[llvm] 5fe3f06 - AMDGPU/GlobalISel: Add new baseline checks for canonicalize
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Thu Apr 23 12:04:43 PDT 2020
Author: Matt Arsenault
Date: 2020-04-23T15:04:32-04:00
New Revision: 5fe3f06596927df988b7aa8797918c49e179f2fc
URL: https://github.com/llvm/llvm-project/commit/5fe3f06596927df988b7aa8797918c49e179f2fc
DIFF: https://github.com/llvm/llvm-project/commit/5fe3f06596927df988b7aa8797918c49e179f2fc.diff
LOG: AMDGPU/GlobalISel: Add new baseline checks for canonicalize
Added:
Modified:
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
index 186afde9a607..b62e24921ebf 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
@@ -1,4 +1,5 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -disable-gisel-legality-check -o - %s | FileCheck -check-prefix=GFX8 %s
# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
---
@@ -14,6 +15,10 @@ machineFunctionInfo:
body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_f16_denorm
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_MAX_F16_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MAX_F16_e64_]]
; GFX9-LABEL: name: fcanonicalize_f16_denorm
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_MAX_F16_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
@@ -37,6 +42,10 @@ machineFunctionInfo:
body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_f16_flush
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_MUL_F16_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F16_e64 0, 15360, 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MUL_F16_e64_]]
; GFX9-LABEL: name: fcanonicalize_f16_flush
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_MUL_F16_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F16_e64 0, 15360, 0, [[COPY]], 0, 0, implicit $exec
@@ -61,6 +70,10 @@ body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_f32_denorm
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
; GFX9-LABEL: name: fcanonicalize_f32_denorm
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
@@ -84,6 +97,10 @@ body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_f32_flush
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
; GFX9-LABEL: name: fcanonicalize_f32_flush
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $exec
@@ -107,6 +124,10 @@ body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_v2f16_denorm
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+ ; GFX8: [[FCANONICALIZE:%[0-9]+]]:vgpr(<2 x s16>) = G_FCANONICALIZE [[COPY]]
+ ; GFX8: S_ENDPGM 0, implicit [[FCANONICALIZE]](<2 x s16>)
; GFX9-LABEL: name: fcanonicalize_v2f16_denorm
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_PK_MAX_F16_:%[0-9]+]]:vgpr_32 = V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $exec
@@ -130,6 +151,10 @@ body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_v2f16_flush
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_PK_MUL_F16_:%[0-9]+]]:vgpr_32 = V_PK_MUL_F16 0, 15360, 8, [[COPY]], 0, 0, 0, 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_PK_MUL_F16_]]
; GFX9-LABEL: name: fcanonicalize_v2f16_flush
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_PK_MUL_F16_:%[0-9]+]]:vgpr_32 = V_PK_MUL_F16 0, 15360, 8, [[COPY]], 0, 0, 0, 0, 0, implicit $exec
@@ -153,6 +178,10 @@ body: |
bb.0:
liveins: $vgpr0_vgpr1
+ ; GFX8-LABEL: name: fcanonicalize_f64_denorm
+ ; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
+ ; GFX8: [[V_MAX_F64_:%[0-9]+]]:vreg_64 = V_MAX_F64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MAX_F64_]]
; GFX9-LABEL: name: fcanonicalize_f64_denorm
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[V_MAX_F64_:%[0-9]+]]:vreg_64 = V_MAX_F64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
@@ -176,6 +205,10 @@ body: |
bb.0:
liveins: $vgpr0_vgpr1
+ ; GFX8-LABEL: name: fcanonicalize_f64_flush
+ ; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
+ ; GFX8: [[V_MUL_F64_:%[0-9]+]]:vreg_64 = V_MUL_F64 0, 4607182418800017408, 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MUL_F64_]]
; GFX9-LABEL: name: fcanonicalize_f64_flush
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[V_MUL_F64_:%[0-9]+]]:vreg_64 = V_MUL_F64 0, 4607182418800017408, 0, [[COPY]], 0, 0, implicit $exec
@@ -198,6 +231,10 @@ machineFunctionInfo:
body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_fabs_f32_denorm
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
; GFX9-LABEL: name: fcanonicalize_fabs_f32_denorm
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $exec
@@ -222,6 +259,10 @@ machineFunctionInfo:
body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_fabs_f32_flush
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 1065353216, 2, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
; GFX9-LABEL: name: fcanonicalize_fabs_f32_flush
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 1065353216, 2, [[COPY]], 0, 0, implicit $exec
@@ -245,6 +286,10 @@ machineFunctionInfo:
body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_fneg_f32_denorm
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
; GFX9-LABEL: name: fcanonicalize_fneg_f32_denorm
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $exec
@@ -268,6 +313,10 @@ machineFunctionInfo:
body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_fneg_f32_flush
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
; GFX9-LABEL: name: fcanonicalize_fneg_f32_flush
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $exec
@@ -291,6 +340,12 @@ machineFunctionInfo:
body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_fneg_fabs_f32_denorm
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
+ ; GFX8: [[V_XOR_B32_e32_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e32 [[S_MOV_B32_]], [[COPY]], implicit $exec
+ ; GFX8: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F32_e64 2, [[V_XOR_B32_e32_]], 2, [[V_XOR_B32_e32_]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
; GFX9-LABEL: name: fcanonicalize_fneg_fabs_f32_denorm
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
@@ -317,6 +372,12 @@ machineFunctionInfo:
body: |
bb.0:
liveins: $vgpr0
+ ; GFX8-LABEL: name: fcanonicalize_fneg_fabs_f32_flush
+ ; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX8: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
+ ; GFX8: [[V_XOR_B32_e32_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e32 [[S_MOV_B32_]], [[COPY]], implicit $exec
+ ; GFX8: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 1065353216, 2, [[V_XOR_B32_e32_]], 0, 0, implicit $exec
+ ; GFX8: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
; GFX9-LABEL: name: fcanonicalize_fneg_fabs_f32_flush
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 2147483648
More information about the llvm-commits
mailing list