diff --git a/llvm/test/CodeGen/AMDGPU/clamp.ll b/llvm/test/CodeGen/AMDGPU/clamp.ll
--- a/llvm/test/CodeGen/AMDGPU/clamp.ll
+++ b/llvm/test/CodeGen/AMDGPU/clamp.ll
@@ -1,11 +1,64 @@
-; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,SI,GFX678 %s
-; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX89,GFX678 %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX89,GFX9 %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX6 %s
+; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX8 %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
+; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11 %s
 
-; GCN-LABEL: {{^}}v_clamp_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -17,10 +70,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_neg_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, -[[A]], -[[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_neg_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_neg_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, -v2, -v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_neg_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, -v3, -v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_neg_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, -v1, -v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_neg_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, -v1, -v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -33,10 +137,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_negabs_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, -|[[A]]|, -|[[A]]| clamp{{$}}
 define amdgpu_kernel void @v_clamp_negabs_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_negabs_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, -|v2|, -|v2| clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_negabs_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, -|v3|, -|v3| clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_negabs_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, -|v1|, -|v1| clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_negabs_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, -|v1|, -|v1| clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -51,12 +206,69 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_negzero_f32:
-; GCN-DAG: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN-DAG: v_add_f32_e32 [[ADD:v[0-9]+]], 0.5, [[A]]
-; GCN: v_max_f32_e32 [[MAX:v[0-9]+]], 0x80000000, [[ADD]]
-; GCN: v_min_f32_e32 v{{[0-9]+}}, 1.0, [[MAX]]
 define amdgpu_kernel void @v_clamp_negzero_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_negzero_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_add_f32_e32 v2, 0.5, v2
+; GFX6-NEXT:    v_max_f32_e32 v2, 0x80000000, v2
+; GFX6-NEXT:    v_min_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_negzero_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_add_f32_e32 v2, 0.5, v3
+; GFX8-NEXT:    v_max_f32_e32 v2, 0x80000000, v2
+; GFX8-NEXT:    v_min_f32_e32 v2, 1.0, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_negzero_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_add_f32_e32 v1, 0.5, v1
+; GFX9-NEXT:    v_max_f32_e32 v1, 0x80000000, v1
+; GFX9-NEXT:    v_min_f32_e32 v1, 1.0, v1
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_negzero_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_add_f32_e32 v1, 0.5, v1
+; GFX11-NEXT:    v_max_f32_e32 v1, 0x80000000, v1
+; GFX11-NEXT:    v_min_f32_e32 v1, 1.0, v1
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -71,14 +283,69 @@
 
 ; FIXME: Weird inconsistency in how -0.0 is treated. Accepted if clamp
 ; matched through med3, not if directly. Is this correct?
-
-; GCN-LABEL: {{^}}v_clamp_negzero_maybe_snan_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX678: v_mul_f32_e32 [[QUIET:v[0-9]+]], 1.0, [[A]]
-; GFX9: v_max_f32_e32 [[QUIET:v[0-9]+]], [[A]], [[A]]
-; GCN: v_max_f32_e32 [[MAX:v[0-9]+]], 0x80000000, [[QUIET]]
-; GCN: v_min_f32_e32 [[MIN:v[0-9]+]], 1.0, [[MAX]]
 define amdgpu_kernel void @v_clamp_negzero_maybe_snan_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_negzero_maybe_snan_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    v_max_f32_e32 v2, 0x80000000, v2
+; GFX6-NEXT:    v_min_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_negzero_maybe_snan_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mul_f32_e32 v2, 1.0, v3
+; GFX8-NEXT:    v_max_f32_e32 v2, 0x80000000, v2
+; GFX8-NEXT:    v_min_f32_e32 v2, 1.0, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_negzero_maybe_snan_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e32 v1, v1, v1
+; GFX9-NEXT:    v_max_f32_e32 v1, 0x80000000, v1
+; GFX9-NEXT:    v_min_f32_e32 v1, 1.0, v1
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_negzero_maybe_snan_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e32 v1, v1, v1
+; GFX11-NEXT:    v_max_f32_e32 v1, 0x80000000, v1
+; GFX11-NEXT:    v_min_f32_e32 v1, 1.0, v1
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -90,21 +357,78 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_multi_use_max_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX678: v_mul_f32_e32 [[QUIET_A:v[0-9]+]], 1.0, [[A]]
-; GFX9: v_max_f32_e32 [[QUIET_A:v[0-9]+]], [[A]], [[A]]
-; GCN: v_max_f32_e32 [[MAX:v[0-9]+]], 0, [[QUIET_A]]
-; GCN: v_min_f32_e32 [[MED:v[0-9]+]], 1.0, [[QUIET_A]]
-; GCN-NOT: [[MAX]]
-; GCN-NOT: [[MED]]
-
-; SI: buffer_store_dword [[MED]]
-; SI: buffer_store_dword [[MAX]]
-
-; GFX89: {{flat|global}}_store_dword v{{.+}}, [[MED]]
-; GFX89: {{flat|global}}_store_dword v{{.+}}, [[MAX]]
 define amdgpu_kernel void @v_clamp_multi_use_max_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_multi_use_max_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_mov_b32 s6, -1
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    v_max_f32_e32 v2, 0, v2
+; GFX6-NEXT:    v_min_f32_e32 v3, 1.0, v2
+; GFX6-NEXT:    buffer_store_dword v3, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    buffer_store_dword v2, off, s[4:7], 0
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_multi_use_max_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mul_f32_e32 v2, 1.0, v3
+; GFX8-NEXT:    v_max_f32_e32 v2, 0, v2
+; GFX8-NEXT:    v_min_f32_e32 v3, 1.0, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v3
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_multi_use_max_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e32 v1, v1, v1
+; GFX9-NEXT:    v_max_f32_e32 v1, 0, v1
+; GFX9-NEXT:    v_min_f32_e32 v2, 1.0, v1
+; GFX9-NEXT:    global_store_dword v0, v2, s[0:1]
+; GFX9-NEXT:    global_store_dword v[0:1], v1, off
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_multi_use_max_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e32 v1, v1, v1
+; GFX11-NEXT:    v_max_f32_e32 v1, 0, v1
+; GFX11-NEXT:    v_min_f32_e32 v2, 1.0, v1
+; GFX11-NEXT:    global_store_b32 v0, v2, s[0:1]
+; GFX11-NEXT:    global_store_b32 v[0:1], v1, off dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -117,13 +441,62 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_f16:
-; GCN: {{buffer|flat|global}}_load_ushort [[A:v[0-9]+]]
-; GFX89: v_max_f16_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
-
-; SI: v_cvt_f32_f16_e64 [[CVT:v[0-9]+]], [[A]] clamp{{$}}
-; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, [[CVT]]
 define amdgpu_kernel void @v_clamp_f16(half addrspace(1)* %out, half addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_f16:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_ushort v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, v2 clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    buffer_store_short v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 1, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_ushort v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_ushort v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f16_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_short v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_u16 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f16_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr half, half addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr half, half addrspace(1)* %out, i32 %tid
@@ -135,14 +508,63 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_neg_f16:
-; GCN: {{buffer|flat|global}}_load_ushort [[A:v[0-9]+]]
-; GFX89: v_max_f16_e64 v{{[0-9]+}}, -[[A]], -[[A]] clamp{{$}}
-
 ; FIXME: Better to fold neg into max
-; SI: v_cvt_f32_f16_e64 [[CVT:v[0-9]+]], -[[A]] clamp{{$}}
-; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, [[CVT]]
 define amdgpu_kernel void @v_clamp_neg_f16(half addrspace(1)* %out, half addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_neg_f16:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_ushort v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, -v2 clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    buffer_store_short v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_neg_f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 1, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_ushort v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_e64 v2, -v3, -v3 clamp
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_neg_f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_ushort v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f16_e64 v1, -v1, -v1 clamp
+; GFX9-NEXT:    global_store_short v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_neg_f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_u16 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f16_e64 v1, -v1, -v1 clamp
+; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr half, half addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr half, half addrspace(1)* %out, i32 %tid
@@ -155,15 +577,63 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_negabs_f16:
-; GCN: {{buffer|flat|global}}_load_ushort [[A:v[0-9]+]]
-; GFX89: v_max_f16_e64 v{{[0-9]+}}, -|[[A]]|, -|[[A]]| clamp{{$}}
-
 ; FIXME: Better to fold neg/abs into max
-
-; SI: v_cvt_f32_f16_e64 [[CVT:v[0-9]+]], -|[[A]]| clamp{{$}}
-; SI: v_cvt_f16_f32_e32 v{{[0-9]+}}, [[CVT]]
 define amdgpu_kernel void @v_clamp_negabs_f16(half addrspace(1)* %out, half addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_negabs_f16:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_ushort v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, -|v2| clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    buffer_store_short v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_negabs_f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 1, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_ushort v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_e64 v2, -|v3|, -|v3| clamp
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_negabs_f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_ushort v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f16_e64 v1, -|v1|, -|v1| clamp
+; GFX9-NEXT:    global_store_short v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_negabs_f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_u16 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f16_e64 v1, -|v1|, -|v1| clamp
+; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr half, half addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr half, half addrspace(1)* %out, i32 %tid
@@ -178,10 +648,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_f64:
-; GCN: {{buffer|flat|global}}_load_dwordx2 [[A:v\[[0-9]+:[0-9]+\]]]
-; GCN: v_max_f64 v{{\[[0-9]+:[0-9]+\]}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_f64(double addrspace(1)* %out, double addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_f64:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f64 v[2:3], v[2:3], v[2:3] clamp
+; GFX6-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_f64:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f64 v[0:1], v[0:1], v[0:1] clamp
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_f64:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; 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:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f64 v[0:1], v[0:1], v[0:1] clamp
+; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_f64:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f64 v[0:1], v[0:1], v[0:1] clamp
+; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr double, double addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr double, double addrspace(1)* %out, i32 %tid
@@ -193,10 +714,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_neg_f64:
-; GCN: {{buffer|flat|global}}_load_dwordx2 [[A:v\[[0-9]+:[0-9]+\]]]
-; GCN: v_max_f64 v{{\[[0-9]+:[0-9]+\]}}, -[[A]], -[[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_neg_f64(double addrspace(1)* %out, double addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_neg_f64:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f64 v[2:3], -v[2:3], -v[2:3] clamp
+; GFX6-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_neg_f64:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f64 v[0:1], -v[0:1], -v[0:1] clamp
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_neg_f64:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; 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:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f64 v[0:1], -v[0:1], -v[0:1] clamp
+; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_neg_f64:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f64 v[0:1], -v[0:1], -v[0:1] clamp
+; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr double, double addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr double, double addrspace(1)* %out, i32 %tid
@@ -209,10 +781,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_negabs_f64:
-; GCN: {{buffer|flat|global}}_load_dwordx2 [[A:v\[[0-9]+:[0-9]+\]]]
-; GCN: v_max_f64 v{{\[[0-9]+:[0-9]+\]}}, -|[[A]]|, -|[[A]]| clamp{{$}}
 define amdgpu_kernel void @v_clamp_negabs_f64(double addrspace(1)* %out, double addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_negabs_f64:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f64 v[2:3], -|v[2:3]|, -|v[2:3]| clamp
+; GFX6-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_negabs_f64:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f64 v[0:1], -|v[0:1]|, -|v[0:1]| clamp
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_negabs_f64:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; 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:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f64 v[0:1], -|v[0:1]|, -|v[0:1]| clamp
+; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_negabs_f64:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f64 v[0:1], -|v[0:1]|, -|v[0:1]| clamp
+; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr double, double addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr double, double addrspace(1)* %out, i32 %tid
@@ -227,10 +850,64 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_aby_negzero_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_med3_f32
 define amdgpu_kernel void @v_clamp_med3_aby_negzero_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_med3_aby_negzero_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_brev_b32 s4, 1
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_med3_f32 v2, s4, 1.0, v2
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_aby_negzero_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    s_brev_b32 s0, 1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_med3_f32 v2, s0, 1.0, v3
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_aby_negzero_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_brev_b32 s2, 1
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_med3_f32 v1, s2, 1.0, v1
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_aby_negzero_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_med3_f32 v1, 0x80000000, 1.0, v1
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -240,10 +917,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_aby_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_med3_aby_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_med3_aby_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_aby_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_aby_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_aby_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -253,10 +981,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_bay_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_med3_bay_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_med3_bay_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_bay_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_bay_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_bay_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -266,10 +1045,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_yab_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_med3_yab_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_med3_yab_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_yab_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_yab_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_yab_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -279,10 +1109,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_yba_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_med3_yba_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_med3_yba_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_yba_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_yba_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_yba_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -292,10 +1173,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_ayb_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_med3_ayb_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_med3_ayb_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_ayb_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_ayb_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_ayb_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -305,10 +1237,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_bya_f32:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_med3_bya_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_med3_bya_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_bya_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_bya_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_bya_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -318,9 +1301,48 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_constants_to_one_f32:
-; GCN: v_mov_b32_e32 v{{[0-9]+}}, 1.0
 define amdgpu_kernel void @v_clamp_constants_to_one_f32(float addrspace(1)* %out) #0 {
+; GFX6-LABEL: v_clamp_constants_to_one_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    v_mov_b32_e32 v2, 1.0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_constants_to_one_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 1.0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_constants_to_one_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 1.0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_constants_to_one_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 1.0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
   %med = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float 4.0)
@@ -328,9 +1350,47 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_constants_to_zero_f32:
-; GCN: v_mov_b32_e32 v{{[0-9]+}}, 0{{$}}
 define amdgpu_kernel void @v_clamp_constants_to_zero_f32(float addrspace(1)* %out) #0 {
+; GFX6-LABEL: v_clamp_constants_to_zero_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v1, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_constants_to_zero_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_constants_to_zero_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_constants_to_zero_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
   %med = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float -4.0)
@@ -338,9 +1398,48 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_constant_preserve_f32:
-; GCN: v_mov_b32_e32 v{{[0-9]+}}, 0.5
 define amdgpu_kernel void @v_clamp_constant_preserve_f32(float addrspace(1)* %out) #0 {
+; GFX6-LABEL: v_clamp_constant_preserve_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    v_mov_b32_e32 v2, 0.5
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_constant_preserve_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0.5
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_constant_preserve_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0.5
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_constant_preserve_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0.5
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
   %med = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float 0.5)
@@ -348,9 +1447,48 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_constant_preserve_denorm_f32:
-; GCN: v_mov_b32_e32 v{{[0-9]+}}, 0x7fffff{{$}}
 define amdgpu_kernel void @v_clamp_constant_preserve_denorm_f32(float addrspace(1)* %out) #0 {
+; GFX6-LABEL: v_clamp_constant_preserve_denorm_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    v_mov_b32_e32 v2, 0x7fffff
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_constant_preserve_denorm_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0x7fffff
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_constant_preserve_denorm_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7fffff
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_constant_preserve_denorm_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0x7fffff
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
   %med = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float bitcast (i32 8388607 to float))
@@ -358,9 +1496,47 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_constant_qnan_f32:
-; GCN: v_mov_b32_e32 v{{[0-9]+}}, 0{{$}}
 define amdgpu_kernel void @v_clamp_constant_qnan_f32(float addrspace(1)* %out) #0 {
+; GFX6-LABEL: v_clamp_constant_qnan_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v1, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_constant_qnan_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_constant_qnan_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_constant_qnan_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
   %med = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float 0x7FF8000000000000)
@@ -368,9 +1544,47 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_constant_snan_f32:
-; GCN: v_mov_b32_e32 v{{[0-9]+}}, 0{{$}}
 define amdgpu_kernel void @v_clamp_constant_snan_f32(float addrspace(1)* %out) #0 {
+; GFX6-LABEL: v_clamp_constant_snan_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v1, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_constant_snan_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_constant_snan_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_constant_snan_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
   %med = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float bitcast (i32 2139095041 to float))
@@ -382,11 +1596,65 @@
 ; Test non-default behaviors enabling snans and disabling dx10_clamp
 ; ---------------------------------------------------------------------
 
-; GCN-LABEL: {{^}}v_clamp_f32_no_dx10_clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_add_f32_e32 [[ADD:v[0-9]+]], 0.5, [[A]]
-; GCN: v_med3_f32 v{{[0-9]+}}, [[ADD]], 0, 1.0
 define amdgpu_kernel void @v_clamp_f32_no_dx10_clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #2 {
+; GFX6-LABEL: v_clamp_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_add_f32_e32 v2, 0.5, v2
+; GFX6-NEXT:    v_med3_f32 v2, v2, 0, 1.0
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_add_f32_e32 v2, 0.5, v3
+; GFX8-NEXT:    v_med3_f32 v2, v2, 0, 1.0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_add_f32_e32 v1, 0.5, v1
+; GFX9-NEXT:    v_med3_f32 v1, v1, 0, 1.0
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_add_f32_e32 v1, 0.5, v1
+; GFX11-NEXT:    v_med3_f32 v1, v1, 0, 1.0
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -399,10 +1667,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_f32_snan_dx10clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_add_f32_e64 [[ADD:v[0-9]+]], [[A]], 0.5 clamp{{$}}
 define amdgpu_kernel void @v_clamp_f32_snan_dx10clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #3 {
+; GFX6-LABEL: v_clamp_f32_snan_dx10clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_add_f32_e64 v2, v2, 0.5 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_f32_snan_dx10clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_add_f32_e64 v2, v3, 0.5 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_f32_snan_dx10clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_add_f32_e64 v1, v1, 0.5 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_f32_snan_dx10clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_add_f32_e64 v1, v1, 0.5 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -415,12 +1734,65 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_f32_snan_no_dx10clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX678: v_mul_f32_e32 [[QUIET_A:v[0-9]+]], 1.0, [[A]]
-; GFX9: v_max_f32_e32 [[QUIET_A:v[0-9]+]], [[A]], [[A]]
-; GCN: v_med3_f32 {{v[0-9]+}}, [[QUIET_A]], 0, 1.0
 define amdgpu_kernel void @v_clamp_f32_snan_no_dx10clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #4 {
+; GFX6-LABEL: v_clamp_f32_snan_no_dx10clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    v_med3_f32 v2, v2, 0, 1.0
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_f32_snan_no_dx10clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_mul_f32_e32 v2, 1.0, v3
+; GFX8-NEXT:    v_med3_f32 v2, v2, 0, 1.0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_f32_snan_no_dx10clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e32 v1, v1, v1
+; GFX9-NEXT:    v_med3_f32 v1, v1, 0, 1.0
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_f32_snan_no_dx10clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e32 v1, v1, v1
+; GFX11-NEXT:    v_med3_f32 v1, v1, 0, 1.0
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -432,11 +1804,65 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_f32_snan_no_dx10clamp_nnan_src:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_add_f32_e32 [[ADD:v[0-9]+]], 1.0, [[A]]
-; GCN: v_med3_f32 v{{[0-9]+}}, [[ADD]], 0, 1.0
 define amdgpu_kernel void @v_clamp_f32_snan_no_dx10clamp_nnan_src(float addrspace(1)* %out, float addrspace(1)* %aptr) #4 {
+; GFX6-LABEL: v_clamp_f32_snan_no_dx10clamp_nnan_src:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_add_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    v_med3_f32 v2, v2, 0, 1.0
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_f32_snan_no_dx10clamp_nnan_src:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_add_f32_e32 v2, 1.0, v3
+; GFX8-NEXT:    v_med3_f32 v2, v2, 0, 1.0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_f32_snan_no_dx10clamp_nnan_src:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_add_f32_e32 v1, 1.0, v1
+; GFX9-NEXT:    v_med3_f32 v1, v1, 0, 1.0
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_f32_snan_no_dx10clamp_nnan_src:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_add_f32_e32 v1, 1.0, v1
+; GFX11-NEXT:    v_med3_f32 v1, v1, 0, 1.0
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -449,10 +1875,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_aby_f32_no_dx10_clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_med3_aby_f32_no_dx10_clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #2 {
+; GFX6-LABEL: v_clamp_med3_aby_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_aby_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_aby_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_aby_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -462,10 +1939,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_bay_f32_no_dx10_clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_med3_bay_f32_no_dx10_clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #2 {
+; GFX6-LABEL: v_clamp_med3_bay_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_max_f32_e64 v2, v2, v2 clamp
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_bay_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f32_e64 v2, v3, v3 clamp
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_bay_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_bay_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_max_f32_e64 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -475,10 +2003,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_yab_f32_no_dx10_clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_med3_f32 v{{[0-9]+}}, [[A]], 0, 1.0
 define amdgpu_kernel void @v_clamp_med3_yab_f32_no_dx10_clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #2 {
+; GFX6-LABEL: v_clamp_med3_yab_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_med3_f32 v2, v2, 0, 1.0
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_yab_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_med3_f32 v2, v3, 0, 1.0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_yab_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_med3_f32 v1, v1, 0, 1.0
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_yab_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_med3_f32 v1, v1, 0, 1.0
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -488,10 +2067,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_yba_f32_no_dx10_clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_med3_f32 v{{[0-9]+}}, [[A]], 1.0, 0
 define amdgpu_kernel void @v_clamp_med3_yba_f32_no_dx10_clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #2 {
+; GFX6-LABEL: v_clamp_med3_yba_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_med3_f32 v2, v2, 1.0, 0
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_yba_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_med3_f32 v2, v3, 1.0, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_yba_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_med3_f32 v1, v1, 1.0, 0
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_yba_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_med3_f32 v1, v1, 1.0, 0
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -501,10 +2131,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_ayb_f32_no_dx10_clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_med3_f32 v{{[0-9]+}}, 0, [[A]], 1.0
 define amdgpu_kernel void @v_clamp_med3_ayb_f32_no_dx10_clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #2 {
+; GFX6-LABEL: v_clamp_med3_ayb_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_med3_f32 v2, 0, v2, 1.0
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_ayb_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_med3_f32 v2, 0, v3, 1.0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_ayb_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_med3_f32 v1, 0, v1, 1.0
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_ayb_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_med3_f32 v1, 0, v1, 1.0
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -514,10 +2195,61 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_med3_bya_f32_no_dx10_clamp:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GCN: v_med3_f32 v{{[0-9]+}}, 1.0, [[A]], 0
 define amdgpu_kernel void @v_clamp_med3_bya_f32_no_dx10_clamp(float addrspace(1)* %out, float addrspace(1)* %aptr) #2 {
+; GFX6-LABEL: v_clamp_med3_bya_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_med3_f32 v2, 1.0, v2, 0
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_med3_bya_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_med3_f32 v2, 1.0, v3, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_med3_bya_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_med3_f32 v1, 1.0, v1, 0
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_med3_bya_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_med3_f32 v1, 1.0, v1, 0
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
@@ -527,9 +2259,48 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_constant_qnan_f32_no_dx10_clamp:
-; GCN: v_mov_b32_e32 v{{[0-9]+}}, 0x7fc00000
 define amdgpu_kernel void @v_clamp_constant_qnan_f32_no_dx10_clamp(float addrspace(1)* %out) #2 {
+; GFX6-LABEL: v_clamp_constant_qnan_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    v_mov_b32_e32 v2, 0x7fc00000
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_constant_qnan_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0x7fc00000
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_constant_qnan_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7fc00000
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_constant_qnan_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0x7fc00000
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
   %med = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float 0x7FF8000000000000)
@@ -537,9 +2308,48 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_constant_snan_f32_no_dx10_clamp:
-; GCN: v_mov_b32_e32 v{{[0-9]+}}, 0x7f800001
 define amdgpu_kernel void @v_clamp_constant_snan_f32_no_dx10_clamp(float addrspace(1)* %out) #2 {
+; GFX6-LABEL: v_clamp_constant_snan_f32_no_dx10_clamp:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    v_mov_b32_e32 v2, 0x7f800001
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_constant_snan_f32_no_dx10_clamp:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0x7f800001
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_constant_snan_f32_no_dx10_clamp:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x7f800001
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_constant_snan_f32_no_dx10_clamp:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0x7f800001
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
   %med = call float @llvm.amdgcn.fmed3.f32(float 0.0, float 1.0, float bitcast (i32 2139095041 to float))
@@ -547,11 +2357,69 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_v2f16:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9-NOT: [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_v2f16:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v3, v3 clamp
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, v2 clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_v2f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, v3, v3 clamp dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_max_f16_e64 v3, v3, v3 clamp
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_v2f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_v2f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -563,11 +2431,80 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_v2f16_undef_elt:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9-NOT: [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_v2f16_undef_elt(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_v2f16_undef_elt:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b32 s2, 0x7fc00000
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_cvt_f32_f16_e32 v3, v2
+; GFX6-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e32 v2, v2
+; GFX6-NEXT:    v_mul_f32_e32 v3, 1.0, v3
+; GFX6-NEXT:    v_max_f32_e32 v3, 0x7fc00000, v3
+; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    v_med3_f32 v2, v2, 0, s2
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_min_f32_e32 v3, 1.0, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_v2f16_undef_elt:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0x7e00
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, v3, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_max_f16_e32 v3, v3, v3
+; GFX8-NEXT:    v_max_f16_e32 v2, 0, v2
+; GFX8-NEXT:    v_max_f16_e32 v3, 0x7e00, v3
+; GFX8-NEXT:    v_min_f16_e32 v3, 1.0, v3
+; GFX8-NEXT:    v_min_f16_sdwa v2, v2, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_v2f16_undef_elt:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_v2f16_undef_elt:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -579,10 +2516,78 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_v2f16_not_zero:
-; GFX9: v_pk_max_f16
-; GFX9: v_pk_min_f16
 define amdgpu_kernel void @v_clamp_v2f16_not_zero(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_v2f16_not_zero:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_cvt_f32_f16_e32 v3, v2
+; GFX6-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, v2 clamp
+; GFX6-NEXT:    v_mul_f32_e32 v3, 1.0, v3
+; GFX6-NEXT:    v_max_f32_e32 v3, 2.0, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_min_f32_e32 v3, 1.0, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_v2f16_not_zero:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_e32 v2, v3, v3
+; GFX8-NEXT:    v_max_f16_e32 v2, 2.0, v2
+; GFX8-NEXT:    v_max_f16_sdwa v3, v3, v3 clamp dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_min_f16_e32 v2, 1.0, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_v2f16_not_zero:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, 2.0
+; GFX9-NEXT:    v_pk_min_f16 v1, v1, 1.0 op_sel_hi:[1,0]
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_v2f16_not_zero:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, 2.0
+; GFX11-NEXT:    v_pk_min_f16 v1, v1, 1.0 op_sel_hi:[1,0]
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -594,10 +2599,77 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_v2f16_not_one:
-; GFX9: v_pk_max_f16
-; GFX9: v_pk_min_f16
 define amdgpu_kernel void @v_clamp_v2f16_not_one(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_v2f16_not_one:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e32 v2, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v3, v3 clamp
+; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_med3_f32 v2, v2, 0, 0
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_v2f16_not_one:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_e32 v2, v3, v3
+; GFX8-NEXT:    v_max_f16_e32 v2, 0, v2
+; GFX8-NEXT:    v_max_f16_sdwa v3, v3, v3 clamp dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_min_f16_e32 v2, 0, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_v2f16_not_one:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, 0
+; GFX9-NEXT:    v_pk_min_f16 v1, v1, 1.0 op_sel:[0,1] op_sel_hi:[1,0]
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_v2f16_not_one:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, 0
+; GFX11-NEXT:    v_pk_min_f16 v1, v1, 1.0 op_sel:[0,1] op_sel_hi:[1,0]
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -609,11 +2681,70 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_neg_v2f16:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9-NOT: [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[A]], [[A]] neg_lo:[1,1] neg_hi:[1,1] clamp{{$}}
 define amdgpu_kernel void @v_clamp_neg_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_neg_v2f16:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_xor_b32_e32 v2, 0x80008000, v2
+; GFX6-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v3, v3 clamp
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, v2 clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_neg_v2f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, -v3, -v3 clamp dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_max_f16_e64 v3, -v3, -v3 clamp
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_neg_v2f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 neg_lo:[1,1] neg_hi:[1,1] clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_neg_v2f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 neg_lo:[1,1] neg_hi:[1,1] clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -626,11 +2757,72 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_negabs_v2f16:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9: v_and_b32_e32 [[ABS:v[0-9]+]], 0x7fff7fff, [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[ABS]], [[ABS]] neg_lo:[1,1] neg_hi:[1,1] clamp{{$}}
 define amdgpu_kernel void @v_clamp_negabs_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_negabs_v2f16:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_or_b32_e32 v2, 0x80008000, v2
+; GFX6-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v3, v3 clamp
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, v2 clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_negabs_v2f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, -|v3|, -|v3| clamp dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_max_f16_e64 v3, -|v3|, -|v3| clamp
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_negabs_v2f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_b32_e32 v1, 0x7fff7fff, v1
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 neg_lo:[1,1] neg_hi:[1,1] clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_negabs_v2f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_and_b32_e32 v1, 0x7fff7fff, v1
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 neg_lo:[1,1] neg_hi:[1,1] clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -645,11 +2837,70 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_neglo_v2f16:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9-NOT: [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[A]], [[A]] neg_lo:[1,1] clamp{{$}}
 define amdgpu_kernel void @v_clamp_neglo_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_neglo_v2f16:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v3, v3 clamp
+; GFX6-NEXT:    v_and_b32_e32 v2, 0xffff, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, -v2 clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_neglo_v2f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, v3, v3 clamp dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_max_f16_e64 v3, -v3, -v3 clamp
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_neglo_v2f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 neg_lo:[1,1] clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_neglo_v2f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 neg_lo:[1,1] clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -664,11 +2915,69 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_neghi_v2f16:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9-NOT: [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[A]], [[A]] neg_hi:[1,1] clamp{{$}}
 define amdgpu_kernel void @v_clamp_neghi_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_neghi_v2f16:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v3, -v3 clamp
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, v2 clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_neghi_v2f16:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, -v3, -v3 clamp dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_max_f16_e64 v3, v3, v3 clamp
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_neghi_v2f16:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 neg_hi:[1,1] clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_neghi_v2f16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 neg_hi:[1,1] clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -683,11 +2992,69 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_v2f16_shuffle:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9-NOT: [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[A]], [[A]] op_sel:[1,1] op_sel_hi:[0,0] clamp{{$}}
 define amdgpu_kernel void @v_clamp_v2f16_shuffle(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_v2f16_shuffle:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v2, v2 clamp
+; GFX6-NEXT:    v_cvt_f32_f16_e64 v3, v3 clamp
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_v2f16_shuffle:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, v3, v3 clamp dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX8-NEXT:    v_max_f16_sdwa v3, v3, v3 clamp dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_v2f16_shuffle:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 op_sel:[1,1] op_sel_hi:[0,0] clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_v2f16_shuffle:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 op_sel:[1,1] op_sel_hi:[0,0] clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -700,11 +3067,79 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_v2f16_undef_limit_elts0:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9-NOT: [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_v2f16_undef_limit_elts0(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_v2f16_undef_limit_elts0:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b32 s2, 0x7fc00000
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e32 v3, v3
+; GFX6-NEXT:    v_cvt_f32_f16_e32 v2, v2
+; GFX6-NEXT:    v_mul_f32_e32 v3, 1.0, v3
+; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    v_med3_f32 v3, v3, s2, 1.0
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    v_med3_f32 v2, v2, 0, s2
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_v2f16_undef_limit_elts0:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0x3c00
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, v3, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_max_f16_e32 v3, v3, v3
+; GFX8-NEXT:    v_max_f16_e32 v2, 0x7e00, v2
+; GFX8-NEXT:    v_max_f16_e32 v3, 0, v3
+; GFX8-NEXT:    v_min_f16_e32 v3, 0x7e00, v3
+; GFX8-NEXT:    v_min_f16_sdwa v2, v2, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_v2f16_undef_limit_elts0:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_v2f16_undef_limit_elts0:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -716,11 +3151,80 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_v2f16_undef_limit_elts1:
-; GCN: {{buffer|flat|global}}_load_dword [[A:v[0-9]+]]
-; GFX9-NOT: [[A]]
-; GFX9: v_pk_max_f16 [[CLAMP:v[0-9]+]], [[A]], [[A]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_v2f16_undef_limit_elts1(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %aptr) #0 {
+; GFX6-LABEL: v_clamp_v2f16_undef_limit_elts1:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s6, 0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v1, 0
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b64 s[4:5], s[2:3]
+; GFX6-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
+; GFX6-NEXT:    s_mov_b32 s2, 0x7fc00000
+; GFX6-NEXT:    s_waitcnt vmcnt(0)
+; GFX6-NEXT:    v_cvt_f32_f16_e32 v3, v2
+; GFX6-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_cvt_f32_f16_e32 v2, v2
+; GFX6-NEXT:    v_mul_f32_e32 v3, 1.0, v3
+; GFX6-NEXT:    v_max_f32_e32 v3, 0x7fc00000, v3
+; GFX6-NEXT:    v_mul_f32_e32 v2, 1.0, v2
+; GFX6-NEXT:    v_med3_f32 v2, v2, 0, s2
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v2, v2
+; GFX6-NEXT:    v_min_f32_e32 v3, 1.0, v3
+; GFX6-NEXT:    v_cvt_f16_f32_e32 v3, v3
+; GFX6-NEXT:    s_mov_b64 s[2:3], s[6:7]
+; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX6-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_v2f16_undef_limit_elts1:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0x7e00
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dword v3, v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    s_waitcnt vmcnt(0)
+; GFX8-NEXT:    v_max_f16_sdwa v2, v3, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_max_f16_e32 v3, v3, v3
+; GFX8-NEXT:    v_max_f16_e32 v2, 0, v2
+; GFX8-NEXT:    v_max_f16_e32 v3, 0x7e00, v3
+; GFX8-NEXT:    v_min_f16_e32 v3, 1.0, v3
+; GFX8-NEXT:    v_min_f16_sdwa v2, v2, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX9-LABEL: v_clamp_v2f16_undef_limit_elts1:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v0, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_pk_max_f16 v1, v1, v1 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_v2f16_undef_limit_elts1:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_pk_max_f16 v1, v1, v1 clamp
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %aptr, i32 %tid
   %out.gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -732,11 +3236,73 @@
   ret void
 }
 
-; GCN-LABEL: {{^}}v_clamp_diff_source_f32:
-; GCN: v_add_f32_e32 [[A:v[0-9]+]]
-; GCN: v_add_f32_e32 [[B:v[0-9]+]]
-; GCN: v_max_f32_e64 v{{[0-9]+}}, [[A]], [[B]] clamp{{$}}
 define amdgpu_kernel void @v_clamp_diff_source_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #0
+; GFX6-LABEL: v_clamp_diff_source_f32:
+; GFX6:       ; %bb.0:
+; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x0
+; GFX6-NEXT:    s_load_dword s2, s[2:3], 0x2
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    v_mov_b32_e32 v0, s5
+; GFX6-NEXT:    v_mov_b32_e32 v1, s2
+; GFX6-NEXT:    v_add_f32_e32 v0, s4, v0
+; GFX6-NEXT:    v_add_f32_e32 v1, s4, v1
+; GFX6-NEXT:    v_max_f32_e64 v0, v0, v1 clamp
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0 offset:12
+; GFX6-NEXT:    s_endpgm
+;
+; GFX8-LABEL: v_clamp_diff_source_f32:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x0
+; GFX8-NEXT:    s_load_dword s2, s[2:3], 0x8
+; GFX8-NEXT:    s_add_u32 s0, s0, 12
+; GFX8-NEXT:    s_addc_u32 s1, s1, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v0, s5
+; GFX8-NEXT:    v_mov_b32_e32 v1, s2
+; GFX8-NEXT:    v_add_f32_e32 v0, s4, v0
+; GFX8-NEXT:    v_add_f32_e32 v1, s4, v1
+; GFX8-NEXT:    v_max_f32_e64 v2, v0, v1 clamp
+; 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
+;
+; GFX9-LABEL: v_clamp_diff_source_f32:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x0
+; GFX9-NEXT:    s_load_dword s6, s[2:3], 0x8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NEXT:    v_mov_b32_e32 v2, s6
+; GFX9-NEXT:    v_add_f32_e32 v1, s4, v1
+; GFX9-NEXT:    v_add_f32_e32 v2, s4, v2
+; GFX9-NEXT:    v_max_f32_e64 v1, v1, v2 clamp
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1] offset:12
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: v_clamp_diff_source_f32:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
+; GFX11-NEXT:    v_mov_b32_e32 v2, 0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b64 s[4:5], s[2:3], 0x0
+; GFX11-NEXT:    s_load_b32 s2, s[2:3], 0x8
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_add_f32_e64 v0, s4, s5
+; GFX11-NEXT:    v_add_f32_e64 v1, s4, s2
+; GFX11-NEXT:    v_max_f32_e64 v0, v0, v1 clamp
+; GFX11-NEXT:    global_store_b32 v2, v0, s[0:1] offset:12
+; GFX11-NEXT:    s_endpgm
 {
   %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 0
   %gep1 = getelementptr float, float addrspace(1)* %aptr, i32 1