This is an archive of the discontinued LLVM Phabricator instance.

Differential D74753

AMDGPU/GlobalISel: Use packed for G_ADD/G_SUB/G_MUL v2s16
ClosedPublic

Authored by arsenm on Feb 17 2020, 7:12 PM.

Details

Reviewers
nhaehnle
kerbowa
foad

Diff Detail

Event Timeline

arsenm created this revision.Feb 17 2020, 7:12 PM
Herald added a project: Restricted Project. · View Herald TranscriptFeb 17 2020, 7:12 PM
Herald added subscribers: hiraditya, t-tye, tpr and 6 others. · View Herald Transcript
foad added a comment.Feb 18 2020, 10:28 AM

Any tests showing the generated code for add and sub?

arsenm added a comment.Feb 18 2020, 10:37 AM
In D74753#1881023, @foad wrote:

Any tests showing the generated code for add and sub?

At some point there needs to be mass port of end to end selection tests, but at this point it's exhausting to test them all and I'm anticipating a lot more merge conflicts as I fix the source modifier matching etc.

arsenm added a comment.Feb 24 2020, 6:27 PM

ping

foad accepted this revision.Feb 25 2020, 1:27 AM

LGTM.

This revision is now accepted and ready to land.Feb 25 2020, 1:27 AM
arsenm closed this revision.Feb 25 2020, 9:12 AM

86e13ec194a01744822606f4473d577cbc2c211a

Revision Contents

PathSize
llvm/
lib/
Target/
AMDGPU/
9 lines
test/
CodeGen/
AMDGPU/
GlobalISel/
233 lines
233 lines
233 lines
88 lines

Diff 245071

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

Show First 20 Lines • Show All 287 Lines • ▼ Show 20 Lines getActionDefinitionsBuilder(G_PHI)
.legalFor(AddrSpaces64) .legalFor(AddrSpaces64)
.legalFor(AddrSpaces32) .legalFor(AddrSpaces32)
.clampScalar(0, S32, S256) .clampScalar(0, S32, S256)
.widenScalarToNextPow2(0, 32) .widenScalarToNextPow2(0, 32)
.clampMaxNumElements(0, S32, 16) .clampMaxNumElements(0, S32, 16)
.moreElementsIf(isSmallOddVector(0), oneMoreElement(0)) .moreElementsIf(isSmallOddVector(0), oneMoreElement(0))
.legalIf(isPointer(0)); .legalIf(isPointer(0));
if (ST.has16BitInsts()) { if (ST.hasVOP3PInsts()) {
getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL})
.legalFor({S32, S16, V2S16})
.clampScalar(0, S16, S32)
.clampMaxNumElements(0, S16, 2)
.scalarize(0)
.widenScalarToNextPow2(0, 32);
} else if (ST.has16BitInsts()) {
getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL}) getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL})
.legalFor({S32, S16}) .legalFor({S32, S16})
.clampScalar(0, S16, S32) .clampScalar(0, S16, S32)
.scalarize(0) .scalarize(0)
.widenScalarToNextPow2(0, 32); .widenScalarToNextPow2(0, 32);
} else { } else {
getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL}) getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL})
.legalFor({S32}) .legalFor({S32})
▲ Show 20 Lines • Show All 3,384 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-add.mir

Show First 20 Lines • Show All 156 Lines • ▼ Show 20 Lines bb.0:
; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[ADD1]](s16) ; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[ADD1]](s16)
; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32) ; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]] ; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; GFX8: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) ; GFX8: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX8: $vgpr0 = COPY [[BITCAST2]](<2 x s16>) ; GFX8: $vgpr0 = COPY [[BITCAST2]](<2 x s16>)
; GFX9-LABEL: name: test_add_v2s16 ; GFX9-LABEL: name: test_add_v2s16
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0 ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1 ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) ; GFX9: [[ADD:%[0-9]+]]:_(<2 x s16>) = G_ADD [[COPY]], [[COPY1]]
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) ; GFX9: $vgpr0 = COPY [[ADD]](<2 x s16>)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; GFX9: [[ADD:%[0-9]+]]:_(s16) = G_ADD [[TRUNC]], [[TRUNC2]]
; GFX9: [[ADD1:%[0-9]+]]:_(s16) = G_ADD [[TRUNC1]], [[TRUNC3]]
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ADD]](s16)
; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ADD1]](s16)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
; GFX9: $vgpr0 = COPY [[BUILD_VECTOR_TRUNC]](<2 x s16>)
%0:_(<2 x s16>) = COPY $vgpr0 %0:_(<2 x s16>) = COPY $vgpr0
%1:_(<2 x s16>) = COPY $vgpr1 %1:_(<2 x s16>) = COPY $vgpr1
%2:_(<2 x s16>) = G_ADD %0, %1 %2:_(<2 x s16>) = G_ADD %0, %1
$vgpr0 = COPY %2 $vgpr0 = COPY %2
... ...
--- ---
name: test_add_v3s16
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX6-LABEL: name: test_add_v3s16
; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
; GFX6: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[COPY6]], [[COPY7]]
; GFX6: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[ADD]](s32)
; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY [[COPY4]](s32)
; GFX6: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[COPY8]], [[COPY9]]
; GFX6: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[ADD1]](s32)
; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; GFX6: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY5]](s32)
; GFX6: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[COPY10]], [[COPY11]]
; GFX6: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[ADD2]](s32)
; GFX6: S_ENDPGM 0, implicit [[TRUNC]](s16), implicit [[TRUNC1]](s16), implicit [[TRUNC2]](s16)
; GFX8-LABEL: name: test_add_v3s16
; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX8: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
; GFX8: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX8: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
; GFX8: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[COPY3]](s32)
; GFX8: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[COPY4]](s32)
; GFX8: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[COPY5]](s32)
; GFX8: [[ADD:%[0-9]+]]:_(s16) = G_ADD [[TRUNC]], [[TRUNC3]]
; GFX8: [[ADD1:%[0-9]+]]:_(s16) = G_ADD [[TRUNC1]], [[TRUNC4]]
; GFX8: [[ADD2:%[0-9]+]]:_(s16) = G_ADD [[TRUNC2]], [[TRUNC5]]
; GFX8: S_ENDPGM 0, implicit [[ADD]](s16), implicit [[ADD1]](s16), implicit [[ADD2]](s16)
; GFX9-LABEL: name: test_add_v3s16
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY5]](s32)
; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY6]](s32), [[COPY7]](s32)
; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[DEF]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY8]](s32), [[COPY9]](s32)
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
; GFX9: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0
; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY4]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY5]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
; GFX9: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS1]](<4 x s16>), 0
; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; GFX9: [[EXTRACT2:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0
; GFX9: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0
; GFX9: [[EXTRACT3:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
; GFX9: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT1]](<3 x s16>), 0
; GFX9: [[EXTRACT4:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT1]](<4 x s16>), 0
; GFX9: [[ADD:%[0-9]+]]:_(<2 x s16>) = G_ADD [[EXTRACT3]], [[EXTRACT4]]
; GFX9: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT2]](<3 x s16>), 0
; GFX9: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[ADD]](<2 x s16>), 0
; GFX9: [[EXTRACT5:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[INSERT3]](<4 x s16>), 0
; GFX9: [[COPY13:%[0-9]+]]:_(s16) = COPY [[TRUNC]](s16)
; GFX9: [[COPY14:%[0-9]+]]:_(s16) = COPY [[TRUNC1]](s16)
; GFX9: [[ADD1:%[0-9]+]]:_(s16) = G_ADD [[COPY13]], [[COPY14]]
; GFX9: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT5]](<3 x s16>), 0
; GFX9: [[INSERT5:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT4]], [[ADD1]](s16), 32
; GFX9: [[EXTRACT6:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[INSERT5]](<4 x s16>), 0
; GFX9: [[INSERT6:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT6]](<3 x s16>), 0
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT6]](<4 x s16>)
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: S_ENDPGM 0, implicit [[TRUNC2]](s16), implicit [[TRUNC3]](s16), implicit [[TRUNC4]](s16)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s32) = COPY $vgpr2
%3:_(s32) = COPY $vgpr3
%4:_(s32) = COPY $vgpr4
%5:_(s32) = COPY $vgpr5
%6:_(s16) = G_TRUNC %0
%7:_(s16) = G_TRUNC %1
%8:_(s16) = G_TRUNC %2
%9:_(s16) = G_TRUNC %3
%10:_(s16) = G_TRUNC %4
%11:_(s16) = G_TRUNC %5
%12:_(<3 x s16>) = G_BUILD_VECTOR %6, %7, %8
%13:_(<3 x s16>) = G_BUILD_VECTOR %9, %10, %11
%14:_(<3 x s16>) = G_ADD %12, %13
%15:_(s16), %16:_(s16), %17:_(s16) = G_UNMERGE_VALUES %14
S_ENDPGM 0, implicit %15, implicit %16, implicit %17
...
---
name: test_add_v4s16
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX6-LABEL: name: test_add_v4s16
; GFX6: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX6: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX6: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX6: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX6: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX6: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX6: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX6: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX6: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; GFX6: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; GFX6: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; GFX6: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; GFX6: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[COPY2]], [[COPY3]]
; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; GFX6: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[COPY4]], [[COPY5]]
; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; GFX6: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[COPY6]], [[COPY7]]
; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; GFX6: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[COPY8]], [[COPY9]]
; GFX6: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY [[ADD]](s32)
; GFX6: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
; GFX6: [[COPY11:%[0-9]+]]:_(s32) = COPY [[ADD1]](s32)
; GFX6: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
; GFX6: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
; GFX6: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
; GFX6: [[BITCAST4:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX6: [[COPY12:%[0-9]+]]:_(s32) = COPY [[ADD2]](s32)
; GFX6: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
; GFX6: [[COPY13:%[0-9]+]]:_(s32) = COPY [[ADD3]](s32)
; GFX6: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
; GFX6: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C]](s32)
; GFX6: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
; GFX6: [[BITCAST5:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; GFX6: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST4]](<2 x s16>), [[BITCAST5]](<2 x s16>)
; GFX6: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
; GFX8-LABEL: name: test_add_v4s16
; GFX8: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX8: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX8: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX8: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX8: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX8: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX8: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX8: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX8: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX8: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; GFX8: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX8: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; GFX8: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; GFX8: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; GFX8: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; GFX8: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; GFX8: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; GFX8: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; GFX8: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; GFX8: [[ADD:%[0-9]+]]:_(s16) = G_ADD [[TRUNC]], [[TRUNC4]]
; GFX8: [[ADD1:%[0-9]+]]:_(s16) = G_ADD [[TRUNC1]], [[TRUNC5]]
; GFX8: [[ADD2:%[0-9]+]]:_(s16) = G_ADD [[TRUNC2]], [[TRUNC6]]
; GFX8: [[ADD3:%[0-9]+]]:_(s16) = G_ADD [[TRUNC3]], [[TRUNC7]]
; GFX8: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ADD]](s16)
; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[ADD1]](s16)
; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; GFX8: [[BITCAST4:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX8: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[ADD2]](s16)
; GFX8: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[ADD3]](s16)
; GFX8: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
; GFX8: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
; GFX8: [[BITCAST5:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; GFX8: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST4]](<2 x s16>), [[BITCAST5]](<2 x s16>)
; GFX8: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
; GFX9-LABEL: name: test_add_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[ADD:%[0-9]+]]:_(<2 x s16>) = G_ADD [[UV]], [[UV2]]
; GFX9: [[ADD1:%[0-9]+]]:_(<2 x s16>) = G_ADD [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[ADD]](<2 x s16>), [[ADD1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_ADD %0, %1
$vgpr0_vgpr1 = COPY %2
...
---
name: test_add_s64 name: test_add_s64
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX6-LABEL: name: test_add_s64 ; GFX6-LABEL: name: test_add_s64
; GFX6: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX6: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 ; GFX6: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
▲ Show 20 Lines • Show All 129 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-mul.mir

Show First 20 Lines • Show All 294 Lines • ▼ Show 20 Lines bb.0:
; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[MUL1]](s16) ; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[MUL1]](s16)
; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32) ; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]] ; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; GFX8: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) ; GFX8: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX8: $vgpr0 = COPY [[BITCAST2]](<2 x s16>) ; GFX8: $vgpr0 = COPY [[BITCAST2]](<2 x s16>)
; GFX9-LABEL: name: test_mul_v2s16 ; GFX9-LABEL: name: test_mul_v2s16
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0 ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1 ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) ; GFX9: [[MUL:%[0-9]+]]:_(<2 x s16>) = G_MUL [[COPY]], [[COPY1]]
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) ; GFX9: $vgpr0 = COPY [[MUL]](<2 x s16>)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; GFX9: [[MUL:%[0-9]+]]:_(s16) = G_MUL [[TRUNC]], [[TRUNC2]]
; GFX9: [[MUL1:%[0-9]+]]:_(s16) = G_MUL [[TRUNC1]], [[TRUNC3]]
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MUL]](s16)
; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[MUL1]](s16)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
; GFX9: $vgpr0 = COPY [[BUILD_VECTOR_TRUNC]](<2 x s16>)
%0:_(<2 x s16>) = COPY $vgpr0 %0:_(<2 x s16>) = COPY $vgpr0
%1:_(<2 x s16>) = COPY $vgpr1 %1:_(<2 x s16>) = COPY $vgpr1
%2:_(<2 x s16>) = G_MUL %0, %1 %2:_(<2 x s16>) = G_MUL %0, %1
$vgpr0 = COPY %2 $vgpr0 = COPY %2
... ...
--- ---
name: test_mul_v3s16
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX6-LABEL: name: test_mul_v3s16
; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
; GFX6: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY6]], [[COPY7]]
; GFX6: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[MUL]](s32)
; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY [[COPY4]](s32)
; GFX6: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[COPY8]], [[COPY9]]
; GFX6: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[MUL1]](s32)
; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; GFX6: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY5]](s32)
; GFX6: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[COPY10]], [[COPY11]]
; GFX6: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[MUL2]](s32)
; GFX6: S_ENDPGM 0, implicit [[TRUNC]](s16), implicit [[TRUNC1]](s16), implicit [[TRUNC2]](s16)
; GFX8-LABEL: name: test_mul_v3s16
; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX8: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
; GFX8: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX8: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
; GFX8: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[COPY3]](s32)
; GFX8: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[COPY4]](s32)
; GFX8: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[COPY5]](s32)
; GFX8: [[MUL:%[0-9]+]]:_(s16) = G_MUL [[TRUNC]], [[TRUNC3]]
; GFX8: [[MUL1:%[0-9]+]]:_(s16) = G_MUL [[TRUNC1]], [[TRUNC4]]
; GFX8: [[MUL2:%[0-9]+]]:_(s16) = G_MUL [[TRUNC2]], [[TRUNC5]]
; GFX8: S_ENDPGM 0, implicit [[MUL]](s16), implicit [[MUL1]](s16), implicit [[MUL2]](s16)
; GFX9-LABEL: name: test_mul_v3s16
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY5]](s32)
; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY6]](s32), [[COPY7]](s32)
; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[DEF]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY8]](s32), [[COPY9]](s32)
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
; GFX9: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0
; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY4]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY5]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
; GFX9: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS1]](<4 x s16>), 0
; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; GFX9: [[EXTRACT2:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0
; GFX9: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0
; GFX9: [[EXTRACT3:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
; GFX9: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT1]](<3 x s16>), 0
; GFX9: [[EXTRACT4:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT1]](<4 x s16>), 0
; GFX9: [[MUL:%[0-9]+]]:_(<2 x s16>) = G_MUL [[EXTRACT3]], [[EXTRACT4]]
; GFX9: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT2]](<3 x s16>), 0
; GFX9: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[MUL]](<2 x s16>), 0
; GFX9: [[EXTRACT5:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[INSERT3]](<4 x s16>), 0
; GFX9: [[COPY13:%[0-9]+]]:_(s16) = COPY [[TRUNC]](s16)
; GFX9: [[COPY14:%[0-9]+]]:_(s16) = COPY [[TRUNC1]](s16)
; GFX9: [[MUL1:%[0-9]+]]:_(s16) = G_MUL [[COPY13]], [[COPY14]]
; GFX9: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT5]](<3 x s16>), 0
; GFX9: [[INSERT5:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT4]], [[MUL1]](s16), 32
; GFX9: [[EXTRACT6:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[INSERT5]](<4 x s16>), 0
; GFX9: [[INSERT6:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT6]](<3 x s16>), 0
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT6]](<4 x s16>)
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: S_ENDPGM 0, implicit [[TRUNC2]](s16), implicit [[TRUNC3]](s16), implicit [[TRUNC4]](s16)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s32) = COPY $vgpr2
%3:_(s32) = COPY $vgpr3
%4:_(s32) = COPY $vgpr4
%5:_(s32) = COPY $vgpr5
%6:_(s16) = G_TRUNC %0
%7:_(s16) = G_TRUNC %1
%8:_(s16) = G_TRUNC %2
%9:_(s16) = G_TRUNC %3
%10:_(s16) = G_TRUNC %4
%11:_(s16) = G_TRUNC %5
%12:_(<3 x s16>) = G_BUILD_VECTOR %6, %7, %8
%13:_(<3 x s16>) = G_BUILD_VECTOR %9, %10, %11
%14:_(<3 x s16>) = G_MUL %12, %13
%15:_(s16), %16:_(s16), %17:_(s16) = G_UNMERGE_VALUES %14
S_ENDPGM 0, implicit %15, implicit %16, implicit %17
...
---
name: test_mul_v4s16
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX6-LABEL: name: test_mul_v4s16
; GFX6: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX6: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX6: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX6: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX6: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX6: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX6: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX6: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX6: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; GFX6: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; GFX6: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; GFX6: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; GFX6: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[COPY2]], [[COPY3]]
; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; GFX6: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[COPY4]], [[COPY5]]
; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; GFX6: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[COPY6]], [[COPY7]]
; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; GFX6: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[COPY8]], [[COPY9]]
; GFX6: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY [[MUL]](s32)
; GFX6: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
; GFX6: [[COPY11:%[0-9]+]]:_(s32) = COPY [[MUL1]](s32)
; GFX6: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
; GFX6: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
; GFX6: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
; GFX6: [[BITCAST4:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX6: [[COPY12:%[0-9]+]]:_(s32) = COPY [[MUL2]](s32)
; GFX6: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
; GFX6: [[COPY13:%[0-9]+]]:_(s32) = COPY [[MUL3]](s32)
; GFX6: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
; GFX6: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C]](s32)
; GFX6: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
; GFX6: [[BITCAST5:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; GFX6: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST4]](<2 x s16>), [[BITCAST5]](<2 x s16>)
; GFX6: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
; GFX8-LABEL: name: test_mul_v4s16
; GFX8: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX8: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX8: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX8: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX8: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX8: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX8: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX8: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX8: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX8: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; GFX8: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX8: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; GFX8: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; GFX8: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; GFX8: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; GFX8: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; GFX8: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; GFX8: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; GFX8: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; GFX8: [[MUL:%[0-9]+]]:_(s16) = G_MUL [[TRUNC]], [[TRUNC4]]
; GFX8: [[MUL1:%[0-9]+]]:_(s16) = G_MUL [[TRUNC1]], [[TRUNC5]]
; GFX8: [[MUL2:%[0-9]+]]:_(s16) = G_MUL [[TRUNC2]], [[TRUNC6]]
; GFX8: [[MUL3:%[0-9]+]]:_(s16) = G_MUL [[TRUNC3]], [[TRUNC7]]
; GFX8: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[MUL]](s16)
; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[MUL1]](s16)
; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; GFX8: [[BITCAST4:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX8: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[MUL2]](s16)
; GFX8: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[MUL3]](s16)
; GFX8: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
; GFX8: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
; GFX8: [[BITCAST5:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; GFX8: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST4]](<2 x s16>), [[BITCAST5]](<2 x s16>)
; GFX8: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
; GFX9-LABEL: name: test_mul_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[MUL:%[0-9]+]]:_(<2 x s16>) = G_MUL [[UV]], [[UV2]]
; GFX9: [[MUL1:%[0-9]+]]:_(<2 x s16>) = G_MUL [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[MUL]](<2 x s16>), [[MUL1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_MUL %0, %1
$vgpr0_vgpr1 = COPY %2
...
---
name: test_mul_s24 name: test_mul_s24
body: | body: |
bb.0: bb.0:
liveins: $vgpr0, $vgpr1 liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: test_mul_s24 ; GFX6-LABEL: name: test_mul_s24
; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 ; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 ; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
▲ Show 20 Lines • Show All 45 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sub.mir

Show First 20 Lines • Show All 156 Lines • ▼ Show 20 Lines bb.0:
; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[SUB1]](s16) ; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[SUB1]](s16)
; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32) ; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]] ; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; GFX8: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32) ; GFX8: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX8: $vgpr0 = COPY [[BITCAST2]](<2 x s16>) ; GFX8: $vgpr0 = COPY [[BITCAST2]](<2 x s16>)
; GFX9-LABEL: name: test_sub_v2s16 ; GFX9-LABEL: name: test_sub_v2s16
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0 ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1 ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>) ; GFX9: [[SUB:%[0-9]+]]:_(<2 x s16>) = G_SUB [[COPY]], [[COPY1]]
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) ; GFX9: $vgpr0 = COPY [[SUB]](<2 x s16>)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; GFX9: [[SUB:%[0-9]+]]:_(s16) = G_SUB [[TRUNC]], [[TRUNC2]]
; GFX9: [[SUB1:%[0-9]+]]:_(s16) = G_SUB [[TRUNC1]], [[TRUNC3]]
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SUB]](s16)
; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[SUB1]](s16)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
; GFX9: $vgpr0 = COPY [[BUILD_VECTOR_TRUNC]](<2 x s16>)
%0:_(<2 x s16>) = COPY $vgpr0 %0:_(<2 x s16>) = COPY $vgpr0
%1:_(<2 x s16>) = COPY $vgpr1 %1:_(<2 x s16>) = COPY $vgpr1
%2:_(<2 x s16>) = G_SUB %0, %1 %2:_(<2 x s16>) = G_SUB %0, %1
$vgpr0 = COPY %2 $vgpr0 = COPY %2
... ...
--- ---
name: test_sub_v3s16
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX6-LABEL: name: test_sub_v3s16
; GFX6: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX6: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
; GFX6: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[COPY6]], [[COPY7]]
; GFX6: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SUB]](s32)
; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY [[COPY4]](s32)
; GFX6: [[SUB1:%[0-9]+]]:_(s32) = G_SUB [[COPY8]], [[COPY9]]
; GFX6: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SUB1]](s32)
; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; GFX6: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY5]](s32)
; GFX6: [[SUB2:%[0-9]+]]:_(s32) = G_SUB [[COPY10]], [[COPY11]]
; GFX6: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SUB2]](s32)
; GFX6: S_ENDPGM 0, implicit [[TRUNC]](s16), implicit [[TRUNC1]](s16), implicit [[TRUNC2]](s16)
; GFX8-LABEL: name: test_sub_v3s16
; GFX8: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX8: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX8: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX8: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX8: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX8: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX8: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
; GFX8: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX8: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
; GFX8: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[COPY3]](s32)
; GFX8: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[COPY4]](s32)
; GFX8: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[COPY5]](s32)
; GFX8: [[SUB:%[0-9]+]]:_(s16) = G_SUB [[TRUNC]], [[TRUNC3]]
; GFX8: [[SUB1:%[0-9]+]]:_(s16) = G_SUB [[TRUNC1]], [[TRUNC4]]
; GFX8: [[SUB2:%[0-9]+]]:_(s16) = G_SUB [[TRUNC2]], [[TRUNC5]]
; GFX8: S_ENDPGM 0, implicit [[SUB]](s16), implicit [[SUB1]](s16), implicit [[SUB2]](s16)
; GFX9-LABEL: name: test_sub_v3s16
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX9: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY5]](s32)
; GFX9: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; GFX9: [[COPY7:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY6]](s32), [[COPY7]](s32)
; GFX9: [[COPY8:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
; GFX9: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GFX9: [[COPY9:%[0-9]+]]:_(s32) = COPY [[DEF]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY8]](s32), [[COPY9]](s32)
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
; GFX9: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS]](<4 x s16>), 0
; GFX9: [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY3]](s32)
; GFX9: [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY4]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
; GFX9: [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY5]](s32)
; GFX9: [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
; GFX9: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS1]](<4 x s16>), 0
; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; GFX9: [[EXTRACT2:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0
; GFX9: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0
; GFX9: [[EXTRACT3:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
; GFX9: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT1]](<3 x s16>), 0
; GFX9: [[EXTRACT4:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT1]](<4 x s16>), 0
; GFX9: [[SUB:%[0-9]+]]:_(<2 x s16>) = G_SUB [[EXTRACT3]], [[EXTRACT4]]
; GFX9: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT2]](<3 x s16>), 0
; GFX9: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT2]], [[SUB]](<2 x s16>), 0
; GFX9: [[EXTRACT5:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[INSERT3]](<4 x s16>), 0
; GFX9: [[COPY13:%[0-9]+]]:_(s16) = COPY [[TRUNC]](s16)
; GFX9: [[COPY14:%[0-9]+]]:_(s16) = COPY [[TRUNC1]](s16)
; GFX9: [[SUB1:%[0-9]+]]:_(s16) = G_SUB [[COPY13]], [[COPY14]]
; GFX9: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT5]](<3 x s16>), 0
; GFX9: [[INSERT5:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[INSERT4]], [[SUB1]](s16), 32
; GFX9: [[EXTRACT6:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[INSERT5]](<4 x s16>), 0
; GFX9: [[INSERT6:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT6]](<3 x s16>), 0
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT6]](<4 x s16>)
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: S_ENDPGM 0, implicit [[TRUNC2]](s16), implicit [[TRUNC3]](s16), implicit [[TRUNC4]](s16)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s32) = COPY $vgpr2
%3:_(s32) = COPY $vgpr3
%4:_(s32) = COPY $vgpr4
%5:_(s32) = COPY $vgpr5
%6:_(s16) = G_TRUNC %0
%7:_(s16) = G_TRUNC %1
%8:_(s16) = G_TRUNC %2
%9:_(s16) = G_TRUNC %3
%10:_(s16) = G_TRUNC %4
%11:_(s16) = G_TRUNC %5
%12:_(<3 x s16>) = G_BUILD_VECTOR %6, %7, %8
%13:_(<3 x s16>) = G_BUILD_VECTOR %9, %10, %11
%14:_(<3 x s16>) = G_SUB %12, %13
%15:_(s16), %16:_(s16), %17:_(s16) = G_UNMERGE_VALUES %14
S_ENDPGM 0, implicit %15, implicit %16, implicit %17
...
---
name: test_sub_v4s16
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX6-LABEL: name: test_sub_v4s16
; GFX6: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX6: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX6: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX6: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX6: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX6: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX6: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX6: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX6: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; GFX6: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; GFX6: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; GFX6: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; GFX6: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; GFX6: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; GFX6: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[COPY2]], [[COPY3]]
; GFX6: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; GFX6: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; GFX6: [[SUB1:%[0-9]+]]:_(s32) = G_SUB [[COPY4]], [[COPY5]]
; GFX6: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; GFX6: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; GFX6: [[SUB2:%[0-9]+]]:_(s32) = G_SUB [[COPY6]], [[COPY7]]
; GFX6: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; GFX6: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; GFX6: [[SUB3:%[0-9]+]]:_(s32) = G_SUB [[COPY8]], [[COPY9]]
; GFX6: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; GFX6: [[COPY10:%[0-9]+]]:_(s32) = COPY [[SUB]](s32)
; GFX6: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
; GFX6: [[COPY11:%[0-9]+]]:_(s32) = COPY [[SUB1]](s32)
; GFX6: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
; GFX6: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
; GFX6: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
; GFX6: [[BITCAST4:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX6: [[COPY12:%[0-9]+]]:_(s32) = COPY [[SUB2]](s32)
; GFX6: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
; GFX6: [[COPY13:%[0-9]+]]:_(s32) = COPY [[SUB3]](s32)
; GFX6: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
; GFX6: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C]](s32)
; GFX6: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
; GFX6: [[BITCAST5:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; GFX6: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST4]](<2 x s16>), [[BITCAST5]](<2 x s16>)
; GFX6: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
; GFX8-LABEL: name: test_sub_v4s16
; GFX8: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX8: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX8: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX8: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX8: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX8: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX8: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX8: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX8: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX8: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX8: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; GFX8: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX8: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; GFX8: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; GFX8: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; GFX8: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; GFX8: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; GFX8: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; GFX8: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; GFX8: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; GFX8: [[SUB:%[0-9]+]]:_(s16) = G_SUB [[TRUNC]], [[TRUNC4]]
; GFX8: [[SUB1:%[0-9]+]]:_(s16) = G_SUB [[TRUNC1]], [[TRUNC5]]
; GFX8: [[SUB2:%[0-9]+]]:_(s16) = G_SUB [[TRUNC2]], [[TRUNC6]]
; GFX8: [[SUB3:%[0-9]+]]:_(s16) = G_SUB [[TRUNC3]], [[TRUNC7]]
; GFX8: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[SUB]](s16)
; GFX8: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[SUB1]](s16)
; GFX8: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; GFX8: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; GFX8: [[BITCAST4:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; GFX8: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[SUB2]](s16)
; GFX8: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[SUB3]](s16)
; GFX8: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
; GFX8: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
; GFX8: [[BITCAST5:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; GFX8: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST4]](<2 x s16>), [[BITCAST5]](<2 x s16>)
; GFX8: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
; GFX9-LABEL: name: test_sub_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[SUB:%[0-9]+]]:_(<2 x s16>) = G_SUB [[UV]], [[UV2]]
; GFX9: [[SUB1:%[0-9]+]]:_(<2 x s16>) = G_SUB [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[SUB]](<2 x s16>), [[SUB1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_SUB %0, %1
$vgpr0_vgpr1 = COPY %2
...
---
name: test_sub_s64 name: test_sub_s64
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX6-LABEL: name: test_sub_s64 ; GFX6-LABEL: name: test_sub_s64
; GFX6: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX6: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 ; GFX6: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
▲ Show 20 Lines • Show All 123 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/mul.v2i16.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=fiji < %s | FileCheck -check-prefix=GFX8 %s
define <2 x i16> @v_mul_v2i16(<2 x i16> %a, <2 x i16> %b) {
; GFX9-LABEL: v_mul_v2i16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_pk_mul_lo_u16 v0, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_mul_v2i16:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_mul_lo_u16_e32 v2, v0, v1
; GFX8-NEXT: v_mul_lo_u16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; GFX8-NEXT: v_or_b32_e32 v0, v2, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
%mul = mul <2 x i16> %a, %b
ret <2 x i16> %mul
}
define <2 x i16> @v_mul_v2i16_fneg_lhs(<2 x half> %a, <2 x i16> %b) {
; GFX9-LABEL: v_mul_v2i16_fneg_lhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_pk_mul_lo_u16 v0, v0, v1 neg_lo:[1,0] neg_hi:[1,0]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_mul_v2i16_fneg_lhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
; GFX8-NEXT: v_mul_lo_u16_e32 v2, v0, v1
; GFX8-NEXT: v_mul_lo_u16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; GFX8-NEXT: v_or_b32_e32 v0, v2, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
%neg.a = fneg <2 x half> %a
%cast.neg.a = bitcast <2 x half> %neg.a to <2 x i16>
%mul = mul <2 x i16> %cast.neg.a, %b
ret <2 x i16> %mul
}
define <2 x i16> @v_mul_v2i16_fneg_rhs(<2 x i16> %a, <2 x half> %b) {
; GFX9-LABEL: v_mul_v2i16_fneg_rhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_pk_mul_lo_u16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_mul_v2i16_fneg_rhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_xor_b32_e32 v1, 0x80008000, v1
; GFX8-NEXT: v_mul_lo_u16_e32 v2, v0, v1
; GFX8-NEXT: v_mul_lo_u16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; GFX8-NEXT: v_or_b32_e32 v0, v2, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
%neg.b = fneg <2 x half> %b
%cast.neg.b = bitcast <2 x half> %neg.b to <2 x i16>
%mul = mul <2 x i16> %a, %cast.neg.b
ret <2 x i16> %mul
}
define <2 x i16> @v_mul_v2i16_fneg_lhs_fneg_rhs(<2 x half> %a, <2 x half> %b) {
; GFX9-LABEL: v_mul_v2i16_fneg_lhs_fneg_rhs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_pk_mul_lo_u16 v0, v0, v1 neg_lo:[1,1] neg_hi:[1,1]
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_mul_v2i16_fneg_lhs_fneg_rhs:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: s_mov_b32 s4, 0x80008000
; GFX8-NEXT: v_xor_b32_e32 v0, s4, v0
; GFX8-NEXT: v_xor_b32_e32 v1, s4, v1
; GFX8-NEXT: v_mul_lo_u16_e32 v2, v0, v1
; GFX8-NEXT: v_mul_lo_u16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; GFX8-NEXT: v_or_b32_e32 v0, v2, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
%neg.a = fneg <2 x half> %a
%neg.b = fneg <2 x half> %b
%cast.neg.a = bitcast <2 x half> %neg.a to <2 x i16>
%cast.neg.b = bitcast <2 x half> %neg.b to <2 x i16>
%mul = mul <2 x i16> %cast.neg.a, %cast.neg.b
ret <2 x i16> %mul
}