, |s2|
+; GFX10-FLUSH-NEXT:    s_lshr_b32 s2, s2, 16
+; GFX10-FLUSH-NEXT:    v_add_f16_e32 v2, s2, v0
+; GFX10-FLUSH-NEXT:    global_store_short v1, v0, s[0:1]
+; GFX10-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-FLUSH-NEXT:    global_store_short v1, v2, s[0:1] offset:2
+; GFX10-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-FLUSH-NEXT:    s_endpgm
+;
+; GFX11-DENORM-LABEL: multiple_use_fadd_fmad_f16:
+; GFX11-DENORM:       ; %bb.0:
+; GFX11-DENORM-NEXT:    s_clause 0x1
+; GFX11-DENORM-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-DENORM-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-DENORM-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-DENORM-NEXT:    s_lshr_b32 s3, s2, 16
+; GFX11-DENORM-NEXT:    v_add_f16_e64 v1, |s2|, |s2|
+; GFX11-DENORM-NEXT:    v_fma_f16 v2, |s2|, 2.0, s3
+; GFX11-DENORM-NEXT:    global_store_b16 v0, v1, s[0:1] dlc
+; GFX11-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-DENORM-NEXT:    global_store_b16 v0, v2, s[0:1] offset:2 dlc
+; GFX11-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-DENORM-NEXT:    s_nop 0
+; GFX11-DENORM-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-DENORM-NEXT:    s_endpgm
+;
+; GFX11-FLUSH-LABEL: multiple_use_fadd_fmad_f16:
+; GFX11-FLUSH:       ; %bb.0:
+; GFX11-FLUSH-NEXT:    s_clause 0x1
+; GFX11-FLUSH-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-FLUSH-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-FLUSH-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FLUSH-NEXT:    v_add_f16_e64 v0, |s2|, |s2|
+; GFX11-FLUSH-NEXT:    s_lshr_b32 s2, s2, 16
+; GFX11-FLUSH-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-FLUSH-NEXT:    v_add_f16_e32 v2, s2, v0
+; GFX11-FLUSH-NEXT:    global_store_b16 v1, v0, s[0:1] dlc
+; GFX11-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-FLUSH-NEXT:    global_store_b16 v1, v2, s[0:1] offset:2 dlc
+; GFX11-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-FLUSH-NEXT:    s_nop 0
+; GFX11-FLUSH-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-FLUSH-NEXT:    s_endpgm
   %x = bitcast i16 %x.arg to half
   %y = bitcast i16 %y.arg to half
   %out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1
@@ -196,22 +736,127 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}multiple_use_fadd_multi_fmad_f16:
-; VI-FLUSH: v_mad_f16 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, v{{[0-9]+}}
-; VI-FLUSH: v_mad_f16 {{v[0-9]+}}, |[[X]]|, 2.0, v{{[0-9]+}}
-
-; VI-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, v{{[0-9]+}}
-; VI-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X]]|, 2.0, v{{[0-9]+}}
-
-; GFX10-FLUSH:  v_add_f16_e64 [[MUL2:v[0-9]+]], |[[X:s[0-9]+]]|, |{{s[0-9]+}}|
-; GFX10-FLUSH:  v_add_f16_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[MUL2]]
-; GFX10-FLUSH:  v_add_f16_e32 {{v[0-9]+}}, {{s[0-9]+}}, [[MUL2]]
-; GFX10-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, s{{[0-9]+}}
-; GFX11-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X:s[0-9]+]]|, 2.0, s{{[0-9]+}}
-; GFX10-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X]]|, 2.0, s{{[0-9]+}}
-; GFX11-DENORM: v_fma_f16 {{v[0-9]+}}, |[[X]]|, 2.0, s{{[0-9]+}}
-
 define amdgpu_kernel void @multiple_use_fadd_multi_fmad_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg, i16 zeroext %z.arg) #0 {
+; VI-DENORM-LABEL: multiple_use_fadd_multi_fmad_f16:
+; VI-DENORM:       ; %bb.0:
+; VI-DENORM-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; VI-DENORM-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x0
+; VI-DENORM-NEXT:    s_load_dword s6, s[4:5], 0x8
+; VI-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-DENORM-NEXT:    s_lshr_b32 s0, s0, 16
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s0
+; VI-DENORM-NEXT:    v_fma_f16 v2, |s6|, 2.0, v0
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s1
+; VI-DENORM-NEXT:    v_fma_f16 v3, |s6|, 2.0, v0
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s2
+; VI-DENORM-NEXT:    s_add_u32 s4, s2, 2
+; VI-DENORM-NEXT:    v_mov_b32_e32 v1, s3
+; VI-DENORM-NEXT:    s_addc_u32 s5, s3, 0
+; VI-DENORM-NEXT:    flat_store_short v[0:1], v2
+; VI-DENORM-NEXT:    s_waitcnt vmcnt(0)
+; VI-DENORM-NEXT:    v_mov_b32_e32 v0, s4
+; VI-DENORM-NEXT:    v_mov_b32_e32 v1, s5
+; VI-DENORM-NEXT:    flat_store_short v[0:1], v3
+; VI-DENORM-NEXT:    s_waitcnt vmcnt(0)
+; VI-DENORM-NEXT:    s_endpgm
+;
+; VI-FLUSH-LABEL: multiple_use_fadd_multi_fmad_f16:
+; VI-FLUSH:       ; %bb.0:
+; VI-FLUSH-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; VI-FLUSH-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x0
+; VI-FLUSH-NEXT:    s_load_dword s6, s[4:5], 0x8
+; VI-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-FLUSH-NEXT:    s_lshr_b32 s0, s0, 16
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s0
+; VI-FLUSH-NEXT:    v_mad_f16 v2, |s6|, 2.0, v0
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s1
+; VI-FLUSH-NEXT:    v_mad_f16 v3, |s6|, 2.0, v0
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s2
+; VI-FLUSH-NEXT:    s_add_u32 s4, s2, 2
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v1, s3
+; VI-FLUSH-NEXT:    s_addc_u32 s5, s3, 0
+; VI-FLUSH-NEXT:    flat_store_short v[0:1], v2
+; VI-FLUSH-NEXT:    s_waitcnt vmcnt(0)
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v0, s4
+; VI-FLUSH-NEXT:    v_mov_b32_e32 v1, s5
+; VI-FLUSH-NEXT:    flat_store_short v[0:1], v3
+; VI-FLUSH-NEXT:    s_waitcnt vmcnt(0)
+; VI-FLUSH-NEXT:    s_endpgm
+;
+; GFX10-DENORM-LABEL: multiple_use_fadd_multi_fmad_f16:
+; GFX10-DENORM:       ; %bb.0:
+; GFX10-DENORM-NEXT:    s_clause 0x2
+; GFX10-DENORM-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-DENORM-NEXT:    s_load_dword s6, s[4:5], 0x8
+; GFX10-DENORM-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x0
+; GFX10-DENORM-NEXT:    v_mov_b32_e32 v0, 0
+; GFX10-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-DENORM-NEXT:    s_lshr_b32 s0, s0, 16
+; GFX10-DENORM-NEXT:    v_fma_f16 v2, |s6|, 2.0, s1
+; GFX10-DENORM-NEXT:    v_fma_f16 v1, |s6|, 2.0, s0
+; GFX10-DENORM-NEXT:    global_store_short v0, v1, s[2:3]
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    global_store_short v0, v2, s[2:3] offset:2
+; GFX10-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-DENORM-NEXT:    s_endpgm
+;
+; GFX10-FLUSH-LABEL: multiple_use_fadd_multi_fmad_f16:
+; GFX10-FLUSH:       ; %bb.0:
+; GFX10-FLUSH-NEXT:    s_clause 0x2
+; GFX10-FLUSH-NEXT:    s_load_dword s6, s[4:5], 0x8
+; GFX10-FLUSH-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX10-FLUSH-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x0
+; GFX10-FLUSH-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-FLUSH-NEXT:    v_add_f16_e64 v0, |s6|, |s6|
+; GFX10-FLUSH-NEXT:    s_lshr_b32 s0, s0, 16
+; GFX10-FLUSH-NEXT:    v_add_f16_e32 v2, s0, v0
+; GFX10-FLUSH-NEXT:    v_add_f16_e32 v0, s1, v0
+; GFX10-FLUSH-NEXT:    global_store_short v1, v2, s[2:3]
+; GFX10-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-FLUSH-NEXT:    global_store_short v1, v0, s[2:3] offset:2
+; GFX10-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-FLUSH-NEXT:    s_endpgm
+;
+; GFX11-DENORM-LABEL: multiple_use_fadd_multi_fmad_f16:
+; GFX11-DENORM:       ; %bb.0:
+; GFX11-DENORM-NEXT:    s_clause 0x2
+; GFX11-DENORM-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
+; GFX11-DENORM-NEXT:    s_load_b32 s4, s[0:1], 0x8
+; GFX11-DENORM-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-DENORM-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-DENORM-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-DENORM-NEXT:    s_lshr_b32 s2, s2, 16
+; GFX11-DENORM-NEXT:    v_fma_f16 v2, |s4|, 2.0, s3
+; GFX11-DENORM-NEXT:    v_fma_f16 v1, |s4|, 2.0, s2
+; GFX11-DENORM-NEXT:    global_store_b16 v0, v1, s[0:1] dlc
+; GFX11-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-DENORM-NEXT:    global_store_b16 v0, v2, s[0:1] offset:2 dlc
+; GFX11-DENORM-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-DENORM-NEXT:    s_nop 0
+; GFX11-DENORM-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-DENORM-NEXT:    s_endpgm
+;
+; GFX11-FLUSH-LABEL: multiple_use_fadd_multi_fmad_f16:
+; GFX11-FLUSH:       ; %bb.0:
+; GFX11-FLUSH-NEXT:    s_clause 0x2
+; GFX11-FLUSH-NEXT:    s_load_b32 s4, s[0:1], 0x8
+; GFX11-FLUSH-NEXT:    s_load_b64 s[2:3], s[0:1], 0x8
+; GFX11-FLUSH-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-FLUSH-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-FLUSH-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FLUSH-NEXT:    v_add_f16_e64 v0, |s4|, |s4|
+; GFX11-FLUSH-NEXT:    s_lshr_b32 s2, s2, 16
+; GFX11-FLUSH-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX11-FLUSH-NEXT:    v_add_f16_e32 v2, s2, v0
+; GFX11-FLUSH-NEXT:    v_add_f16_e32 v0, s3, v0
+; GFX11-FLUSH-NEXT:    global_store_b16 v1, v2, s[0:1] dlc
+; GFX11-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-FLUSH-NEXT:    global_store_b16 v1, v0, s[0:1] offset:2 dlc
+; GFX11-FLUSH-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-FLUSH-NEXT:    s_nop 0
+; GFX11-FLUSH-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-FLUSH-NEXT:    s_endpgm
   %x = bitcast i16 %x.arg to half
   %y = bitcast i16 %y.arg to half
   %z = bitcast i16 %z.arg to half
@@ -225,11 +870,48 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}fmul_x2_xn2_f16:
-; GCN: v_mul_f16_e64 [[TMP0:v[0-9]+]], [[X:s[0-9]+]], -4.0
-; GCN: v_mul_f16_e32 [[RESULT:v[0-9]+]], [[X]], [[TMP0]]
-; GCN: buffer_store_{{short|b16}} [[RESULT]]
 define amdgpu_kernel void @fmul_x2_xn2_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg) #0 {
+; VI-LABEL: fmul_x2_xn2_f16:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_load_dword s2, s[4:5], 0x8
+; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    v_mul_f16_e64 v0, s2, -4.0
+; VI-NEXT:    v_mul_f16_e32 v2, s2, v0
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    flat_store_short v[0:1], v2
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    s_endpgm
+;
+; GFX10-LABEL: fmul_x2_xn2_f16:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mul_f16_e64 v0, s2, -4.0
+; GFX10-NEXT:    v_mul_f16_e32 v0, s2, v0
+; GFX10-NEXT:    global_store_short v1, v0, s[0:1]
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fmul_x2_xn2_f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_mul_f16_e64 v0, s2, -4.0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_mul_f16_e32 v0, s2, v0
+; GFX11-NEXT:    global_store_b16 v1, v0, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_nop 0
+; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT:    s_endpgm
   %x = bitcast i16 %x.arg to half
   %y = bitcast i16 %y.arg to half
   %out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1
@@ -240,13 +922,49 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}fmul_x2_xn3_f16:
-; SIVI:  v_mov_b32_e32 [[K:v[0-9]+]], 0xc600
-; SIVI:  v_mul_f16_e32 [[TMP0:v[0-9]+]], [[X:s[0-9]+]], [[K]]
-; GFX10: v_mul_f16_e64 [[TMP0:v[0-9]+]], 0xc600, [[X:s[0-9]+]]
-; GCN:   v_mul_f16_e32 [[RESULT:v[0-9]+]], [[X]], [[TMP0]]
-; GCN:   buffer_store_{{short|b16}} [[RESULT]]
 define amdgpu_kernel void @fmul_x2_xn3_f16(ptr addrspace(1) %out, i16 zeroext %x.arg, i16 zeroext %y.arg) #0 {
+; VI-LABEL: fmul_x2_xn3_f16:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_load_dword s2, s[4:5], 0x8
+; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; VI-NEXT:    v_mov_b32_e32 v0, 0xc600
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    v_mul_f16_e32 v0, s2, v0
+; VI-NEXT:    v_mul_f16_e32 v2, s2, v0
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    flat_store_short v[0:1], v2
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    s_endpgm
+;
+; GFX10-LABEL: fmul_x2_xn3_f16:
+; GFX10:       ; %bb.0:
+; GFX10-NEXT:    s_clause 0x1
+; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x8
+; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mul_f16_e64 v0, 0xc600, s2
+; GFX10-NEXT:    v_mul_f16_e32 v0, s2, v0
+; GFX10-NEXT:    global_store_short v1, v0, s[0:1]
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX10-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fmul_x2_xn3_f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[0:1], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_mul_f16_e64 v0, 0xc600, s2
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_mul_f16_e32 v0, s2, v0
+; GFX11-NEXT:    global_store_b16 v1, v0, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_nop 0
+; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT:    s_endpgm
   %x = bitcast i16 %x.arg to half
   %y = bitcast i16 %y.arg to half
   %out.gep.1 = getelementptr half, ptr addrspace(1) %out, i32 1