You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
393 lines
13 KiB
393 lines
13 KiB
# 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
|
|
|
|
---
|
|
|
|
name: fcanonicalize_f16_denorm
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp64-fp16-input-denormals: true
|
|
fp64-fp16-output-denormals: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
; GFX8-LABEL: name: fcanonicalize_f16_denorm
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %2:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %2
|
|
; GFX9-LABEL: name: fcanonicalize_f16_denorm
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %2:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %2
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s16) = G_TRUNC %0
|
|
%2:vgpr(s16) = G_FCANONICALIZE %1
|
|
S_ENDPGM 0, implicit %2
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_f16_flush
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp64-fp16-input-denormals: false
|
|
fp64-fp16-output-denormals: false
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
; GFX8-LABEL: name: fcanonicalize_f16_flush
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %2:vgpr_32 = nofpexcept V_MUL_F16_e64 0, 15360, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %2
|
|
; GFX9-LABEL: name: fcanonicalize_f16_flush
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %2:vgpr_32 = nofpexcept V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %2
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s16) = G_TRUNC %0
|
|
%2:vgpr(s16) = G_FCANONICALIZE %1
|
|
S_ENDPGM 0, implicit %2
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_f32_denorm
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp32-input-denormals: true
|
|
fp32-output-denormals: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
|
|
; GFX8-LABEL: name: fcanonicalize_f32_denorm
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %1:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %1
|
|
; GFX9-LABEL: name: fcanonicalize_f32_denorm
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %1:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %1
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s32) = G_FCANONICALIZE %0
|
|
S_ENDPGM 0, implicit %1
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_f32_flush
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp32-input-denormals: false
|
|
fp32-output-denormals: false
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
|
|
; GFX8-LABEL: name: fcanonicalize_f32_flush
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %1:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %1
|
|
; GFX9-LABEL: name: fcanonicalize_f32_flush
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %1:vgpr_32 = nofpexcept V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %1
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s32) = G_FCANONICALIZE %0
|
|
S_ENDPGM 0, implicit %1
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_v2f16_denorm
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp64-fp16-input-denormals: true
|
|
fp64-fp16-output-denormals: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
|
|
; GFX8-LABEL: name: fcanonicalize_v2f16_denorm
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %1
|
|
; GFX9-LABEL: name: fcanonicalize_v2f16_denorm
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %1
|
|
%0:vgpr(<2 x s16>) = COPY $vgpr0
|
|
%1:vgpr(<2 x s16>) = G_FCANONICALIZE %0
|
|
S_ENDPGM 0, implicit %1
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_v2f16_flush
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp64-fp16-input-denormals: false
|
|
fp64-fp16-output-denormals: false
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
|
|
; GFX8-LABEL: name: fcanonicalize_v2f16_flush
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %1:vgpr_32 = nofpexcept V_PK_MUL_F16 0, 15360, 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %1
|
|
; GFX9-LABEL: name: fcanonicalize_v2f16_flush
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %1:vgpr_32 = nofpexcept V_PK_MAX_F16 8, [[COPY]], 8, [[COPY]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %1
|
|
%0:vgpr(<2 x s16>) = COPY $vgpr0
|
|
%1:vgpr(<2 x s16>) = G_FCANONICALIZE %0
|
|
S_ENDPGM 0, implicit %1
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_f64_denorm
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp64-fp16-input-denormals: true
|
|
fp64-fp16-output-denormals: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1
|
|
|
|
; GFX8-LABEL: name: fcanonicalize_f64_denorm
|
|
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX8: %1:vreg_64 = nofpexcept V_MAX_F64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %1
|
|
; GFX9-LABEL: name: fcanonicalize_f64_denorm
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: %1:vreg_64 = nofpexcept V_MAX_F64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %1
|
|
%0:vgpr(s64) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s64) = G_FCANONICALIZE %0
|
|
S_ENDPGM 0, implicit %1
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_f64_flush
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp64-fp16-input-denormals: false
|
|
fp64-fp16-output-denormals: false
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1
|
|
|
|
; GFX8-LABEL: name: fcanonicalize_f64_flush
|
|
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX8: %1:vreg_64 = nofpexcept V_MUL_F64 0, 4607182418800017408, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %1
|
|
; GFX9-LABEL: name: fcanonicalize_f64_flush
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: %1:vreg_64 = nofpexcept V_MAX_F64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %1
|
|
%0:vgpr(s64) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s64) = G_FCANONICALIZE %0
|
|
S_ENDPGM 0, implicit %1
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_fabs_f32_denorm
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp32-input-denormals: true
|
|
fp32-output-denormals: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
; GFX8-LABEL: name: fcanonicalize_fabs_f32_denorm
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %2:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %2
|
|
; GFX9-LABEL: name: fcanonicalize_fabs_f32_denorm
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %2
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s32) = G_FABS %0
|
|
%2:vgpr(s32) = G_FCANONICALIZE %1
|
|
S_ENDPGM 0, implicit %2
|
|
...
|
|
|
|
|
|
---
|
|
|
|
name: fcanonicalize_fabs_f32_flush
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp32-input-denormals: false
|
|
fp32-output-denormals: false
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
; GFX8-LABEL: name: fcanonicalize_fabs_f32_flush
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %2:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %2
|
|
; GFX9-LABEL: name: fcanonicalize_fabs_f32_flush
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[COPY]], 2, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %2
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s32) = G_FABS %0
|
|
%2:vgpr(s32) = G_FCANONICALIZE %1
|
|
S_ENDPGM 0, implicit %2
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_fneg_f32_denorm
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp32-input-denormals: true
|
|
fp32-output-denormals: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
; GFX8-LABEL: name: fcanonicalize_fneg_f32_denorm
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %2:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %2
|
|
; GFX9-LABEL: name: fcanonicalize_fneg_f32_denorm
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %2
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s32) = G_FNEG %0
|
|
%2:vgpr(s32) = G_FCANONICALIZE %1
|
|
S_ENDPGM 0, implicit %2
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_fneg_f32_flush
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp32-input-denormals: false
|
|
fp32-output-denormals: false
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0
|
|
; GFX8-LABEL: name: fcanonicalize_fneg_f32_flush
|
|
; GFX8: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX8: %2:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %2
|
|
; GFX9-LABEL: name: fcanonicalize_fneg_f32_flush
|
|
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: %2:vgpr_32 = nofpexcept V_MAX_F32_e64 1, [[COPY]], 1, [[COPY]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %2
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s32) = G_FNEG %0
|
|
%2:vgpr(s32) = G_FCANONICALIZE %1
|
|
S_ENDPGM 0, implicit %2
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_fneg_fabs_f32_denorm
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp32-input-denormals: true
|
|
fp32-output-denormals: true
|
|
|
|
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: %3:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[V_XOR_B32_e32_]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %3
|
|
; 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
|
|
; GFX9: [[V_XOR_B32_e32_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e32 [[S_MOV_B32_]], [[COPY]], implicit $exec
|
|
; GFX9: %3:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e32_]], 2, [[V_XOR_B32_e32_]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %3
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s32) = G_FNEG %0
|
|
%2:vgpr(s32) = G_FABS %1
|
|
%3:vgpr(s32) = G_FCANONICALIZE %2
|
|
S_ENDPGM 0, implicit %3
|
|
...
|
|
|
|
---
|
|
|
|
name: fcanonicalize_fneg_fabs_f32_flush
|
|
legalized: true
|
|
regBankSelected: true
|
|
machineFunctionInfo:
|
|
mode:
|
|
fp32-input-denormals: false
|
|
fp32-output-denormals: false
|
|
|
|
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: %3:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1065353216, 2, [[V_XOR_B32_e32_]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX8: S_ENDPGM 0, implicit %3
|
|
; 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
|
|
; GFX9: [[V_XOR_B32_e32_:%[0-9]+]]:vgpr_32 = V_XOR_B32_e32 [[S_MOV_B32_]], [[COPY]], implicit $exec
|
|
; GFX9: %3:vgpr_32 = nofpexcept V_MAX_F32_e64 2, [[V_XOR_B32_e32_]], 2, [[V_XOR_B32_e32_]], 0, 0, implicit $mode, implicit $exec
|
|
; GFX9: S_ENDPGM 0, implicit %3
|
|
%0:vgpr(s32) = COPY $vgpr0
|
|
%1:vgpr(s32) = G_FNEG %0
|
|
%2:vgpr(s32) = G_FABS %1
|
|
%3:vgpr(s32) = G_FCANONICALIZE %2
|
|
S_ENDPGM 0, implicit %3
|
|
...
|