DAG: Fix PromoteFloatResult for fcanonicalize

llvm-svn: 338382

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index 8b9bb57..9aa0ea1 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -1898,7 +1898,8 @@ void DAGTypeLegalizer::PromoteFloatResult(SDNode *N, unsigned ResNo) {
     case ISD::FROUND:
     case ISD::FSIN:
     case ISD::FSQRT:
-    case ISD::FTRUNC:     R = PromoteFloatRes_UnaryOp(N); break;
+    case ISD::FTRUNC:
+    case ISD::FCANONICALIZE: R = PromoteFloatRes_UnaryOp(N); break;
 
     // Binary FP Operations
     case ISD::FADD:

diff --git a/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll b/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
index f251e30..52d8919 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
@@ -1,5 +1,6 @@
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9 %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI,GFX89 %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9,GFX89 %s
+; RUN: llc -march=amdgcn -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI %s
 
 declare half @llvm.fabs.f16(half) #0
 declare half @llvm.canonicalize.f16(half) #0
@@ -9,18 +10,21 @@ declare i32 @llvm.amdgcn.workitem.id.x() #0
 
 
 ; GCN-LABEL: {{^}}v_test_canonicalize_var_f16:
-; GCN: v_max_f16_e32 [[REG:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_max_f16_e32 [[REG:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+
+; CI: v_cvt_f32_f16_e32
+; CI: v_mul_f32_e32 {{v[0-9]+}}, 1.0, {{v[0-9]+}}
 define amdgpu_kernel void @v_test_canonicalize_var_f16(half addrspace(1)* %out) #1 {
   %val = load half, half addrspace(1)* %out
   %canonicalized = call half @llvm.canonicalize.f16(half %val)
-  store half %canonicalized, half addrspace(1)* %out
+  store half %canonicalized, half addrspace(1)* undef
   ret void
 }
 
 ; GCN-LABEL: {{^}}s_test_canonicalize_var_f16:
-; GCN: v_max_f16_e64 [[REG:v[0-9]+]], {{s[0-9]+}}, {{s[0-9]+}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_max_f16_e64 [[REG:v[0-9]+]], {{s[0-9]+}}, {{s[0-9]+}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @s_test_canonicalize_var_f16(half addrspace(1)* %out, i16 zeroext %val.arg) #1 {
   %val = bitcast i16 %val.arg to half
   %canonicalized = call half @llvm.canonicalize.f16(half %val)
@@ -29,8 +33,8 @@ define amdgpu_kernel void @s_test_canonicalize_var_f16(half addrspace(1)* %out,
 }
 
 ; GCN-LABEL: {{^}}v_test_canonicalize_fabs_var_f16:
-; GCN: v_max_f16_e64 [[REG:v[0-9]+]], |{{v[0-9]+}}|, |{{v[0-9]+}}|
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_max_f16_e64 [[REG:v[0-9]+]], |{{v[0-9]+}}|, |{{v[0-9]+}}|
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @v_test_canonicalize_fabs_var_f16(half addrspace(1)* %out) #1 {
   %val = load half, half addrspace(1)* %out
   %val.fabs = call half @llvm.fabs.f16(half %val)
@@ -40,8 +44,11 @@ define amdgpu_kernel void @v_test_canonicalize_fabs_var_f16(half addrspace(1)* %
 }
 
 ; GCN-LABEL: {{^}}v_test_canonicalize_fneg_fabs_var_f16:
-; GCN: v_max_f16_e64 [[REG:v[0-9]+]], -|{{v[0-9]+}}|, -|{{v[0-9]+}}|
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_max_f16_e64 [[REG:v[0-9]+]], -|{{v[0-9]+}}|, -|{{v[0-9]+}}|
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+
+; CI: v_cvt_f32_f16_e64 v{{[0-9]+}}, -|v{{[0-9]+}}|
+; CI: v_mul_f32_e32 {{v[0-9]+}}, 1.0, {{v[0-9]+}}
 define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_f16(half addrspace(1)* %out) #1 {
   %val = load half, half addrspace(1)* %out
   %val.fabs = call half @llvm.fabs.f16(half %val)
@@ -52,8 +59,11 @@ define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_f16(half addrspace(
 }
 
 ; GCN-LABEL: {{^}}v_test_canonicalize_fneg_var_f16:
-; GCN: v_max_f16_e64 [[REG:v[0-9]+]], -{{v[0-9]+}}, -{{v[0-9]+}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_max_f16_e64 [[REG:v[0-9]+]], -{{v[0-9]+}}, -{{v[0-9]+}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+
+; CI: v_cvt_f32_f16_e64 {{v[0-9]+}}, -{{v[0-9]+}}
+; CI: v_mul_f32_e32 {{v[0-9]+}}, 1.0, {{v[0-9]+}}
 define amdgpu_kernel void @v_test_canonicalize_fneg_var_f16(half addrspace(1)* %out) #1 {
   %val = load half, half addrspace(1)* %out
   %val.fneg = fsub half -0.0, %val
@@ -63,8 +73,8 @@ define amdgpu_kernel void @v_test_canonicalize_fneg_var_f16(half addrspace(1)* %
 }
 
 ; GCN-LABEL: {{^}}v_test_no_denormals_canonicalize_fneg_var_f16:
-; GCN: v_mul_f16_e32 [[REG:v[0-9]+]], -1.0, v{{[0-9]+}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mul_f16_e32 [[REG:v[0-9]+]], -1.0, v{{[0-9]+}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @v_test_no_denormals_canonicalize_fneg_var_f16(half addrspace(1)* %out) #2 {
   %val = load half, half addrspace(1)* %out
   %val.fneg = fsub half -0.0, %val
@@ -74,8 +84,11 @@ define amdgpu_kernel void @v_test_no_denormals_canonicalize_fneg_var_f16(half ad
 }
 
 ; GCN-LABEL: {{^}}v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
-; GCN: v_mul_f16_e64 [[REG:v[0-9]+]], -1.0, |v{{[0-9]+}}|
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mul_f16_e64 [[REG:v[0-9]+]], -1.0, |v{{[0-9]+}}|
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+
+; CI: v_cvt_f32_f16_e64 {{v[0-9]+}}, -|{{v[0-9]+}}|
+; CI: v_mul_f32_e32 {{v[0-9]+}}, 1.0, {{v[0-9]+}}
 define amdgpu_kernel void @v_test_no_denormals_canonicalize_fneg_fabs_var_f16(half addrspace(1)* %out) #2 {
   %val = load half, half addrspace(1)* %out
   %val.fabs = call half @llvm.fabs.f16(half %val)
@@ -86,8 +99,8 @@ define amdgpu_kernel void @v_test_no_denormals_canonicalize_fneg_fabs_var_f16(ha
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_p0_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_p0_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0.0)
   store half %canonicalized, half addrspace(1)* %out
@@ -95,8 +108,8 @@ define amdgpu_kernel void @test_fold_canonicalize_p0_f16(half addrspace(1)* %out
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_n0_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0xffff8000{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0xffff8000{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_n0_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half -0.0)
   store half %canonicalized, half addrspace(1)* %out
@@ -104,8 +117,8 @@ define amdgpu_kernel void @test_fold_canonicalize_n0_f16(half addrspace(1)* %out
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_p1_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3c00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3c00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_p1_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 1.0)
   store half %canonicalized, half addrspace(1)* %out
@@ -113,8 +126,8 @@ define amdgpu_kernel void @test_fold_canonicalize_p1_f16(half addrspace(1)* %out
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_n1_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0xffffbc00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0xffffbc00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_n1_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half -1.0)
   store half %canonicalized, half addrspace(1)* %out
@@ -122,8 +135,8 @@ define amdgpu_kernel void @test_fold_canonicalize_n1_f16(half addrspace(1)* %out
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_literal_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x4c00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x4c00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_literal_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 16.0)
   store half %canonicalized, half addrspace(1)* %out
@@ -131,8 +144,8 @@ define amdgpu_kernel void @test_fold_canonicalize_literal_f16(half addrspace(1)*
 }
 
 ; GCN-LABEL: {{^}}test_default_denormals_fold_canonicalize_denormal0_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3ff{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3ff{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_default_denormals_fold_canonicalize_denormal0_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xH03FF)
   store half %canonicalized, half addrspace(1)* %out
@@ -140,8 +153,8 @@ define amdgpu_kernel void @test_default_denormals_fold_canonicalize_denormal0_f1
 }
 
 ; GCN-LABEL: {{^}}test_denormals_fold_canonicalize_denormal0_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3ff{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3ff{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal0_f16(half addrspace(1)* %out) #3 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xH03FF)
   store half %canonicalized, half addrspace(1)* %out
@@ -149,8 +162,8 @@ define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal0_f16(half a
 }
 
 ; GCN-LABEL: {{^}}test_default_denormals_fold_canonicalize_denormal1_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0xffff83ff{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0xffff83ff{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_default_denormals_fold_canonicalize_denormal1_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xH83FF)
   store half %canonicalized, half addrspace(1)* %out
@@ -158,8 +171,8 @@ define amdgpu_kernel void @test_default_denormals_fold_canonicalize_denormal1_f1
 }
 
 ; GCN-LABEL: {{^}}test_denormals_fold_canonicalize_denormal1_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0xffff83ff{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0xffff83ff{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal1_f16(half addrspace(1)* %out) #3 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xH83FF)
   store half %canonicalized, half addrspace(1)* %out
@@ -167,8 +180,8 @@ define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal1_f16(half a
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_qnan_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7c00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7c00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_qnan_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xH7C00)
   store half %canonicalized, half addrspace(1)* %out
@@ -176,8 +189,8 @@ define amdgpu_kernel void @test_fold_canonicalize_qnan_f16(half addrspace(1)* %o
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_qnan_value_neg1_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg1_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half bitcast (i16 -1 to half))
   store half %canonicalized, half addrspace(1)* %out
@@ -185,8 +198,8 @@ define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg1_f16(half addrs
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_qnan_value_neg2_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg2_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half bitcast (i16 -2 to half))
   store half %canonicalized, half addrspace(1)* %out
@@ -194,8 +207,8 @@ define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg2_f16(half addrs
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_snan0_value_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_snan0_value_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xH7C01)
   store half %canonicalized, half addrspace(1)* %out
@@ -203,8 +216,8 @@ define amdgpu_kernel void @test_fold_canonicalize_snan0_value_f16(half addrspace
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_snan1_value_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_snan1_value_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xH7DFF)
   store half %canonicalized, half addrspace(1)* %out
@@ -212,8 +225,8 @@ define amdgpu_kernel void @test_fold_canonicalize_snan1_value_f16(half addrspace
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_snan2_value_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_snan2_value_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xHFDFF)
   store half %canonicalized, half addrspace(1)* %out
@@ -221,8 +234,8 @@ define amdgpu_kernel void @test_fold_canonicalize_snan2_value_f16(half addrspace
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_snan3_value_f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
-; GCN: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e00{{$}}
+; GFX89: {{flat|global}}_store_short v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_snan3_value_f16(half addrspace(1)* %out) #1 {
   %canonicalized = call half @llvm.canonicalize.f16(half 0xHFC01)
   store half %canonicalized, half addrspace(1)* %out
@@ -253,7 +266,7 @@ define amdgpu_kernel void @v_test_canonicalize_var_v2f16(<2 x half> addrspace(1)
 
 ; GFX9: v_and_b32_e32 [[ABS:v[0-9]+]], 0x7fff7fff, v{{[0-9]+}}
 ; GFX9: v_pk_max_f16 [[REG:v[0-9]+]], [[ABS]], [[ABS]]{{$}}
-; GCN: {{flat|global}}_store_dword
+; GFX89: {{flat|global}}_store_dword
 define amdgpu_kernel void @v_test_canonicalize_fabs_var_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -271,7 +284,12 @@ define amdgpu_kernel void @v_test_canonicalize_fabs_var_v2f16(<2 x half> addrspa
 
 ; GFX9: v_and_b32_e32 [[ABS:v[0-9]+]], 0x7fff7fff, v{{[0-9]+}}
 ; GFX9: v_pk_max_f16 [[REG:v[0-9]+]], [[ABS]], [[ABS]] neg_lo:[1,1] neg_hi:[1,1]{{$}}
-; GCN: {{flat|global}}_store_dword
+; GFX89: {{flat|global}}_store_dword
+
+; CI: v_cvt_f32_f16
+; CI: v_cvt_f32_f16
+; CI: v_mul_f32_e32 v{{[0-9]+}}, 1.0
+; CI: v_mul_f32_e32 v{{[0-9]+}}, 1.0
 define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
@@ -315,8 +333,8 @@ define amdgpu_kernel void @s_test_canonicalize_var_v2f16(<2 x half> addrspace(1)
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_p0_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_p0_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> zeroinitializer)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -324,8 +342,8 @@ define amdgpu_kernel void @test_fold_canonicalize_p0_v2f16(<2 x half> addrspace(
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_n0_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x80008000{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x80008000{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_n0_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half -0.0, half -0.0>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -333,8 +351,8 @@ define amdgpu_kernel void @test_fold_canonicalize_n0_v2f16(<2 x half> addrspace(
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_p1_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3c003c00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3c003c00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_p1_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 1.0, half 1.0>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -342,8 +360,8 @@ define amdgpu_kernel void @test_fold_canonicalize_p1_v2f16(<2 x half> addrspace(
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_n1_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0xbc00bc00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0xbc00bc00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_n1_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half -1.0, half -1.0>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -351,8 +369,8 @@ define amdgpu_kernel void @test_fold_canonicalize_n1_v2f16(<2 x half> addrspace(
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_literal_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x4c004c00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x4c004c00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_literal_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 16.0, half 16.0>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -360,8 +378,8 @@ define amdgpu_kernel void @test_fold_canonicalize_literal_v2f16(<2 x half> addrs
 }
 
 ; GCN-LABEL: {{^}}test_no_denormals_fold_canonicalize_denormal0_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3ff03ff{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3ff03ff{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_no_denormals_fold_canonicalize_denormal0_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH03FF, half 0xH03FF>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -369,8 +387,8 @@ define amdgpu_kernel void @test_no_denormals_fold_canonicalize_denormal0_v2f16(<
 }
 
 ; GCN-LABEL: {{^}}test_denormals_fold_canonicalize_denormal0_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3ff03ff{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x3ff03ff{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal0_v2f16(<2 x half> addrspace(1)* %out) #3 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH03FF, half 0xH03FF>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -378,8 +396,8 @@ define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal0_v2f16(<2 x
 }
 
 ; GCN-LABEL: {{^}}test_no_denormals_fold_canonicalize_denormal1_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x83ff83ff{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x83ff83ff{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_no_denormals_fold_canonicalize_denormal1_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH83FF, half 0xH83FF>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -387,8 +405,8 @@ define amdgpu_kernel void @test_no_denormals_fold_canonicalize_denormal1_v2f16(<
 }
 
 ; GCN-LABEL: {{^}}test_denormals_fold_canonicalize_denormal1_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x83ff83ff{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x83ff83ff{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal1_v2f16(<2 x half> addrspace(1)* %out) #3 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH83FF, half 0xH83FF>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -396,8 +414,8 @@ define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal1_v2f16(<2 x
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_qnan_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7c007c00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7c007c00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_qnan_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7C00, half 0xH7C00>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -405,8 +423,8 @@ define amdgpu_kernel void @test_fold_canonicalize_qnan_v2f16(<2 x half> addrspac
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_qnan_value_neg1_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg1_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> bitcast (i32 -1 to <2 x half>))
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -414,8 +432,8 @@ define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg1_v2f16(<2 x hal
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_qnan_value_neg2_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg2_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half bitcast (i16 -2 to half), half bitcast (i16 -2 to half)>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -423,8 +441,8 @@ define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg2_v2f16(<2 x hal
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_snan0_value_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_snan0_value_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7C01, half 0xH7C01>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -432,8 +450,8 @@ define amdgpu_kernel void @test_fold_canonicalize_snan0_value_v2f16(<2 x half> a
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_snan1_value_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_snan1_value_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7DFF, half 0xH7DFF>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -441,8 +459,8 @@ define amdgpu_kernel void @test_fold_canonicalize_snan1_value_v2f16(<2 x half> a
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_snan2_value_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_snan2_value_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xHFDFF, half 0xHFDFF>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out
@@ -450,8 +468,8 @@ define amdgpu_kernel void @test_fold_canonicalize_snan2_value_v2f16(<2 x half> a
 }
 
 ; GCN-LABEL: {{^}}test_fold_canonicalize_snan3_value_v2f16:
-; GCN: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
-; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
+; GFX89: v_mov_b32_e32 [[REG:v[0-9]+]], 0x7e007e00{{$}}
+; GFX89: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[REG]]
 define amdgpu_kernel void @test_fold_canonicalize_snan3_value_v2f16(<2 x half> addrspace(1)* %out) #1 {
   %canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xHFC01, half 0xHFC01>)
   store <2 x half> %canonicalized, <2 x half> addrspace(1)* %out