llvm-for-llvmta/test/CodeGen/AMDGPU/GlobalISel/inst-select-fmul.v2s16.mir

75 lines
2.5 KiB
YAML

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=instruction-select -global-isel-abort=0 -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
---
name: fmul_v2f16_vv
legalized: true
regBankSelected: true
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX9-LABEL: name: fmul_v2f16_vv
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9: %2:vgpr_32 = nofpexcept V_PK_MUL_F16 8, [[COPY]], 8, [[COPY1]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX9: S_ENDPGM 0, implicit %2
%0:vgpr(<2 x s16>) = COPY $vgpr0
%1:vgpr(<2 x s16>) = COPY $vgpr1
%2:vgpr(<2 x s16>) = G_FMUL %0, %1
S_ENDPGM 0, implicit %2
...
---
name: fmul_v2f16_fneg_v_fneg_v
legalized: true
regBankSelected: true
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX9-LABEL: name: fmul_v2f16_fneg_v_fneg_v
; GFX9: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9: %4:vgpr_32 = nofpexcept V_PK_MUL_F16 11, [[COPY]], 11, [[COPY1]], 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX9: S_ENDPGM 0, implicit %4
%0:vgpr(<2 x s16>) = COPY $vgpr0
%1:vgpr(<2 x s16>) = COPY $vgpr1
%2:vgpr(<2 x s16>) = G_FNEG %0
%3:vgpr(<2 x s16>) = G_FNEG %1
%4:vgpr(<2 x s16>) = G_FMUL %2, %3
S_ENDPGM 0, implicit %4
...
---
name: fmul_v2f16_fneg_lo_v_v
legalized: true
regBankSelected: true
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-LABEL: name: fmul_v2f16_fneg_lo_v_v
; GFX9: [[COPY:%[0-9]+]]:vgpr_32(<2 x s16>) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
; GFX9: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX9: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY1]](s32)
; GFX9: [[FNEG:%[0-9]+]]:vgpr(s16) = G_FNEG [[TRUNC]]
; GFX9: [[ANYEXT:%[0-9]+]]:vgpr(s32) = G_ANYEXT [[FNEG]](s16)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:vgpr_32(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[COPY2]](s32)
; GFX9: %7:vgpr_32(<2 x s16>) = nofpexcept V_PK_MUL_F16 8, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 8, [[COPY]](<2 x s16>), 0, 0, 0, 0, 0, implicit $mode, implicit $exec
; GFX9: S_ENDPGM 0, implicit %7(<2 x s16>)
%0:vgpr(<2 x s16>) = COPY $vgpr0
%1:vgpr(s32) = COPY $vgpr1
%2:vgpr(s32) = COPY $vgpr2
%3:vgpr(s16) = G_TRUNC %1
%4:vgpr(s16) = G_FNEG %3
%5:vgpr(s32) = G_ANYEXT %4
%6:vgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC %5, %2
%7:vgpr(<2 x s16>) = G_FMUL %6, %0
S_ENDPGM 0, implicit %7
...