This is an archive of the discontinued LLVM Phabricator instance.

Differential D65412

GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES
ClosedPublic

Authored by arsenm on Jul 29 2019, 1:02 PM.

Details

Reviewers
aemerson
paquette
aditya_nandakumar
dsanders
Summary

Odd sized vectors aren't handled yet.

Diff Detail

Event Timeline

arsenm created this revision.Jul 29 2019, 1:02 PM
Herald added subscribers: Petar.Avramovic, volkan, tschuett and 4 others. · View Herald TranscriptJul 29 2019, 1:03 PM
arsenm updated this revision to Diff 212883.Aug 1 2019, 12:34 PM
arsenm added a reviewer: dsanders.

Rebase

paquette accepted this revision.Aug 13 2019, 9:01 AM

LGTM

This revision is now accepted and ready to land.Aug 13 2019, 9:01 AM
arsenm closed this revision.Aug 13 2019, 9:25 AM

r368713

Revision Contents

PathSize
include/
llvm/
CodeGen/
GlobalISel/
3 lines
lib/
CodeGen/
GlobalISel/
62 lines
Target/
AMDGPU/
11 lines
test/
CodeGen/
AMDGPU/
GlobalISel/
87 lines
38 lines
66 lines
38 lines
94 lines
74 lines
74 lines
66 lines
38 lines
19 lines
57 lines
111 lines
85 lines
37 lines
25 lines
71 lines
82 lines
82 lines
75 lines
75 lines
56 lines
22 lines

Diff 212883

include/llvm/CodeGen/GlobalISel/LegalizerHelper.h

Show First 20 Lines • Show All 194 Lines • ▼ Show 20 Linespublic:
fewerElementsVectorSelect(MachineInstr &MI, unsigned TypeIdx, LLT NarrowTy); fewerElementsVectorSelect(MachineInstr &MI, unsigned TypeIdx, LLT NarrowTy);
LegalizeResult fewerElementsVectorPhi(MachineInstr &MI, LegalizeResult fewerElementsVectorPhi(MachineInstr &MI,
unsigned TypeIdx, LLT NarrowTy); unsigned TypeIdx, LLT NarrowTy);
LegalizeResult moreElementsVectorPhi(MachineInstr &MI, unsigned TypeIdx, LegalizeResult moreElementsVectorPhi(MachineInstr &MI, unsigned TypeIdx,
LLT MoreTy); LLT MoreTy);
LegalizeResult fewerElementsVectorUnmergeValues(MachineInstr &MI,
unsigned TypeIdx,
LLT NarrowTy);
LegalizeResult LegalizeResult
reduceLoadStoreWidth(MachineInstr &MI, unsigned TypeIdx, LLT NarrowTy); reduceLoadStoreWidth(MachineInstr &MI, unsigned TypeIdx, LLT NarrowTy);
LegalizeResult narrowScalarShiftByConstant(MachineInstr &MI, const APInt &Amt, LegalizeResult narrowScalarShiftByConstant(MachineInstr &MI, const APInt &Amt,
LLT HalfTy, LLT ShiftAmtTy); LLT HalfTy, LLT ShiftAmtTy);
LegalizeResult narrowScalarShift(MachineInstr &MI, unsigned TypeIdx, LLT Ty); LegalizeResult narrowScalarShift(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
LegalizeResult narrowScalarMul(MachineInstr &MI, LLT Ty); LegalizeResult narrowScalarMul(MachineInstr &MI, LLT Ty);
Show All 37 Lines

lib/CodeGen/GlobalISel/LegalizerHelper.cpp

Show First 20 Lines • Show All 165 Lines • ▼ Show 20 Lines for (unsigned Offset = MainSize * NumParts; Offset < RegSize;
Register NewReg = MRI.createGenericVirtualRegister(LeftoverTy); Register NewReg = MRI.createGenericVirtualRegister(LeftoverTy);
LeftoverRegs.push_back(NewReg); LeftoverRegs.push_back(NewReg);
MIRBuilder.buildExtract(NewReg, Reg, Offset); MIRBuilder.buildExtract(NewReg, Reg, Offset);
} }
return true; return true;
} }
static LLT getGCDType(LLT OrigTy, LLT TargetTy) {
if (OrigTy.isVector() && TargetTy.isVector()) {
assert(OrigTy.getElementType() == TargetTy.getElementType());
int GCD = greatestCommonDivisor(OrigTy.getNumElements(),
TargetTy.getNumElements());
return LLT::scalarOrVector(GCD, OrigTy.getElementType());
}
if (OrigTy.isVector() && !TargetTy.isVector()) {
assert(OrigTy.getElementType() == TargetTy);
return TargetTy;
}
assert(!OrigTy.isVector() && !TargetTy.isVector());
int GCD = greatestCommonDivisor(OrigTy.getSizeInBits(),
TargetTy.getSizeInBits());
return LLT::scalar(GCD);
}
void LegalizerHelper::insertParts(Register DstReg, void LegalizerHelper::insertParts(Register DstReg,
LLT ResultTy, LLT PartTy, LLT ResultTy, LLT PartTy,
ArrayRef<Register> PartRegs, ArrayRef<Register> PartRegs,
LLT LeftoverTy, LLT LeftoverTy,
ArrayRef<Register> LeftoverRegs) { ArrayRef<Register> LeftoverRegs) {
if (!LeftoverTy.isValid()) { if (!LeftoverTy.isValid()) {
assert(LeftoverRegs.empty()); assert(LeftoverRegs.empty());
▲ Show 20 Lines • Show All 2,169 Lines • ▼ Show 20 Lines for (unsigned I = 1, E = MI.getNumOperands(); I != E; I += 2) {
} }
} }
MI.eraseFromParent(); MI.eraseFromParent();
return Legalized; return Legalized;
} }
LegalizerHelper::LegalizeResult LegalizerHelper::LegalizeResult
LegalizerHelper::fewerElementsVectorUnmergeValues(MachineInstr &MI,
unsigned TypeIdx,
LLT NarrowTy) {
if (TypeIdx != 1)
return UnableToLegalize;
const int NumDst = MI.getNumOperands() - 1;
const Register SrcReg = MI.getOperand(NumDst).getReg();
LLT SrcTy = MRI.getType(SrcReg);
LLT DstTy = MRI.getType(MI.getOperand(0).getReg());
// TODO: Create sequence of extracts.
if (DstTy == NarrowTy)
return UnableToLegalize;
LLT GCDTy = getGCDType(SrcTy, NarrowTy);
if (DstTy == GCDTy) {
// This would just be a copy of the same unmerge.
// TODO: Create extracts, pad with undef and create intermediate merges.
return UnableToLegalize;
}
auto Unmerge = MIRBuilder.buildUnmerge(GCDTy, SrcReg);
const int NumUnmerge = Unmerge->getNumOperands() - 1;
const int PartsPerUnmerge = NumDst / NumUnmerge;
for (int I = 0; I != NumUnmerge; ++I) {
auto MIB = MIRBuilder.buildInstr(TargetOpcode::G_UNMERGE_VALUES);
for (int J = 0; J != PartsPerUnmerge; ++J)
MIB.addDef(MI.getOperand(I * PartsPerUnmerge + J).getReg());
MIB.addUse(Unmerge.getReg(I));
}
MI.eraseFromParent();
return Legalized;
}
LegalizerHelper::LegalizeResult
LegalizerHelper::reduceLoadStoreWidth(MachineInstr &MI, unsigned TypeIdx, LegalizerHelper::reduceLoadStoreWidth(MachineInstr &MI, unsigned TypeIdx,
LLT NarrowTy) { LLT NarrowTy) {
// FIXME: Don't know how to handle secondary types yet. // FIXME: Don't know how to handle secondary types yet.
if (TypeIdx != 0) if (TypeIdx != 0)
return UnableToLegalize; return UnableToLegalize;
MachineMemOperand *MMO = *MI.memoperands_begin(); MachineMemOperand *MMO = *MI.memoperands_begin();
▲ Show 20 Lines • Show All 154 Lines • ▼ Show 20 Lines case G_ADDRSPACE_CAST:
return fewerElementsVectorCasts(MI, TypeIdx, NarrowTy); return fewerElementsVectorCasts(MI, TypeIdx, NarrowTy);
case G_ICMP: case G_ICMP:
case G_FCMP: case G_FCMP:
return fewerElementsVectorCmp(MI, TypeIdx, NarrowTy); return fewerElementsVectorCmp(MI, TypeIdx, NarrowTy);
case G_SELECT: case G_SELECT:
return fewerElementsVectorSelect(MI, TypeIdx, NarrowTy); return fewerElementsVectorSelect(MI, TypeIdx, NarrowTy);
case G_PHI: case G_PHI:
return fewerElementsVectorPhi(MI, TypeIdx, NarrowTy); return fewerElementsVectorPhi(MI, TypeIdx, NarrowTy);
case G_UNMERGE_VALUES:
return fewerElementsVectorUnmergeValues(MI, TypeIdx, NarrowTy);
case G_LOAD: case G_LOAD:
case G_STORE: case G_STORE:
return reduceLoadStoreWidth(MI, TypeIdx, NarrowTy); return reduceLoadStoreWidth(MI, TypeIdx, NarrowTy);
default: default:
return UnableToLegalize; return UnableToLegalize;
} }
} }
▲ Show 20 Lines • Show All 1,001 Lines • Show Last 20 Lines

lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

Show First 20 Lines • Show All 125 Lines • ▼ Show 20 Lines
static LegalityPredicate isWideScalarTruncStore(unsigned TypeIdx) { static LegalityPredicate isWideScalarTruncStore(unsigned TypeIdx) {
return [=](const LegalityQuery &Query) { return [=](const LegalityQuery &Query) {
const LLT Ty = Query.Types[TypeIdx]; const LLT Ty = Query.Types[TypeIdx];
return !Ty.isVector() && Ty.getSizeInBits() > 32 && return !Ty.isVector() && Ty.getSizeInBits() > 32 &&
Query.MMODescrs[0].SizeInBits < Ty.getSizeInBits(); Query.MMODescrs[0].SizeInBits < Ty.getSizeInBits();
}; };
} }
static LegalityPredicate elementTypeIs(unsigned TypeIdx, LLT Type) {
return [=](const LegalityQuery &Query) {
return Query.Types[TypeIdx].getElementType() == Type;
};
}
AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_, AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
const GCNTargetMachine &TM) const GCNTargetMachine &TM)
: ST(ST_) { : ST(ST_) {
using namespace TargetOpcode; using namespace TargetOpcode;
auto GetAddrSpacePtr = [&TM](unsigned AS) { auto GetAddrSpacePtr = [&TM](unsigned AS) {
return LLT::pointer(AS, TM.getPointerSizeInBits(AS)); return LLT::pointer(AS, TM.getPointerSizeInBits(AS));
}; };
▲ Show 20 Lines • Show All 778 Lines • ▼ Show 20 Lines for (unsigned Op : {G_MERGE_VALUES, G_UNMERGE_VALUES}) {
getActionDefinitionsBuilder(Op) getActionDefinitionsBuilder(Op)
.widenScalarToNextPow2(LitTyIdx, /*Min*/ 16) .widenScalarToNextPow2(LitTyIdx, /*Min*/ 16)
// Clamp the little scalar to s8-s256 and make it a power of 2. It's not // Clamp the little scalar to s8-s256 and make it a power of 2. It's not
// worth considering the multiples of 64 since 2*192 and 2*384 are not // worth considering the multiples of 64 since 2*192 and 2*384 are not
// valid. // valid.
.clampScalar(LitTyIdx, S16, S256) .clampScalar(LitTyIdx, S16, S256)
.widenScalarToNextPow2(LitTyIdx, /*Min*/ 32) .widenScalarToNextPow2(LitTyIdx, /*Min*/ 32)
.legalIf(all(typeIs(0, S16), typeIs(1, LLT::vector(3, 16)))) // FIXME: Testing hack
.fewerElementsIf(all(typeIs(0, S16), vectorWiderThan(1, 32),
elementTypeIs(1, S16)),
changeTo(1, V2S16))
// Break up vectors with weird elements into scalars // Break up vectors with weird elements into scalars
.fewerElementsIf( .fewerElementsIf(
[=](const LegalityQuery &Query) { return notValidElt(Query, 0); }, [=](const LegalityQuery &Query) { return notValidElt(Query, 0); },
scalarize(0)) scalarize(0))
.fewerElementsIf( .fewerElementsIf(
[=](const LegalityQuery &Query) { return notValidElt(Query, 1); }, [=](const LegalityQuery &Query) { return notValidElt(Query, 1); },
scalarize(1)) scalarize(1))
.clampScalar(BigTyIdx, S32, S512) .clampScalar(BigTyIdx, S32, S512)
▲ Show 20 Lines • Show All 700 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-ashr.mir

Show First 20 Lines • Show All 689 Lines • ▼ Show 20 Lines
name: test_ashr_v4s16_v4s16 name: test_ashr_v4s16_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_ashr_v4s16_v4s16 ; SI-LABEL: name: test_ashr_v4s16_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SEXT]], [[ZEXT]](s32) ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[ASHR]](s32) ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) ; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) ; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SEXT1]], [[ZEXT1]](s32) ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[ASHR1]](s32) ; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV6]](s16) ; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; SI: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SEXT2]], [[ZEXT2]](s32) ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[ASHR2]](s32) ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV7]](s16) ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; SI: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY3]], [[C]](s32)
; SI: [[ASHR3:%[0-9]+]]:_(s32) = G_ASHR [[SEXT3]], [[ZEXT3]](s32) ; SI: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[ASHR3]](s32) ; SI: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[ASHR]], [[AND]](s32)
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[ASHR1]](s32)
; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY5]], [[C]](s32)
; SI: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C]](s32)
; SI: [[ASHR3:%[0-9]+]]:_(s32) = G_ASHR [[ASHR2]], [[AND1]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[ASHR3]](s32)
; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY7]], [[C]](s32)
; SI: [[ASHR4:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C]](s32)
; SI: [[ASHR5:%[0-9]+]]:_(s32) = G_ASHR [[ASHR4]], [[AND2]](s32)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[ASHR5]](s32)
; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[COPY9]], [[C]](s32)
; SI: [[ASHR6:%[0-9]+]]:_(s32) = G_ASHR [[SHL3]], [[C]](s32)
; SI: [[ASHR7:%[0-9]+]]:_(s32) = G_ASHR [[ASHR6]], [[AND3]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[ASHR7]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_ashr_v4s16_v4s16 ; VI-LABEL: name: test_ashr_v4s16_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[UV]], [[UV4]](s16) ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[UV1]], [[UV5]](s16) ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[ASHR2:%[0-9]+]]:_(s16) = G_ASHR [[UV2]], [[UV6]](s16) ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[ASHR3:%[0-9]+]]:_(s16) = G_ASHR [[UV3]], [[UV7]](s16) ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[TRUNC]], [[TRUNC4]](s16)
; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[TRUNC1]], [[TRUNC5]](s16)
; VI: [[ASHR2:%[0-9]+]]:_(s16) = G_ASHR [[TRUNC2]], [[TRUNC6]](s16)
; VI: [[ASHR3:%[0-9]+]]:_(s16) = G_ASHR [[TRUNC3]], [[TRUNC7]](s16)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[ASHR]](s16), [[ASHR1]](s16), [[ASHR2]](s16), [[ASHR3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[ASHR]](s16), [[ASHR1]](s16), [[ASHR2]](s16), [[ASHR3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_ashr_v4s16_v4s16 ; GFX9-LABEL: name: test_ashr_v4s16_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[ASHR:%[0-9]+]]:_(<2 x s16>) = G_ASHR [[UV]], [[UV2]](<2 x s16>) ; GFX9: [[ASHR:%[0-9]+]]:_(<2 x s16>) = G_ASHR [[UV]], [[UV2]](<2 x s16>)
▲ Show 20 Lines • Show All 812 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-fabs.mir

Show First 20 Lines • Show All 288 Lines • ▼ Show 20 Lines
--- ---
name: test_fabs_v4s16 name: test_fabs_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; SI-LABEL: name: test_fabs_v4s16 ; SI-LABEL: name: test_fabs_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]] ; SI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]] ; SI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]] ; SI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FABS3:%[0-9]+]]:_(s32) = G_FABS [[FPEXT3]] ; SI: [[FABS3:%[0-9]+]]:_(s32) = G_FABS [[FPEXT3]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS3]](s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_fabs_v4s16 ; VI-LABEL: name: test_fabs_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[FABS:%[0-9]+]]:_(s16) = G_FABS [[UV]] ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[FABS1:%[0-9]+]]:_(s16) = G_FABS [[UV1]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[FABS2:%[0-9]+]]:_(s16) = G_FABS [[UV2]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[FABS3:%[0-9]+]]:_(s16) = G_FABS [[UV3]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[FABS:%[0-9]+]]:_(s16) = G_FABS [[TRUNC]]
; VI: [[FABS1:%[0-9]+]]:_(s16) = G_FABS [[TRUNC1]]
; VI: [[FABS2:%[0-9]+]]:_(s16) = G_FABS [[TRUNC2]]
; VI: [[FABS3:%[0-9]+]]:_(s16) = G_FABS [[TRUNC3]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FABS]](s16), [[FABS1]](s16), [[FABS2]](s16), [[FABS3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FABS]](s16), [[FABS1]](s16), [[FABS2]](s16), [[FABS3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_fabs_v4s16 ; GFX9-LABEL: name: test_fabs_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[FABS:%[0-9]+]]:_(<2 x s16>) = G_FABS [[UV]] ; GFX9: [[FABS:%[0-9]+]]:_(<2 x s16>) = G_FABS [[UV]]
; GFX9: [[FABS1:%[0-9]+]]:_(<2 x s16>) = G_FABS [[UV1]] ; GFX9: [[FABS1:%[0-9]+]]:_(<2 x s16>) = G_FABS [[UV1]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FABS]](<2 x s16>), [[FABS1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FABS]](<2 x s16>), [[FABS1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = G_FABS %0 %1:_(<4 x s16>) = G_FABS %0
$vgpr0_vgpr1 = COPY %1 $vgpr0_vgpr1 = COPY %1
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-fadd.mir

Show First 20 Lines • Show All 400 Lines • ▼ Show 20 Lines
name: test_fadd_v4s16 name: test_fadd_v4s16
body: | body: |
bb.0.entry: bb.0.entry:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_fadd_v4s16 ; SI-LABEL: name: test_fadd_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]] ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]] ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16) ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]] ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16) ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
; SI: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT6]], [[FPEXT7]] ; SI: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT6]], [[FPEXT7]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_fadd_v4s16 ; VI-LABEL: name: test_fadd_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV]], [[UV4]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV1]], [[UV5]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[UV6]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[FADD3:%[0-9]+]]:_(s16) = G_FADD [[UV3]], [[UV7]] ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[TRUNC]], [[TRUNC4]]
; VI: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[TRUNC1]], [[TRUNC5]]
; VI: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[TRUNC2]], [[TRUNC6]]
; VI: [[FADD3:%[0-9]+]]:_(s16) = G_FADD [[TRUNC3]], [[TRUNC7]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FADD]](s16), [[FADD1]](s16), [[FADD2]](s16), [[FADD3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FADD]](s16), [[FADD1]](s16), [[FADD2]](s16), [[FADD3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_fadd_v4s16 ; GFX9-LABEL: name: test_fadd_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[FADD:%[0-9]+]]:_(<2 x s16>) = G_FADD [[UV]], [[UV2]] ; GFX9: [[FADD:%[0-9]+]]:_(<2 x s16>) = G_FADD [[UV]], [[UV2]]
; GFX9: [[FADD1:%[0-9]+]]:_(<2 x s16>) = G_FADD [[UV1]], [[UV3]] ; GFX9: [[FADD1:%[0-9]+]]:_(<2 x s16>) = G_FADD [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FADD]](<2 x s16>), [[FADD1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FADD]](<2 x s16>), [[FADD1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3 %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_FADD %0, %1 %2:_(<4 x s16>) = G_FADD %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-fcanonicalize.mir

Show First 20 Lines • Show All 273 Lines • ▼ Show 20 Lines
--- ---
name: test_fcanonicalize_v4s16 name: test_fcanonicalize_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; SI-LABEL: name: test_fcanonicalize_v4s16 ; SI-LABEL: name: test_fcanonicalize_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FCANONICALIZE:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT]] ; SI: [[FCANONICALIZE:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE]](s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FCANONICALIZE1:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT1]] ; SI: [[FCANONICALIZE1:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT1]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE1]](s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE1]](s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FCANONICALIZE2:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT2]] ; SI: [[FCANONICALIZE2:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT2]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE2]](s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE2]](s32)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FCANONICALIZE3:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT3]] ; SI: [[FCANONICALIZE3:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT3]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE3]](s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_fcanonicalize_v4s16 ; VI-LABEL: name: test_fcanonicalize_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV]] ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV1]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV3]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC]]
; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC1]]
; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC2]]
; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC3]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FCANONICALIZE]](s16), [[FCANONICALIZE1]](s16), [[FCANONICALIZE2]](s16), [[FCANONICALIZE3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FCANONICALIZE]](s16), [[FCANONICALIZE1]](s16), [[FCANONICALIZE2]](s16), [[FCANONICALIZE3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_fcanonicalize_v4s16 ; GFX9-LABEL: name: test_fcanonicalize_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[FCANONICALIZE:%[0-9]+]]:_(<2 x s16>) = G_FCANONICALIZE [[UV]] ; GFX9: [[FCANONICALIZE:%[0-9]+]]:_(<2 x s16>) = G_FCANONICALIZE [[UV]]
; GFX9: [[FCANONICALIZE1:%[0-9]+]]:_(<2 x s16>) = G_FCANONICALIZE [[UV1]] ; GFX9: [[FCANONICALIZE1:%[0-9]+]]:_(<2 x s16>) = G_FCANONICALIZE [[UV1]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FCANONICALIZE]](<2 x s16>), [[FCANONICALIZE1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FCANONICALIZE]](<2 x s16>), [[FCANONICALIZE1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = G_FCANONICALIZE %0 %1:_(<4 x s16>) = G_FCANONICALIZE %0
$vgpr0_vgpr1 = COPY %1 $vgpr0_vgpr1 = COPY %1
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir

Show First 20 Lines • Show All 482 Lines • ▼ Show 20 Lines
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; SI-LABEL: name: test_fma_v4s16 ; SI-LABEL: name: test_fma_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5 ; SI: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16), [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16) ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; SI: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
; SI: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
; SI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
; SI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
; SI: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
; SI: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<2 x s16>)
; SI: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
; SI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
; SI: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR5]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC8]](s16)
; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]] ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16) ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC9]](s16)
; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]] ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16) ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16) ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC10]](s16)
; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]] ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
; SI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16) ; SI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
; SI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[UV11]](s16) ; SI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC11]](s16)
; SI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[FPEXT9]], [[FPEXT10]], [[FPEXT11]] ; SI: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[FPEXT9]], [[FPEXT10]], [[FPEXT11]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA3]](s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_fma_v4s16 ; VI-LABEL: name: test_fma_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5 ; VI: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16), [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>) ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[UV]], [[UV4]], [[UV8]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[UV1]], [[UV5]], [[UV9]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[FMA2:%[0-9]+]]:_(s16) = G_FMA [[UV2]], [[UV6]], [[UV10]] ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[FMA3:%[0-9]+]]:_(s16) = G_FMA [[UV3]], [[UV7]], [[UV11]] ; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
; VI: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
; VI: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
; VI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
; VI: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
; VI: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<2 x s16>)
; VI: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
; VI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
; VI: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR5]](s32)
; VI: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC4]], [[TRUNC8]]
; VI: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[TRUNC1]], [[TRUNC5]], [[TRUNC9]]
; VI: [[FMA2:%[0-9]+]]:_(s16) = G_FMA [[TRUNC2]], [[TRUNC6]], [[TRUNC10]]
; VI: [[FMA3:%[0-9]+]]:_(s16) = G_FMA [[TRUNC3]], [[TRUNC7]], [[TRUNC11]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FMA]](s16), [[FMA1]](s16), [[FMA2]](s16), [[FMA3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FMA]](s16), [[FMA1]](s16), [[FMA2]](s16), [[FMA3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_fma_v4s16 ; GFX9-LABEL: name: test_fma_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX9: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5 ; GFX9: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; 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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
Show All 11 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-fmaxnum.mir

Show First 20 Lines • Show All 482 Lines • ▼ Show 20 Lines
name: test_fminnum_v4s16 name: test_fminnum_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_fminnum_v4s16 ; SI-LABEL: name: test_fminnum_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC %26(s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC %26(s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC %23(s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC %23(s32)
; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16) ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC %20(s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC %20(s32)
; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16) ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC %17(s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC %17(s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; SI: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT6]], [[FPEXT7]] ; SI: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT6]], [[FPEXT7]]
; SI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT4]], [[FPEXT5]] ; SI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT4]], [[FPEXT5]]
; SI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT2]], [[FPEXT3]] ; SI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT2]], [[FPEXT3]]
; SI: [[FMINNUM_IEEE3:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT]], [[FPEXT1]] ; SI: [[FMINNUM_IEEE3:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT]], [[FPEXT1]]
; VI-LABEL: name: test_fminnum_v4s16 ; VI-LABEL: name: test_fminnum_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR %11(s16), %12(s16), %13(s16), %14(s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR %11(s16), %12(s16), %13(s16), %14(s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV3]] ; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC3]]
; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV7]] ; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC7]]
; VI: [[FMINNUM_IEEE:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE]], [[FCANONICALIZE1]] ; VI: [[FMINNUM_IEEE:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE]], [[FCANONICALIZE1]]
; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] ; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC2]]
; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV6]] ; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC6]]
; VI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE2]], [[FCANONICALIZE3]] ; VI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE2]], [[FCANONICALIZE3]]
; VI: [[FCANONICALIZE4:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV1]] ; VI: [[FCANONICALIZE4:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC1]]
; VI: [[FCANONICALIZE5:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV5]] ; VI: [[FCANONICALIZE5:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC5]]
; VI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE4]], [[FCANONICALIZE5]] ; VI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE4]], [[FCANONICALIZE5]]
; VI: [[FCANONICALIZE6:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV]] ; VI: [[FCANONICALIZE6:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC]]
; VI: [[FCANONICALIZE7:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV4]] ; VI: [[FCANONICALIZE7:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC4]]
; VI: [[FMINNUM_IEEE3:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE6]], [[FCANONICALIZE7]] ; VI: [[FMINNUM_IEEE3:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE6]], [[FCANONICALIZE7]]
; GFX9-LABEL: name: test_fminnum_v4s16 ; GFX9-LABEL: name: test_fminnum_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS %7(<2 x s16>), %8(<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS %7(<2 x s16>), %8(<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
Show All 11 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-fminnum.mir

Show First 20 Lines • Show All 482 Lines • ▼ Show 20 Lines
name: test_fminnum_v4s16 name: test_fminnum_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_fminnum_v4s16 ; SI-LABEL: name: test_fminnum_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC %26(s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC %26(s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC %23(s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC %23(s32)
; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16) ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC %20(s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC %20(s32)
; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16) ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC %17(s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC %17(s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; SI: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT6]], [[FPEXT7]] ; SI: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT6]], [[FPEXT7]]
; SI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT4]], [[FPEXT5]] ; SI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT4]], [[FPEXT5]]
; SI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT2]], [[FPEXT3]] ; SI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT2]], [[FPEXT3]]
; SI: [[FMINNUM_IEEE3:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT]], [[FPEXT1]] ; SI: [[FMINNUM_IEEE3:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT]], [[FPEXT1]]
; VI-LABEL: name: test_fminnum_v4s16 ; VI-LABEL: name: test_fminnum_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR %11(s16), %12(s16), %13(s16), %14(s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR %11(s16), %12(s16), %13(s16), %14(s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV3]] ; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC3]]
; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV7]] ; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC7]]
; VI: [[FMINNUM_IEEE:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE]], [[FCANONICALIZE1]] ; VI: [[FMINNUM_IEEE:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE]], [[FCANONICALIZE1]]
; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] ; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC2]]
; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV6]] ; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC6]]
; VI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE2]], [[FCANONICALIZE3]] ; VI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE2]], [[FCANONICALIZE3]]
; VI: [[FCANONICALIZE4:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV1]] ; VI: [[FCANONICALIZE4:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC1]]
; VI: [[FCANONICALIZE5:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV5]] ; VI: [[FCANONICALIZE5:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC5]]
; VI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE4]], [[FCANONICALIZE5]] ; VI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE4]], [[FCANONICALIZE5]]
; VI: [[FCANONICALIZE6:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV]] ; VI: [[FCANONICALIZE6:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC]]
; VI: [[FCANONICALIZE7:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV4]] ; VI: [[FCANONICALIZE7:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[TRUNC4]]
; VI: [[FMINNUM_IEEE3:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE6]], [[FCANONICALIZE7]] ; VI: [[FMINNUM_IEEE3:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE6]], [[FCANONICALIZE7]]
; GFX9-LABEL: name: test_fminnum_v4s16 ; GFX9-LABEL: name: test_fminnum_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS %7(<2 x s16>), %8(<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS %7(<2 x s16>), %8(<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
Show All 11 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-fmul.mir

Show First 20 Lines • Show All 399 Lines • ▼ Show 20 Lines
name: test_fmul_v4s16 name: test_fmul_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_fmul_v4s16 ; SI-LABEL: name: test_fmul_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
; SI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]] ; SI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
; SI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]] ; SI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32)
; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16) ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
; SI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]] ; SI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32)
; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16) ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
; SI: [[FMUL3:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT6]], [[FPEXT7]] ; SI: [[FMUL3:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT6]], [[FPEXT7]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL3]](s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_fmul_v4s16 ; VI-LABEL: name: test_fmul_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[UV]], [[UV4]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[FMUL1:%[0-9]+]]:_(s16) = G_FMUL [[UV1]], [[UV5]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[FMUL2:%[0-9]+]]:_(s16) = G_FMUL [[UV2]], [[UV6]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[FMUL3:%[0-9]+]]:_(s16) = G_FMUL [[UV3]], [[UV7]] ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC]], [[TRUNC4]]
; VI: [[FMUL1:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC1]], [[TRUNC5]]
; VI: [[FMUL2:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC2]], [[TRUNC6]]
; VI: [[FMUL3:%[0-9]+]]:_(s16) = G_FMUL [[TRUNC3]], [[TRUNC7]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FMUL]](s16), [[FMUL1]](s16), [[FMUL2]](s16), [[FMUL3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FMUL]](s16), [[FMUL1]](s16), [[FMUL2]](s16), [[FMUL3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_fmul_v4s16 ; GFX9-LABEL: name: test_fmul_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[FMUL:%[0-9]+]]:_(<2 x s16>) = G_FMUL [[UV]], [[UV2]] ; GFX9: [[FMUL:%[0-9]+]]:_(<2 x s16>) = G_FMUL [[UV]], [[UV2]]
; GFX9: [[FMUL1:%[0-9]+]]:_(<2 x s16>) = G_FMUL [[UV1]], [[UV3]] ; GFX9: [[FMUL1:%[0-9]+]]:_(<2 x s16>) = G_FMUL [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FMUL]](<2 x s16>), [[FMUL1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FMUL]](<2 x s16>), [[FMUL1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3 %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_FMUL %0, %1 %2:_(<4 x s16>) = G_FMUL %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-fneg.mir

Show First 20 Lines • Show All 286 Lines • ▼ Show 20 Lines
--- ---
name: test_fneg_v4s16 name: test_fneg_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; SI-LABEL: name: test_fneg_v4s16 ; SI-LABEL: name: test_fneg_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]] ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]] ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]] ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]] ; SI: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_fneg_v4s16 ; VI-LABEL: name: test_fneg_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV]] ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV1]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV2]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[FNEG3:%[0-9]+]]:_(s16) = G_FNEG [[UV3]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC]]
; VI: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC1]]
; VI: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC2]]
; VI: [[FNEG3:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC3]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FNEG]](s16), [[FNEG1]](s16), [[FNEG2]](s16), [[FNEG3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FNEG]](s16), [[FNEG1]](s16), [[FNEG2]](s16), [[FNEG3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_fneg_v4s16 ; GFX9-LABEL: name: test_fneg_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[FNEG:%[0-9]+]]:_(<2 x s16>) = G_FNEG [[UV]] ; GFX9: [[FNEG:%[0-9]+]]:_(<2 x s16>) = G_FNEG [[UV]]
; GFX9: [[FNEG1:%[0-9]+]]:_(<2 x s16>) = G_FNEG [[UV1]] ; GFX9: [[FNEG1:%[0-9]+]]:_(<2 x s16>) = G_FNEG [[UV1]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FNEG]](<2 x s16>), [[FNEG1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FNEG]](<2 x s16>), [[FNEG1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = G_FNEG %0 %1:_(<4 x s16>) = G_FNEG %0
$vgpr0_vgpr1 = COPY %1 $vgpr0_vgpr1 = COPY %1
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-fpext.mir

Show First 20 Lines • Show All 84 Lines • ▼ Show 20 Lines
--- ---
name: test_fpext_v4f16_to_v4f32 name: test_fpext_v4f16_to_v4f32
body: | body: |
bb.0: bb.0:
liveins: $vgpr0 liveins: $vgpr0
; CHECK-LABEL: name: test_fpext_v4f16_to_v4f32 ; CHECK-LABEL: name: test_fpext_v4f16_to_v4f32
; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[DEF]](<4 x s16>) ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF]](<4 x s16>)
; CHECK: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; CHECK: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; CHECK: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; CHECK: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; CHECK: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; CHECK: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; CHECK: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; CHECK: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FPEXT]](s32), [[FPEXT1]](s32), [[FPEXT2]](s32), [[FPEXT3]](s32) ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FPEXT]](s32), [[FPEXT1]](s32), [[FPEXT2]](s32), [[FPEXT3]](s32)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>) ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
%0:_(<4 x s16>) = G_IMPLICIT_DEF %0:_(<4 x s16>) = G_IMPLICIT_DEF
%1:_(<4 x s32>) = G_FPEXT %0 %1:_(<4 x s32>) = G_FPEXT %0
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1 $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
... ...
--- ---
▲ Show 20 Lines • Show All 71 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-fsqrt.mir

Show First 20 Lines • Show All 286 Lines • ▼ Show 20 Lines
--- ---
name: test_fsqrt_v4s16 name: test_fsqrt_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; SI-LABEL: name: test_fsqrt_v4s16 ; SI-LABEL: name: test_fsqrt_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]] ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]] ; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT2]] ; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT2]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT2]](s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT2]](s32)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FSQRT3:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT3]] ; SI: [[FSQRT3:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT3]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT3]](s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC]](s16), [[FPTRUNC1]](s16), [[FPTRUNC2]](s16), [[FPTRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_fsqrt_v4s16 ; VI-LABEL: name: test_fsqrt_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[UV]] ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[UV1]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[UV2]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[FSQRT3:%[0-9]+]]:_(s16) = G_FSQRT [[UV3]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
; VI: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
; VI: [[FSQRT3:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC3]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16), [[FSQRT2]](s16), [[FSQRT3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16), [[FSQRT2]](s16), [[FSQRT3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_fsqrt_v4s16 ; GFX9-LABEL: name: test_fsqrt_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[UV]] ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[UV1]] ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[UV2]] ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[FSQRT3:%[0-9]+]]:_(s16) = G_FSQRT [[UV3]] ; 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 [[UV1]](<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: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
; GFX9: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
; GFX9: [[FSQRT3:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC3]]
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16), [[FSQRT2]](s16), [[FSQRT3]](s16) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FSQRT]](s16), [[FSQRT1]](s16), [[FSQRT2]](s16), [[FSQRT3]](s16)
; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = G_FSQRT %0 %1:_(<4 x s16>) = G_FSQRT %0
$vgpr0_vgpr1 = COPY %1 $vgpr0_vgpr1 = COPY %1
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-fsub.mir

Show First 20 Lines • Show All 453 Lines • ▼ Show 20 Lines
name: test_fsub_v4s16 name: test_fsub_v4s16
body: | body: |
bb.0.entry: bb.0.entry:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_fsub_v4s16 ; SI-LABEL: name: test_fsub_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]] ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32)
; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16)
; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]] ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32)
; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]] ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32)
; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16) ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16)
; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]] ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32) ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32)
; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16) ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]] ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]]
; SI: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32) ; SI: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32)
; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16) ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16)
; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]] ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]]
; SI: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32) ; SI: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32)
; SI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16) ; SI: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
; SI: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT9]] ; SI: [[FNEG3:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT9]]
; SI: [[FPTRUNC6:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32) ; SI: [[FPTRUNC6:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG3]](s32)
; SI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
; SI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC6]](s16) ; SI: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC6]](s16)
; SI: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT10]], [[FPEXT11]] ; SI: [[FADD3:%[0-9]+]]:_(s32) = G_FADD [[FPEXT10]], [[FPEXT11]]
; SI: [[FPTRUNC7:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32) ; SI: [[FPTRUNC7:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC1]](s16), [[FPTRUNC3]](s16), [[FPTRUNC5]](s16), [[FPTRUNC7]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FPTRUNC1]](s16), [[FPTRUNC3]](s16), [[FPTRUNC5]](s16), [[FPTRUNC7]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_fsub_v4s16 ; VI-LABEL: name: test_fsub_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV4]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV]], [[FNEG]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV5]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV1]], [[FNEG1]] ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV6]] ; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[FNEG2]] ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[FNEG3:%[0-9]+]]:_(s16) = G_FNEG [[UV7]] ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[FADD3:%[0-9]+]]:_(s16) = G_FADD [[UV3]], [[FNEG3]] ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC4]]
; VI: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[TRUNC]], [[FNEG]]
; VI: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC5]]
; VI: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[TRUNC1]], [[FNEG1]]
; VI: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC6]]
; VI: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[TRUNC2]], [[FNEG2]]
; VI: [[FNEG3:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC7]]
; VI: [[FADD3:%[0-9]+]]:_(s16) = G_FADD [[TRUNC3]], [[FNEG3]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FADD]](s16), [[FADD1]](s16), [[FADD2]](s16), [[FADD3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FADD]](s16), [[FADD1]](s16), [[FADD2]](s16), [[FADD3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_fsub_v4s16 ; GFX9-LABEL: name: test_fsub_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX9: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV4]] ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV]], [[FNEG]] ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV5]] ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV1]], [[FNEG1]] ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV6]] ; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX9: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[FNEG2]] ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[FNEG3:%[0-9]+]]:_(s16) = G_FNEG [[UV7]] ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: [[FADD3:%[0-9]+]]:_(s16) = G_FADD [[UV3]], [[FNEG3]] ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; GFX9: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; GFX9: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; GFX9: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; GFX9: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; GFX9: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; GFX9: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; GFX9: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; GFX9: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC4]]
; GFX9: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[TRUNC]], [[FNEG]]
; GFX9: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC5]]
; GFX9: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[TRUNC1]], [[FNEG1]]
; GFX9: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC6]]
; GFX9: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[TRUNC2]], [[FNEG2]]
; GFX9: [[FNEG3:%[0-9]+]]:_(s16) = G_FNEG [[TRUNC7]]
; GFX9: [[FADD3:%[0-9]+]]:_(s16) = G_FADD [[TRUNC3]], [[FNEG3]]
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FADD]](s16), [[FADD1]](s16), [[FADD2]](s16), [[FADD3]](s16) ; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[FADD]](s16), [[FADD1]](s16), [[FADD2]](s16), [[FADD3]](s16)
; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3 %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_FSUB %0, %1 %2:_(<4 x s16>) = G_FSUB %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-lshr.mir

Show First 20 Lines • Show All 674 Lines • ▼ Show 20 Lines
name: test_lshr_v4s16_v4s16 name: test_lshr_v4s16_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_lshr_v4s16_v4s16 ; SI-LABEL: name: test_lshr_v4s16_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT1]], [[ZEXT]](s32) ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) ; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16) ; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT3]], [[ZEXT2]](s32) ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32) ; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV6]](s16) ; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT5]], [[ZEXT4]](s32) ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32) ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
; SI: [[ZEXT6:%[0-9]+]]:_(s32) = G_ZEXT [[UV7]](s16) ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; SI: [[ZEXT7:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT7]], [[ZEXT6]](s32) ; SI: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[AND1]], [[AND]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
; SI: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[AND3]], [[AND2]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR5]](s32)
; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
; SI: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[AND5]], [[AND4]](s32)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR6]](s32)
; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C1]]
; SI: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[AND7]], [[AND6]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR7]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_lshr_v4s16_v4s16 ; VI-LABEL: name: test_lshr_v4s16_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[LSHR:%[0-9]+]]:_(s16) = G_LSHR [[UV]], [[UV4]](s16) ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s16) = G_LSHR [[UV1]], [[UV5]](s16) ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[LSHR2:%[0-9]+]]:_(s16) = G_LSHR [[UV2]], [[UV6]](s16) ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s16) = G_LSHR [[UV3]], [[UV7]](s16) ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[LSHR]](s16), [[LSHR1]](s16), [[LSHR2]](s16), [[LSHR3]](s16) ; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[LSHR4:%[0-9]+]]:_(s16) = G_LSHR [[TRUNC]], [[TRUNC4]](s16)
; VI: [[LSHR5:%[0-9]+]]:_(s16) = G_LSHR [[TRUNC1]], [[TRUNC5]](s16)
; VI: [[LSHR6:%[0-9]+]]:_(s16) = G_LSHR [[TRUNC2]], [[TRUNC6]](s16)
; VI: [[LSHR7:%[0-9]+]]:_(s16) = G_LSHR [[TRUNC3]], [[TRUNC7]](s16)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[LSHR4]](s16), [[LSHR5]](s16), [[LSHR6]](s16), [[LSHR7]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_lshr_v4s16_v4s16 ; GFX9-LABEL: name: test_lshr_v4s16_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[LSHR:%[0-9]+]]:_(<2 x s16>) = G_LSHR [[UV]], [[UV2]](<2 x s16>) ; GFX9: [[LSHR:%[0-9]+]]:_(<2 x s16>) = G_LSHR [[UV]], [[UV2]](<2 x s16>)
; GFX9: [[LSHR1:%[0-9]+]]:_(<2 x s16>) = G_LSHR [[UV1]], [[UV3]](<2 x s16>) ; GFX9: [[LSHR1:%[0-9]+]]:_(<2 x s16>) = G_LSHR [[UV1]], [[UV3]](<2 x s16>)
▲ Show 20 Lines • Show All 781 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-phi.mir

Show First 20 Lines • Show All 190 Lines • ▼ Show 20 Linesbody: |
; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr2 ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr2
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]] ; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[COPY1]](s32), [[C]]
; CHECK: G_BRCOND [[ICMP]](s1), %bb.1 ; CHECK: G_BRCOND [[ICMP]](s1), %bb.1
; CHECK: G_BR %bb.2 ; CHECK: G_BR %bb.2
; CHECK: bb.1: ; CHECK: bb.1:
; CHECK: successors: %bb.2(0x80000000) ; CHECK: successors: %bb.2(0x80000000)
; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; CHECK: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16) ; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s16) ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C1]](s32)
; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[ANYEXT]], [[ANYEXT1]] ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C1]](s32)
; CHECK: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; CHECK: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C1]](s32)
; CHECK: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; CHECK: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C1]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[COPY2]], [[COPY3]]
; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[ADD]](s32) ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[ADD]](s32)
; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16) ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s16) ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ANYEXT2]], [[ANYEXT3]] ; CHECK: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[COPY4]], [[COPY5]]
; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[ADD1]](s32) ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[ADD1]](s32)
; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) ; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV6]](s16) ; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; CHECK: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[ANYEXT4]], [[ANYEXT5]] ; CHECK: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[COPY6]], [[COPY7]]
; CHECK: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[ADD2]](s32) ; CHECK: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[ADD2]](s32)
; CHECK: [[ANYEXT6:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16) ; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[ANYEXT7:%[0-9]+]]:_(s32) = G_ANYEXT [[UV7]](s16) ; CHECK: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; CHECK: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[ANYEXT6]], [[ANYEXT7]] ; CHECK: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[COPY8]], [[COPY9]]
; CHECK: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[ADD3]](s32) ; CHECK: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[ADD3]](s32)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
; CHECK: G_BR %bb.2 ; CHECK: G_BR %bb.2
; CHECK: bb.2: ; CHECK: bb.2:
; CHECK: [[PHI:%[0-9]+]]:_(<4 x s16>) = G_PHI [[COPY]](<4 x s16>), %bb.0, [[BUILD_VECTOR]](<4 x s16>), %bb.1 ; CHECK: [[PHI:%[0-9]+]]:_(<4 x s16>) = G_PHI [[COPY]](<4 x s16>), %bb.0, [[BUILD_VECTOR]](<4 x s16>), %bb.1
; CHECK: $vgpr0_vgpr1 = COPY [[PHI]](<4 x s16>) ; CHECK: $vgpr0_vgpr1 = COPY [[PHI]](<4 x s16>)
; CHECK: S_SETPC_B64 undef $sgpr30_sgpr31 ; CHECK: S_SETPC_B64 undef $sgpr30_sgpr31
bb.0: bb.0:
▲ Show 20 Lines • Show All 1,166 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-sext.mir

Show First 20 Lines • Show All 122 Lines • ▼ Show 20 Lines
--- ---
name: test_sext_v4s16_to_v4s32 name: test_sext_v4s16_to_v4s32
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_sext_v4s16_to_v4s32 ; CHECK-LABEL: name: test_sext_v4s16_to_v4s32
; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; CHECK: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; CHECK: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[SEXT]](s32), [[SEXT1]](s32), [[SEXT2]](s32), [[SEXT3]](s32) ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY1]], [[C]](s32)
; CHECK: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY2]], [[C]](s32)
; CHECK: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY3]], [[C]](s32)
; CHECK: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[COPY4]], [[C]](s32)
; CHECK: [[ASHR3:%[0-9]+]]:_(s32) = G_ASHR [[SHL3]], [[C]](s32)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[ASHR]](s32), [[ASHR1]](s32), [[ASHR2]](s32), [[ASHR3]](s32)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>) ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s32>) = G_SEXT %0 %1:_(<4 x s32>) = G_SEXT %0
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1 $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
... ...
--- ---
name: test_sext_v2s32_to_v2s64 name: test_sext_v2s32_to_v2s64
▲ Show 20 Lines • Show All 55 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-shl.mir

Show First 20 Lines • Show All 662 Lines • ▼ Show 20 Lines
name: test_shl_v4s16_v4s16 name: test_shl_v4s16_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_shl_v4s16_v4s16 ; SI-LABEL: name: test_shl_v4s16_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ANYEXT]], [[ZEXT]](s32) ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY3]], [[AND]](s32)
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; SI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16) ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ANYEXT1]], [[ZEXT1]](s32) ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY5]], [[AND1]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32) ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV6]](s16) ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; SI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ANYEXT2]], [[ZEXT2]](s32) ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY7]], [[AND2]](s32)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32) ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SHL2]](s32)
; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV7]](s16) ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; SI: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16) ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[ANYEXT3]], [[ZEXT3]](s32) ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[COPY9]], [[AND3]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32) ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_shl_v4s16_v4s16 ; VI-LABEL: name: test_shl_v4s16_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[UV]], [[UV4]](s16) ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[UV1]], [[UV5]](s16) ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[UV2]], [[UV6]](s16) ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[UV3]], [[UV7]](s16) ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[TRUNC4]](s16)
; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[TRUNC5]](s16)
; VI: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC6]](s16)
; VI: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[TRUNC3]], [[TRUNC7]](s16)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[SHL]](s16), [[SHL1]](s16), [[SHL2]](s16), [[SHL3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[SHL]](s16), [[SHL1]](s16), [[SHL2]](s16), [[SHL3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_shl_v4s16_v4s16 ; GFX9-LABEL: name: test_shl_v4s16_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[SHL:%[0-9]+]]:_(<2 x s16>) = G_SHL [[UV]], [[UV2]](<2 x s16>) ; GFX9: [[SHL:%[0-9]+]]:_(<2 x s16>) = G_SHL [[UV]], [[UV2]](<2 x s16>)
▲ Show 20 Lines • Show All 864 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir

Show First 20 Lines • Show All 393 Lines • ▼ Show 20 Lines
name: test_smax_v4s16 name: test_smax_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_smax_v4s16 ; SI-LABEL: name: test_smax_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT]], [[SEXT1]] ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY2]], [[C]](s32)
; SI: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32)
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY3]], [[C]](s32)
; SI: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C]](s32)
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[ASHR]], [[ASHR1]]
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SMAX]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SMAX]](s32)
; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; SI: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV5]](s16) ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY4]], [[C]](s32)
; SI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[SEXT2]], [[SEXT3]] ; SI: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C]](s32)
; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[COPY5]], [[C]](s32)
; SI: [[ASHR3:%[0-9]+]]:_(s32) = G_ASHR [[SHL3]], [[C]](s32)
; SI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[ASHR2]], [[ASHR3]]
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SMAX1]](s32) ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SMAX1]](s32)
; SI: [[SEXT4:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; SI: [[SEXT5:%[0-9]+]]:_(s32) = G_SEXT [[UV6]](s16) ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[COPY6]], [[C]](s32)
; SI: [[SMAX2:%[0-9]+]]:_(s32) = G_SMAX [[SEXT4]], [[SEXT5]] ; SI: [[ASHR4:%[0-9]+]]:_(s32) = G_ASHR [[SHL4]], [[C]](s32)
; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[COPY7]], [[C]](s32)
; SI: [[ASHR5:%[0-9]+]]:_(s32) = G_ASHR [[SHL5]], [[C]](s32)
; SI: [[SMAX2:%[0-9]+]]:_(s32) = G_SMAX [[ASHR4]], [[ASHR5]]
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SMAX2]](s32) ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SMAX2]](s32)
; SI: [[SEXT6:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; SI: [[SEXT7:%[0-9]+]]:_(s32) = G_SEXT [[UV7]](s16) ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[COPY8]], [[C]](s32)
; SI: [[SMAX3:%[0-9]+]]:_(s32) = G_SMAX [[SEXT6]], [[SEXT7]] ; SI: [[ASHR6:%[0-9]+]]:_(s32) = G_ASHR [[SHL6]], [[C]](s32)
; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[COPY9]], [[C]](s32)
; SI: [[ASHR7:%[0-9]+]]:_(s32) = G_ASHR [[SHL7]], [[C]](s32)
; SI: [[SMAX3:%[0-9]+]]:_(s32) = G_SMAX [[ASHR6]], [[ASHR7]]
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SMAX3]](s32) ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SMAX3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_smax_v4s16 ; VI-LABEL: name: test_smax_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[UV]], [[UV4]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[SMAX1:%[0-9]+]]:_(s16) = G_SMAX [[UV1]], [[UV5]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[SMAX2:%[0-9]+]]:_(s16) = G_SMAX [[UV2]], [[UV6]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[SMAX3:%[0-9]+]]:_(s16) = G_SMAX [[UV3]], [[UV7]] ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC]], [[TRUNC4]]
; VI: [[SMAX1:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC1]], [[TRUNC5]]
; VI: [[SMAX2:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC2]], [[TRUNC6]]
; VI: [[SMAX3:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC3]], [[TRUNC7]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[SMAX]](s16), [[SMAX1]](s16), [[SMAX2]](s16), [[SMAX3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[SMAX]](s16), [[SMAX1]](s16), [[SMAX2]](s16), [[SMAX3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_smax_v4s16 ; GFX9-LABEL: name: test_smax_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[SMAX:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[UV]], [[UV2]] ; GFX9: [[SMAX:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[UV]], [[UV2]]
; GFX9: [[SMAX1:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[UV1]], [[UV3]] ; GFX9: [[SMAX1:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[SMAX]](<2 x s16>), [[SMAX1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[SMAX]](<2 x s16>), [[SMAX1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3 %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_SMAX %0, %1 %2:_(<4 x s16>) = G_SMAX %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-smin.mir

Show First 20 Lines • Show All 393 Lines • ▼ Show 20 Lines
name: test_smin_v4s16 name: test_smin_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_smin_v4s16 ; SI-LABEL: name: test_smin_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[SMIN:%[0-9]+]]:_(s32) = G_SMIN [[SEXT]], [[SEXT1]] ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY2]], [[C]](s32)
; SI: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32)
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY3]], [[C]](s32)
; SI: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C]](s32)
; SI: [[SMIN:%[0-9]+]]:_(s32) = G_SMIN [[ASHR]], [[ASHR1]]
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SMIN]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SMIN]](s32)
; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; SI: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV5]](s16) ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY4]], [[C]](s32)
; SI: [[SMIN1:%[0-9]+]]:_(s32) = G_SMIN [[SEXT2]], [[SEXT3]] ; SI: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C]](s32)
; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; SI: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[COPY5]], [[C]](s32)
; SI: [[ASHR3:%[0-9]+]]:_(s32) = G_ASHR [[SHL3]], [[C]](s32)
; SI: [[SMIN1:%[0-9]+]]:_(s32) = G_SMIN [[ASHR2]], [[ASHR3]]
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SMIN1]](s32) ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SMIN1]](s32)
; SI: [[SEXT4:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; SI: [[SEXT5:%[0-9]+]]:_(s32) = G_SEXT [[UV6]](s16) ; SI: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[COPY6]], [[C]](s32)
; SI: [[SMIN2:%[0-9]+]]:_(s32) = G_SMIN [[SEXT4]], [[SEXT5]] ; SI: [[ASHR4:%[0-9]+]]:_(s32) = G_ASHR [[SHL4]], [[C]](s32)
; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; SI: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[COPY7]], [[C]](s32)
; SI: [[ASHR5:%[0-9]+]]:_(s32) = G_ASHR [[SHL5]], [[C]](s32)
; SI: [[SMIN2:%[0-9]+]]:_(s32) = G_SMIN [[ASHR4]], [[ASHR5]]
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SMIN2]](s32) ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SMIN2]](s32)
; SI: [[SEXT6:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; SI: [[SEXT7:%[0-9]+]]:_(s32) = G_SEXT [[UV7]](s16) ; SI: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[COPY8]], [[C]](s32)
; SI: [[SMIN3:%[0-9]+]]:_(s32) = G_SMIN [[SEXT6]], [[SEXT7]] ; SI: [[ASHR6:%[0-9]+]]:_(s32) = G_ASHR [[SHL6]], [[C]](s32)
; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; SI: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[COPY9]], [[C]](s32)
; SI: [[ASHR7:%[0-9]+]]:_(s32) = G_ASHR [[SHL7]], [[C]](s32)
; SI: [[SMIN3:%[0-9]+]]:_(s32) = G_SMIN [[ASHR6]], [[ASHR7]]
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SMIN3]](s32) ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SMIN3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_smin_v4s16 ; VI-LABEL: name: test_smin_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[SMIN:%[0-9]+]]:_(s16) = G_SMIN [[UV]], [[UV4]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[SMIN1:%[0-9]+]]:_(s16) = G_SMIN [[UV1]], [[UV5]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[SMIN2:%[0-9]+]]:_(s16) = G_SMIN [[UV2]], [[UV6]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[SMIN3:%[0-9]+]]:_(s16) = G_SMIN [[UV3]], [[UV7]] ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[SMIN:%[0-9]+]]:_(s16) = G_SMIN [[TRUNC]], [[TRUNC4]]
; VI: [[SMIN1:%[0-9]+]]:_(s16) = G_SMIN [[TRUNC1]], [[TRUNC5]]
; VI: [[SMIN2:%[0-9]+]]:_(s16) = G_SMIN [[TRUNC2]], [[TRUNC6]]
; VI: [[SMIN3:%[0-9]+]]:_(s16) = G_SMIN [[TRUNC3]], [[TRUNC7]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[SMIN]](s16), [[SMIN1]](s16), [[SMIN2]](s16), [[SMIN3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[SMIN]](s16), [[SMIN1]](s16), [[SMIN2]](s16), [[SMIN3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_smin_v4s16 ; GFX9-LABEL: name: test_smin_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[SMIN:%[0-9]+]]:_(<2 x s16>) = G_SMIN [[UV]], [[UV2]] ; GFX9: [[SMIN:%[0-9]+]]:_(<2 x s16>) = G_SMIN [[UV]], [[UV2]]
; GFX9: [[SMIN1:%[0-9]+]]:_(<2 x s16>) = G_SMIN [[UV1]], [[UV3]] ; GFX9: [[SMIN1:%[0-9]+]]:_(<2 x s16>) = G_SMIN [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[SMIN]](<2 x s16>), [[SMIN1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[SMIN]](<2 x s16>), [[SMIN1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3 %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_SMIN %0, %1 %2:_(<4 x s16>) = G_SMIN %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-umax.mir

Show First 20 Lines • Show All 376 Lines • ▼ Show 20 Lines
name: test_umax_v4s16 name: test_umax_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_umax_v4s16 ; SI-LABEL: name: test_umax_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[UMAX:%[0-9]+]]:_(s32) = G_UMAX [[ZEXT]], [[ZEXT1]] ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
; SI: [[UMAX:%[0-9]+]]:_(s32) = G_UMAX [[AND]], [[AND1]]
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UMAX]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UMAX]](s32)
; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16) ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
; SI: [[UMAX1:%[0-9]+]]:_(s32) = G_UMAX [[ZEXT2]], [[ZEXT3]] ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
; SI: [[UMAX1:%[0-9]+]]:_(s32) = G_UMAX [[AND2]], [[AND3]]
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UMAX1]](s32) ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UMAX1]](s32)
; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV6]](s16) ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
; SI: [[UMAX2:%[0-9]+]]:_(s32) = G_UMAX [[ZEXT4]], [[ZEXT5]] ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
; SI: [[UMAX2:%[0-9]+]]:_(s32) = G_UMAX [[AND4]], [[AND5]]
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UMAX2]](s32) ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UMAX2]](s32)
; SI: [[ZEXT6:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; SI: [[ZEXT7:%[0-9]+]]:_(s32) = G_ZEXT [[UV7]](s16) ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
; SI: [[UMAX3:%[0-9]+]]:_(s32) = G_UMAX [[ZEXT6]], [[ZEXT7]] ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C1]]
; SI: [[UMAX3:%[0-9]+]]:_(s32) = G_UMAX [[AND6]], [[AND7]]
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[UMAX3]](s32) ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[UMAX3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_umax_v4s16 ; VI-LABEL: name: test_umax_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[UMAX:%[0-9]+]]:_(s16) = G_UMAX [[UV]], [[UV4]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[UMAX1:%[0-9]+]]:_(s16) = G_UMAX [[UV1]], [[UV5]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[UMAX2:%[0-9]+]]:_(s16) = G_UMAX [[UV2]], [[UV6]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[UMAX3:%[0-9]+]]:_(s16) = G_UMAX [[UV3]], [[UV7]] ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[UMAX:%[0-9]+]]:_(s16) = G_UMAX [[TRUNC]], [[TRUNC4]]
; VI: [[UMAX1:%[0-9]+]]:_(s16) = G_UMAX [[TRUNC1]], [[TRUNC5]]
; VI: [[UMAX2:%[0-9]+]]:_(s16) = G_UMAX [[TRUNC2]], [[TRUNC6]]
; VI: [[UMAX3:%[0-9]+]]:_(s16) = G_UMAX [[TRUNC3]], [[TRUNC7]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[UMAX]](s16), [[UMAX1]](s16), [[UMAX2]](s16), [[UMAX3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[UMAX]](s16), [[UMAX1]](s16), [[UMAX2]](s16), [[UMAX3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_umax_v4s16 ; GFX9-LABEL: name: test_umax_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[UMAX:%[0-9]+]]:_(<2 x s16>) = G_UMAX [[UV]], [[UV2]] ; GFX9: [[UMAX:%[0-9]+]]:_(<2 x s16>) = G_UMAX [[UV]], [[UV2]]
; GFX9: [[UMAX1:%[0-9]+]]:_(<2 x s16>) = G_UMAX [[UV1]], [[UV3]] ; GFX9: [[UMAX1:%[0-9]+]]:_(<2 x s16>) = G_UMAX [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[UMAX]](<2 x s16>), [[UMAX1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[UMAX]](<2 x s16>), [[UMAX1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3 %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_UMAX %0, %1 %2:_(<4 x s16>) = G_UMAX %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-umin.mir

Show First 20 Lines • Show All 376 Lines • ▼ Show 20 Lines
name: test_umin_v4s16 name: test_umin_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_umin_v4s16 ; SI-LABEL: name: test_umin_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16) ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[UMIN:%[0-9]+]]:_(s32) = G_UMIN [[ZEXT]], [[ZEXT1]] ; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
; SI: [[UMIN:%[0-9]+]]:_(s32) = G_UMIN [[AND]], [[AND1]]
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UMIN]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UMIN]](s32)
; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16) ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
; SI: [[UMIN1:%[0-9]+]]:_(s32) = G_UMIN [[ZEXT2]], [[ZEXT3]] ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
; SI: [[UMIN1:%[0-9]+]]:_(s32) = G_UMIN [[AND2]], [[AND3]]
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UMIN1]](s32) ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UMIN1]](s32)
; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV6]](s16) ; SI: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
; SI: [[UMIN2:%[0-9]+]]:_(s32) = G_UMIN [[ZEXT4]], [[ZEXT5]] ; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
; SI: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
; SI: [[UMIN2:%[0-9]+]]:_(s32) = G_UMIN [[AND4]], [[AND5]]
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UMIN2]](s32) ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UMIN2]](s32)
; SI: [[ZEXT6:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) ; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; SI: [[ZEXT7:%[0-9]+]]:_(s32) = G_ZEXT [[UV7]](s16) ; SI: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C1]]
; SI: [[UMIN3:%[0-9]+]]:_(s32) = G_UMIN [[ZEXT6]], [[ZEXT7]] ; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; SI: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY9]], [[C1]]
; SI: [[UMIN3:%[0-9]+]]:_(s32) = G_UMIN [[AND6]], [[AND7]]
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[UMIN3]](s32) ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[UMIN3]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; VI-LABEL: name: test_umin_v4s16 ; VI-LABEL: name: test_umin_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>) ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[UMIN:%[0-9]+]]:_(s16) = G_UMIN [[UV]], [[UV4]] ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[UMIN1:%[0-9]+]]:_(s16) = G_UMIN [[UV1]], [[UV5]] ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[UMIN2:%[0-9]+]]:_(s16) = G_UMIN [[UV2]], [[UV6]] ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[UMIN3:%[0-9]+]]:_(s16) = G_UMIN [[UV3]], [[UV7]] ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[UMIN:%[0-9]+]]:_(s16) = G_UMIN [[TRUNC]], [[TRUNC4]]
; VI: [[UMIN1:%[0-9]+]]:_(s16) = G_UMIN [[TRUNC1]], [[TRUNC5]]
; VI: [[UMIN2:%[0-9]+]]:_(s16) = G_UMIN [[TRUNC2]], [[TRUNC6]]
; VI: [[UMIN3:%[0-9]+]]:_(s16) = G_UMIN [[TRUNC3]], [[TRUNC7]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[UMIN]](s16), [[UMIN1]](s16), [[UMIN2]](s16), [[UMIN3]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[UMIN]](s16), [[UMIN1]](s16), [[UMIN2]](s16), [[UMIN3]](s16)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>) ; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<4 x s16>)
; GFX9-LABEL: name: test_umin_v4s16 ; GFX9-LABEL: name: test_umin_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; 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: [[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: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[UMIN:%[0-9]+]]:_(<2 x s16>) = G_UMIN [[UV]], [[UV2]] ; GFX9: [[UMIN:%[0-9]+]]:_(<2 x s16>) = G_UMIN [[UV]], [[UV2]]
; GFX9: [[UMIN1:%[0-9]+]]:_(<2 x s16>) = G_UMIN [[UV1]], [[UV3]] ; GFX9: [[UMIN1:%[0-9]+]]:_(<2 x s16>) = G_UMIN [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[UMIN]](<2 x s16>), [[UMIN1]](<2 x s16>) ; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[UMIN]](<2 x s16>), [[UMIN1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>) ; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3 %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_UMIN %0, %1 %2:_(<4 x s16>) = G_UMIN %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...

test/CodeGen/AMDGPU/GlobalISel/legalize-unmerge-values.mir

Show First 20 Lines • Show All 80 Lines • ▼ Show 20 Lines
--- ---
name: test_unmerge_s16_v4s16 name: test_unmerge_s16_v4s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s16_v4s16 ; CHECK-LABEL: name: test_unmerge_s16_v4s16
; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16) ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16) ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16) ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; CHECK: $vgpr0 = COPY [[ANYEXT]](s32) ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: $vgpr1 = COPY [[ANYEXT1]](s32) ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: $vgpr2 = COPY [[ANYEXT2]](s32) ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: $vgpr3 = COPY [[ANYEXT3]](s32) ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(s16), %2:_(s16), %3:_(s16), %4:_(s16) = G_UNMERGE_VALUES %0 %1:_(s16), %2:_(s16), %3:_(s16), %4:_(s16) = G_UNMERGE_VALUES %0
%5:_(s32) = G_ANYEXT %1 %5:_(s32) = G_ANYEXT %1
%6:_(s32) = G_ANYEXT %2 %6:_(s32) = G_ANYEXT %2
%7:_(s32) = G_ANYEXT %3 %7:_(s32) = G_ANYEXT %3
%8:_(s32) = G_ANYEXT %4 %8:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5 $vgpr0 = COPY %5
$vgpr1 = COPY %6 $vgpr1 = COPY %6
$vgpr2 = COPY %7 $vgpr2 = COPY %7
$vgpr3 = COPY %8 $vgpr3 = COPY %8
... ...
--- ---
name: test_unmerge_s16_v6s16 name: test_unmerge_s16_v6s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s16_v6s16 ; CHECK-LABEL: name: test_unmerge_s16_v6s16
; CHECK: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2 ; CHECK: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<6 x s16>) ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<6 x s16>)
; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16) ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16) ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16) ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s16) ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s16) ; CHECK: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; CHECK: $vgpr0 = COPY [[ANYEXT]](s32) ; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; CHECK: $vgpr1 = COPY [[ANYEXT1]](s32) ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: $vgpr2 = COPY [[ANYEXT2]](s32) ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: $vgpr3 = COPY [[ANYEXT3]](s32) ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: $vgpr4 = COPY [[ANYEXT4]](s32) ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: $vgpr5 = COPY [[ANYEXT5]](s32) ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
; CHECK: $vgpr4 = COPY [[COPY5]](s32)
; CHECK: $vgpr5 = COPY [[COPY6]](s32)
%0:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2 %0:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
%1:_(s16), %2:_(s16), %3:_(s16), %4:_(s16), %5:_(s16), %6:_(s16) = G_UNMERGE_VALUES %0 %1:_(s16), %2:_(s16), %3:_(s16), %4:_(s16), %5:_(s16), %6:_(s16) = G_UNMERGE_VALUES %0
%7:_(s32) = G_ANYEXT %1 %7:_(s32) = G_ANYEXT %1
%8:_(s32) = G_ANYEXT %2 %8:_(s32) = G_ANYEXT %2
%9:_(s32) = G_ANYEXT %3 %9:_(s32) = G_ANYEXT %3
%10:_(s32) = G_ANYEXT %4 %10:_(s32) = G_ANYEXT %4
%11:_(s32) = G_ANYEXT %5 %11:_(s32) = G_ANYEXT %5
%12:_(s32) = G_ANYEXT %6 %12:_(s32) = G_ANYEXT %6
▲ Show 20 Lines • Show All 345 Lines • Show Last 20 Lines

test/CodeGen/AMDGPU/GlobalISel/legalize-zext.mir

Show First 20 Lines • Show All 124 Lines • ▼ Show 20 Lines
--- ---
name: test_zext_v4s16_to_v4s32 name: test_zext_v4s16_to_v4s32
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_zext_v4s16_to_v4s32 ; CHECK-LABEL: name: test_zext_v4s16_to_v4s32
; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>) ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16) ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16) ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[ZEXT]](s32), [[ZEXT1]](s32), [[ZEXT2]](s32), [[ZEXT3]](s32) ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]]
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[AND]](s32), [[AND1]](s32), [[AND2]](s32), [[AND3]](s32)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>) ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s32>) = G_ZEXT %0 %1:_(<4 x s32>) = G_ZEXT %0
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1 $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
... ...
--- ---
name: test_zext_v2s32_to_v2s64 name: test_zext_v2s32_to_v2s64
▲ Show 20 Lines • Show All 55 Lines • Show Last 20 Lines