[GlobalSel][NFC] Remove LLT initializers taking scalar sizes #119725
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
tgymnich
force-pushed
the
llt-remove-int-init
branch
from
c7c3932 to
8af2e4e
Compare
|
@llvm/pr-subscribers-backend-aarch64 @llvm/pr-subscribers-backend-amdgpu Author: Tim Gymnich (tgymnich) ChangesRemove initializers taking a scalar size in favor of initializers taking a LLT. This avoids dropping additional type information that may be included in LLT in the future. progresses on #119667 Patch is 121.16 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/119725.diff 36 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/MachineFunction.h b/llvm/include/llvm/CodeGen/MachineFunction.h
index 547cc26eda2295..5fb397b99e0d77 100644
--- a/llvm/include/llvm/CodeGen/MachineFunction.h
+++ b/llvm/include/llvm/CodeGen/MachineFunction.h
@@ -1063,12 +1063,12 @@ class LLVM_ABI MachineFunction {
int64_t Offset, LLT Ty);
MachineMemOperand *getMachineMemOperand(const MachineMemOperand *MMO,
int64_t Offset, LocationSize Size) {
- return getMachineMemOperand(
- MMO, Offset,
- !Size.hasValue() ? LLT()
- : Size.isScalable()
- ? LLT::scalable_vector(1, 8 * Size.getValue().getKnownMinValue())
- : LLT::scalar(8 * Size.getValue().getKnownMinValue()));
+ if (!Size.hasValue())
+ return getMachineMemOperand(MMO, Offset, LLT());
+
+ ElementCount EC = ElementCount::get(1, Size.isScalable());
+ LLT Ty = LLT::scalar(8 * Size.getValue().getKnownMinValue());
+ return getMachineMemOperand(MMO, Offset, LLT::scalarOrVector(EC, Ty));
}
MachineMemOperand *getMachineMemOperand(const MachineMemOperand *MMO,
int64_t Offset, uint64_t Size) {
diff --git a/llvm/include/llvm/CodeGenTypes/LowLevelType.h b/llvm/include/llvm/CodeGenTypes/LowLevelType.h
index 62ee28cfac99c5..f139800025541d 100644
--- a/llvm/include/llvm/CodeGenTypes/LowLevelType.h
+++ b/llvm/include/llvm/CodeGenTypes/LowLevelType.h
@@ -60,13 +60,6 @@ class LLT {
ElementCount::getFixed(0), SizeInBits, AddressSpace};
}
- /// Get a low-level vector of some number of elements and element width.
- static constexpr LLT vector(ElementCount EC, unsigned ScalarSizeInBits) {
- assert(!EC.isScalar() && "invalid number of vector elements");
- return LLT{/*isPointer=*/false, /*isVector=*/true, /*isScalar=*/false,
- EC, ScalarSizeInBits, /*AddressSpace=*/0};
- }
-
/// Get a low-level vector of some number of elements and element type.
static constexpr LLT vector(ElementCount EC, LLT ScalarTy) {
assert(!EC.isScalar() && "invalid number of vector elements");
@@ -95,26 +88,12 @@ class LLT {
return scalar(64);
}
- /// Get a low-level fixed-width vector of some number of elements and element
- /// width.
- static constexpr LLT fixed_vector(unsigned NumElements,
- unsigned ScalarSizeInBits) {
- return vector(ElementCount::getFixed(NumElements), ScalarSizeInBits);
- }
-
/// Get a low-level fixed-width vector of some number of elements and element
/// type.
static constexpr LLT fixed_vector(unsigned NumElements, LLT ScalarTy) {
return vector(ElementCount::getFixed(NumElements), ScalarTy);
}
- /// Get a low-level scalable vector of some number of elements and element
- /// width.
- static constexpr LLT scalable_vector(unsigned MinNumElements,
- unsigned ScalarSizeInBits) {
- return vector(ElementCount::getScalable(MinNumElements), ScalarSizeInBits);
- }
-
/// Get a low-level scalable vector of some number of elements and element
/// type.
static constexpr LLT scalable_vector(unsigned MinNumElements, LLT ScalarTy) {
@@ -125,12 +104,6 @@ class LLT {
return EC.isScalar() ? ScalarTy : LLT::vector(EC, ScalarTy);
}
- static constexpr LLT scalarOrVector(ElementCount EC, uint64_t ScalarSize) {
- assert(ScalarSize <= std::numeric_limits<unsigned>::max() &&
- "Not enough bits in LLT to represent size");
- return scalarOrVector(EC, LLT::scalar(static_cast<unsigned>(ScalarSize)));
- }
-
explicit constexpr LLT(bool isPointer, bool isVector, bool isScalar,
ElementCount EC, uint64_t SizeInBits,
unsigned AddressSpace)
@@ -215,16 +188,6 @@ class LLT {
return isVector() ? LLT::vector(getElementCount(), NewEltTy) : NewEltTy;
}
- /// If this type is a vector, return a vector with the same number of elements
- /// but the new element size. Otherwise, return the new element type. Invalid
- /// for pointer types. For pointer types, use changeElementType.
- constexpr LLT changeElementSize(unsigned NewEltSize) const {
- assert(!isPointerOrPointerVector() &&
- "invalid to directly change element size for pointers");
- return isVector() ? LLT::vector(getElementCount(), NewEltSize)
- : LLT::scalar(NewEltSize);
- }
-
/// Return a vector or scalar with the same element type and the new element
/// count.
constexpr LLT changeElementCount(ElementCount EC) const {
diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
index a2737995446526..d53de6e6db0754 100644
--- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
@@ -2580,7 +2580,7 @@ static LLT getMidVTForTruncRightShiftCombine(LLT ShiftTy, LLT TruncTy) {
// ShiftTy > 32 > TruncTy -> 32
if (ShiftSize > 32 && TruncSize < 32)
- return ShiftTy.changeElementSize(32);
+ return ShiftTy.changeElementType(LLT::scalar(32));
// TODO: We could also reduce to 16 bits, but that's more target-dependent.
// Some targets like it, some don't, some only like it under certain
@@ -5362,9 +5362,8 @@ MachineInstr *CombinerHelper::buildUDivUsingMul(MachineInstr &MI) {
Q = MIB.buildLShr(Ty, Q, PostShift).getReg(0);
auto One = MIB.buildConstant(Ty, 1);
- auto IsOne = MIB.buildICmp(
- CmpInst::Predicate::ICMP_EQ,
- Ty.isScalar() ? LLT::scalar(1) : Ty.changeElementSize(1), RHS, One);
+ auto IsOne = MIB.buildICmp(CmpInst::Predicate::ICMP_EQ,
+ Ty.changeElementType(LLT::scalar(1)), RHS, One);
return MIB.buildSelect(Ty, IsOne, LHS, Q);
}
@@ -5404,8 +5403,7 @@ bool CombinerHelper::matchUDivByConst(MachineInstr &MI) {
return false;
if (!isLegalOrBeforeLegalizer(
{TargetOpcode::G_ICMP,
- {DstTy.isVector() ? DstTy.changeElementSize(1) : LLT::scalar(1),
- DstTy}}))
+ {DstTy.changeElementType(LLT::scalar(1)), DstTy}}))
return false;
}
@@ -5538,8 +5536,7 @@ void CombinerHelper::applySDivByPow2(MachineInstr &MI) {
Register RHS = SDiv.getReg(2);
LLT Ty = MRI.getType(Dst);
LLT ShiftAmtTy = getTargetLowering().getPreferredShiftAmountTy(Ty);
- LLT CCVT =
- Ty.isVector() ? LLT::vector(Ty.getElementCount(), 1) : LLT::scalar(1);
+ LLT CCVT = Ty.changeElementType(LLT::scalar(1));
// Effectively we want to lower G_SDIV %lhs, %rhs, where %rhs is a power of 2,
// to the following version:
diff --git a/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp b/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
index 45b403bdd07658..93561b04150406 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
@@ -343,12 +343,11 @@ LegacyLegalizerInfo::findVectorLegalAction(const InstrAspect &Aspect) const {
ScalarInVectorActions[OpcodeIdx][TypeIdx];
LLT IntermediateType;
- auto ElementSizeAndAction =
+ auto &&[ElementSize, Action1] =
findAction(ElemSizeVec, Aspect.Type.getScalarSizeInBits());
- IntermediateType = LLT::fixed_vector(Aspect.Type.getNumElements(),
- ElementSizeAndAction.first);
- if (ElementSizeAndAction.second != Legal)
- return {ElementSizeAndAction.second, IntermediateType};
+ IntermediateType = Aspect.Type.changeElementType(LLT::scalar(ElementSize));
+ if (Action1 != Legal)
+ return {Action1, IntermediateType};
auto i = NumElements2Actions[OpcodeIdx].find(
IntermediateType.getScalarSizeInBits());
@@ -356,11 +355,10 @@ LegacyLegalizerInfo::findVectorLegalAction(const InstrAspect &Aspect) const {
return {NotFound, IntermediateType};
}
const SizeAndActionsVec &NumElementsVec = (*i).second[TypeIdx];
- auto NumElementsAndAction =
+ auto &&[NumElements, Action2] =
findAction(NumElementsVec, IntermediateType.getNumElements());
- return {NumElementsAndAction.second,
- LLT::fixed_vector(NumElementsAndAction.first,
- IntermediateType.getScalarSizeInBits())};
+ return {Action2,
+ LLT::fixed_vector(NumElements, IntermediateType.getScalarType())};
}
unsigned LegacyLegalizerInfo::getOpcodeIdxForOpcode(unsigned Opcode) const {
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
index 25c1db91b05d8e..b718300537ee21 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
@@ -80,7 +80,8 @@ LegalizeMutation LegalizeMutations::widenScalarOrEltToNextPow2(unsigned TypeIdx,
const LLT Ty = Query.Types[TypeIdx];
unsigned NewEltSizeInBits =
std::max(1u << Log2_32_Ceil(Ty.getScalarSizeInBits()), Min);
- return std::make_pair(TypeIdx, Ty.changeElementSize(NewEltSizeInBits));
+ return std::make_pair(TypeIdx,
+ Ty.changeElementType(LLT::scalar(NewEltSizeInBits)));
};
}
@@ -90,7 +91,8 @@ LegalizeMutations::widenScalarOrEltToNextMultipleOf(unsigned TypeIdx,
return [=](const LegalityQuery &Query) {
const LLT Ty = Query.Types[TypeIdx];
unsigned NewEltSizeInBits = alignTo(Ty.getScalarSizeInBits(), Size);
- return std::make_pair(TypeIdx, Ty.changeElementSize(NewEltSizeInBits));
+ return std::make_pair(TypeIdx,
+ Ty.changeElementType(LLT::scalar(NewEltSizeInBits)));
};
}
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index f56be39838ba7d..d5ec612b19cbd1 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -3081,9 +3081,8 @@ LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
LLT VecTy = MRI.getType(VecReg);
Observer.changingInstr(MI);
- widenScalarSrc(
- MI, LLT::vector(VecTy.getElementCount(), WideTy.getSizeInBits()), 1,
- TargetOpcode::G_ANYEXT);
+ widenScalarSrc(MI, VecTy.changeElementType(WideTy), 1,
+ TargetOpcode::G_ANYEXT);
widenScalarDst(MI, WideTy, 0);
Observer.changedInstr(MI);
@@ -3815,7 +3814,7 @@ LegalizerHelper::bitcastExtractSubvector(MachineInstr &MI, unsigned TypeIdx,
return UnableToLegalize;
Idx /= AdjustAmt;
- SrcTy = LLT::vector(SrcTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
+ SrcTy = SrcTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
auto CastVec = MIRBuilder.buildBitcast(SrcTy, Src);
auto PromotedES = MIRBuilder.buildExtractSubvector(CastTy, CastVec, Idx);
MIRBuilder.buildBitcast(Dst, PromotedES);
@@ -3883,8 +3882,10 @@ LegalizerHelper::bitcastInsertSubvector(MachineInstr &MI, unsigned TypeIdx,
return UnableToLegalize;
Idx /= AdjustAmt;
- BigVecTy = LLT::vector(BigVecTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
- SubVecTy = LLT::vector(SubVecTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
+ BigVecTy =
+ BigVecTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
+ SubVecTy =
+ SubVecTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
auto CastBigVec = MIRBuilder.buildBitcast(BigVecTy, BigVec);
auto CastSubVec = MIRBuilder.buildBitcast(SubVecTy, SubVec);
auto PromotedIS =
@@ -4632,7 +4633,8 @@ Register LegalizerHelper::getVectorElementPointer(Register VecPtr, LLT VecTy,
const DataLayout &DL = MIRBuilder.getDataLayout();
unsigned AS = MRI.getType(VecPtr).getAddressSpace();
unsigned IndexSizeInBits = DL.getIndexSize(AS) * 8;
- LLT IdxTy = MRI.getType(Index).changeElementSize(IndexSizeInBits);
+ LLT IdxTy =
+ MRI.getType(Index).changeElementType(LLT::scalar(IndexSizeInBits));
if (IdxTy != MRI.getType(Index))
Index = MIRBuilder.buildSExtOrTrunc(IdxTy, Index).getReg(0);
@@ -5348,8 +5350,8 @@ LegalizerHelper::fewerElementsBitcast(MachineInstr &MI, unsigned int TypeIdx,
auto [DstReg, DstTy, SrcReg, SrcTy] = MI.getFirst2RegLLTs();
unsigned SrcScalSize = SrcTy.getScalarSizeInBits();
- LLT SrcNarrowTy =
- LLT::fixed_vector(NarrowTy.getSizeInBits() / SrcScalSize, SrcScalSize);
+ LLT SrcNarrowTy = LLT::fixed_vector(NarrowTy.getSizeInBits() / SrcScalSize,
+ SrcTy.getScalarType());
// Split the Src and Dst Reg into smaller registers
SmallVector<Register> SrcVRegs, BitcastVRegs;
@@ -6883,8 +6885,9 @@ LegalizerHelper::lowerBitCount(MachineInstr &MI) {
// If CTLZ_ZERO_UNDEF is supported, emit that and a select for zero.
auto CtlzZU = MIRBuilder.buildCTLZ_ZERO_UNDEF(DstTy, SrcReg);
auto ZeroSrc = MIRBuilder.buildConstant(SrcTy, 0);
- auto ICmp = MIRBuilder.buildICmp(
- CmpInst::ICMP_EQ, SrcTy.changeElementSize(1), SrcReg, ZeroSrc);
+ auto ICmp = MIRBuilder.buildICmp(CmpInst::ICMP_EQ,
+ SrcTy.changeElementType(LLT::scalar(1)),
+ SrcReg, ZeroSrc);
auto LenConst = MIRBuilder.buildConstant(DstTy, Len);
MIRBuilder.buildSelect(DstReg, ICmp, LenConst, CtlzZU);
MI.eraseFromParent();
@@ -6931,8 +6934,9 @@ LegalizerHelper::lowerBitCount(MachineInstr &MI) {
// zero.
auto CttzZU = MIRBuilder.buildCTTZ_ZERO_UNDEF(DstTy, SrcReg);
auto Zero = MIRBuilder.buildConstant(SrcTy, 0);
- auto ICmp = MIRBuilder.buildICmp(
- CmpInst::ICMP_EQ, DstTy.changeElementSize(1), SrcReg, Zero);
+ auto ICmp = MIRBuilder.buildICmp(CmpInst::ICMP_EQ,
+ DstTy.changeElementType(LLT::scalar(1)),
+ SrcReg, Zero);
auto LenConst = MIRBuilder.buildConstant(DstTy, Len);
MIRBuilder.buildSelect(DstReg, ICmp, LenConst, CttzZU);
MI.eraseFromParent();
@@ -7177,7 +7181,7 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerEXT(MachineInstr &MI) {
// The step between extend is too large, split it by creating an intermediate
// extend instruction
if (SrcTyScalarSize * 2 < DstTyScalarSize) {
- LLT MidTy = SrcTy.changeElementSize(SrcTyScalarSize * 2);
+ LLT MidTy = SrcTy.changeElementType(LLT::scalar(SrcTyScalarSize * 2));
// If the destination type is illegal, split it into multiple statements
// zext x -> zext(merge(zext(unmerge), zext(unmerge)))
auto NewExt = MIRBuilder.buildInstr(MI.getOpcode(), {MidTy}, {Src});
@@ -7236,16 +7240,17 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerTRUNC(MachineInstr &MI) {
// Truncate the splits into intermediate narrower elements.
LLT InterTy;
if (DstTy.getScalarSizeInBits() * 2 < SrcTy.getScalarSizeInBits())
- InterTy = SplitSrcTy.changeElementSize(DstTy.getScalarSizeInBits() * 2);
+ InterTy = SplitSrcTy.changeElementType(
+ LLT::scalar(DstTy.getScalarSizeInBits() * 2));
else
- InterTy = SplitSrcTy.changeElementSize(DstTy.getScalarSizeInBits());
+ InterTy = SplitSrcTy.changeElementType(DstTy.getScalarType());
for (unsigned I = 0; I < SplitSrcs.size(); ++I) {
SplitSrcs[I] = MIRBuilder.buildTrunc(InterTy, SplitSrcs[I]).getReg(0);
}
// Combine the new truncates into one vector
auto Merge = MIRBuilder.buildMergeLikeInstr(
- DstTy.changeElementSize(InterTy.getScalarSizeInBits()), SplitSrcs);
+ DstTy.changeElementType(InterTy.getScalarType()), SplitSrcs);
// Truncate the new vector to the final result type
if (DstTy.getScalarSizeInBits() * 2 < SrcTy.getScalarSizeInBits())
@@ -7672,6 +7677,8 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
bool AreExactFloatBounds = !(MinStatus & APFloat::opStatus::opInexact) &&
!(MaxStatus & APFloat::opStatus::opInexact);
+ const LLT S1 = LLT::scalar(1);
+
// If the integer bounds are exactly representable as floats, emit a
// min+max+fptoi sequence. Otherwise we have to use a sequence of comparisons
// and selects.
@@ -7679,13 +7686,13 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// Clamp Src by MinFloat from below. If Src is NaN the result is MinFloat.
auto MaxC = MIRBuilder.buildFConstant(SrcTy, MinFloat);
auto MaxP = MIRBuilder.buildFCmp(CmpInst::FCMP_ULT,
- SrcTy.changeElementSize(1), Src, MaxC);
+ SrcTy.changeElementType(S1), Src, MaxC);
auto Max = MIRBuilder.buildSelect(SrcTy, MaxP, Src, MaxC);
// Clamp by MaxFloat from above. NaN cannot occur.
auto MinC = MIRBuilder.buildFConstant(SrcTy, MaxFloat);
auto MinP =
- MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementSize(1), Max,
- MinC, MachineInstr::FmNoNans);
+ MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementType(S1),
+ Max, MinC, MachineInstr::FmNoNans);
auto Min =
MIRBuilder.buildSelect(SrcTy, MinP, Max, MinC, MachineInstr::FmNoNans);
// Convert clamped value to integer. In the unsigned case we're done,
@@ -7699,7 +7706,7 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// Otherwise, select 0 if Src is NaN.
auto FpToInt = MIRBuilder.buildFPTOSI(DstTy, Min);
auto IsZero = MIRBuilder.buildFCmp(CmpInst::FCMP_UNO,
- DstTy.changeElementSize(1), Src, Src);
+ DstTy.changeElementType(S1), Src, Src);
MIRBuilder.buildSelect(Dst, IsZero, MIRBuilder.buildConstant(DstTy, 0),
FpToInt);
MI.eraseFromParent();
@@ -7715,13 +7722,13 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// If Src ULT MinFloat, select MinInt. In particular, this also selects
// MinInt if Src is NaN.
auto ULT =
- MIRBuilder.buildFCmp(CmpInst::FCMP_ULT, SrcTy.changeElementSize(1), Src,
+ MIRBuilder.buildFCmp(CmpInst::FCMP_ULT, SrcTy.changeElementType(S1), Src,
MIRBuilder.buildFConstant(SrcTy, MinFloat));
auto Max = MIRBuilder.buildSelect(
DstTy, ULT, MIRBuilder.buildConstant(DstTy, MinInt), FpToInt);
// If Src OGT MaxFloat, select MaxInt.
auto OGT =
- MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementSize(1), Src,
+ MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementType(S1), Src,
MIRBuilder.buildFConstant(SrcTy, MaxFloat));
// In the unsigned case we are done, because we mapped NaN to MinInt, which
@@ -7737,7 +7744,7 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
auto Min = MIRBuilder.buildSelect(
DstTy, OGT, MIRBuilder.buildConstant(DstTy, MaxInt), Max);
auto IsZero = MIRBuilder.buildFCmp(CmpInst::FCMP_UNO,
- DstTy.changeElementSize(1), Src, Src);
+ DstTy.changeElementType(S1), Src, Src);
MIRBuilder.buildSelect(Dst, IsZero, MIRBuilder.buildConstant(DstTy, 0), Min);
MI.eraseFromParent();
return Legalized;
@@ -7900,7 +7907,7 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerMinMax(MachineInstr &MI) {
auto [Dst, Src0, Src1] = MI.getFirst3Regs();
const CmpInst::Predicate Pred = minMaxToCompare(MI.getOpcode());
- LLT CmpType = MRI.getType(Dst).changeElementSize(1);
+ LLT CmpType = MRI.getType(Dst).changeElementType(LLT::scalar(1));
auto Cmp = MIRBuilder.buildICmp(Pred, CmpType, Src0, Src1);
MIRBuilder.buildSelect(Dst, Cmp, Src0, Src1);
@@ -7915,7 +7922,7 @@ LegalizerHelper::lowerThreewayCompare(MachineInstr &MI) {
Register Dst = Cmp->getReg(0);
LLT DstTy = MRI.getType(Dst);
- LLT CmpTy = DstTy.changeElementSize(1);
+ LLT CmpTy = DstTy.changeElementType(LLT::scalar(1));
CmpInst::Predicate LTPredicate = Cmp->isSigned()
? CmpInst::Predicate::ICMP_SLT
@@ -8028,7 +8035,7 @@ LegalizerHelper::lowerIntrinsicRound(MachineInstr &MI) {
auto [DstReg, X] = MI.getFirst2Regs();
const unsigned Flags = MI.getFlags();
const LLT Ty = MRI.getType(DstReg);
- const LLT CondTy = Ty.changeElementSize(1);
+ const LLT CondTy = Ty.changeElementType(LLT::scalar(1));
// round(x) =>
// t = trunc(x);
@@ -8061,7 +8068,7 @@ LegalizerHelp...
[truncated]
|
|
@llvm/pr-subscribers-llvm-globalisel Author: Tim Gymnich (tgymnich) ChangesRemove initializers taking a scalar size in favor of initializers taking a LLT. This avoids dropping additional type information that may be included in LLT in the future. progresses on #119667 Patch is 121.16 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/119725.diff 36 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/MachineFunction.h b/llvm/include/llvm/CodeGen/MachineFunction.h
index 547cc26eda2295..5fb397b99e0d77 100644
--- a/llvm/include/llvm/CodeGen/MachineFunction.h
+++ b/llvm/include/llvm/CodeGen/MachineFunction.h
@@ -1063,12 +1063,12 @@ class LLVM_ABI MachineFunction {
int64_t Offset, LLT Ty);
MachineMemOperand *getMachineMemOperand(const MachineMemOperand *MMO,
int64_t Offset, LocationSize Size) {
- return getMachineMemOperand(
- MMO, Offset,
- !Size.hasValue() ? LLT()
- : Size.isScalable()
- ? LLT::scalable_vector(1, 8 * Size.getValue().getKnownMinValue())
- : LLT::scalar(8 * Size.getValue().getKnownMinValue()));
+ if (!Size.hasValue())
+ return getMachineMemOperand(MMO, Offset, LLT());
+
+ ElementCount EC = ElementCount::get(1, Size.isScalable());
+ LLT Ty = LLT::scalar(8 * Size.getValue().getKnownMinValue());
+ return getMachineMemOperand(MMO, Offset, LLT::scalarOrVector(EC, Ty));
}
MachineMemOperand *getMachineMemOperand(const MachineMemOperand *MMO,
int64_t Offset, uint64_t Size) {
diff --git a/llvm/include/llvm/CodeGenTypes/LowLevelType.h b/llvm/include/llvm/CodeGenTypes/LowLevelType.h
index 62ee28cfac99c5..f139800025541d 100644
--- a/llvm/include/llvm/CodeGenTypes/LowLevelType.h
+++ b/llvm/include/llvm/CodeGenTypes/LowLevelType.h
@@ -60,13 +60,6 @@ class LLT {
ElementCount::getFixed(0), SizeInBits, AddressSpace};
}
- /// Get a low-level vector of some number of elements and element width.
- static constexpr LLT vector(ElementCount EC, unsigned ScalarSizeInBits) {
- assert(!EC.isScalar() && "invalid number of vector elements");
- return LLT{/*isPointer=*/false, /*isVector=*/true, /*isScalar=*/false,
- EC, ScalarSizeInBits, /*AddressSpace=*/0};
- }
-
/// Get a low-level vector of some number of elements and element type.
static constexpr LLT vector(ElementCount EC, LLT ScalarTy) {
assert(!EC.isScalar() && "invalid number of vector elements");
@@ -95,26 +88,12 @@ class LLT {
return scalar(64);
}
- /// Get a low-level fixed-width vector of some number of elements and element
- /// width.
- static constexpr LLT fixed_vector(unsigned NumElements,
- unsigned ScalarSizeInBits) {
- return vector(ElementCount::getFixed(NumElements), ScalarSizeInBits);
- }
-
/// Get a low-level fixed-width vector of some number of elements and element
/// type.
static constexpr LLT fixed_vector(unsigned NumElements, LLT ScalarTy) {
return vector(ElementCount::getFixed(NumElements), ScalarTy);
}
- /// Get a low-level scalable vector of some number of elements and element
- /// width.
- static constexpr LLT scalable_vector(unsigned MinNumElements,
- unsigned ScalarSizeInBits) {
- return vector(ElementCount::getScalable(MinNumElements), ScalarSizeInBits);
- }
-
/// Get a low-level scalable vector of some number of elements and element
/// type.
static constexpr LLT scalable_vector(unsigned MinNumElements, LLT ScalarTy) {
@@ -125,12 +104,6 @@ class LLT {
return EC.isScalar() ? ScalarTy : LLT::vector(EC, ScalarTy);
}
- static constexpr LLT scalarOrVector(ElementCount EC, uint64_t ScalarSize) {
- assert(ScalarSize <= std::numeric_limits<unsigned>::max() &&
- "Not enough bits in LLT to represent size");
- return scalarOrVector(EC, LLT::scalar(static_cast<unsigned>(ScalarSize)));
- }
-
explicit constexpr LLT(bool isPointer, bool isVector, bool isScalar,
ElementCount EC, uint64_t SizeInBits,
unsigned AddressSpace)
@@ -215,16 +188,6 @@ class LLT {
return isVector() ? LLT::vector(getElementCount(), NewEltTy) : NewEltTy;
}
- /// If this type is a vector, return a vector with the same number of elements
- /// but the new element size. Otherwise, return the new element type. Invalid
- /// for pointer types. For pointer types, use changeElementType.
- constexpr LLT changeElementSize(unsigned NewEltSize) const {
- assert(!isPointerOrPointerVector() &&
- "invalid to directly change element size for pointers");
- return isVector() ? LLT::vector(getElementCount(), NewEltSize)
- : LLT::scalar(NewEltSize);
- }
-
/// Return a vector or scalar with the same element type and the new element
/// count.
constexpr LLT changeElementCount(ElementCount EC) const {
diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
index a2737995446526..d53de6e6db0754 100644
--- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
@@ -2580,7 +2580,7 @@ static LLT getMidVTForTruncRightShiftCombine(LLT ShiftTy, LLT TruncTy) {
// ShiftTy > 32 > TruncTy -> 32
if (ShiftSize > 32 && TruncSize < 32)
- return ShiftTy.changeElementSize(32);
+ return ShiftTy.changeElementType(LLT::scalar(32));
// TODO: We could also reduce to 16 bits, but that's more target-dependent.
// Some targets like it, some don't, some only like it under certain
@@ -5362,9 +5362,8 @@ MachineInstr *CombinerHelper::buildUDivUsingMul(MachineInstr &MI) {
Q = MIB.buildLShr(Ty, Q, PostShift).getReg(0);
auto One = MIB.buildConstant(Ty, 1);
- auto IsOne = MIB.buildICmp(
- CmpInst::Predicate::ICMP_EQ,
- Ty.isScalar() ? LLT::scalar(1) : Ty.changeElementSize(1), RHS, One);
+ auto IsOne = MIB.buildICmp(CmpInst::Predicate::ICMP_EQ,
+ Ty.changeElementType(LLT::scalar(1)), RHS, One);
return MIB.buildSelect(Ty, IsOne, LHS, Q);
}
@@ -5404,8 +5403,7 @@ bool CombinerHelper::matchUDivByConst(MachineInstr &MI) {
return false;
if (!isLegalOrBeforeLegalizer(
{TargetOpcode::G_ICMP,
- {DstTy.isVector() ? DstTy.changeElementSize(1) : LLT::scalar(1),
- DstTy}}))
+ {DstTy.changeElementType(LLT::scalar(1)), DstTy}}))
return false;
}
@@ -5538,8 +5536,7 @@ void CombinerHelper::applySDivByPow2(MachineInstr &MI) {
Register RHS = SDiv.getReg(2);
LLT Ty = MRI.getType(Dst);
LLT ShiftAmtTy = getTargetLowering().getPreferredShiftAmountTy(Ty);
- LLT CCVT =
- Ty.isVector() ? LLT::vector(Ty.getElementCount(), 1) : LLT::scalar(1);
+ LLT CCVT = Ty.changeElementType(LLT::scalar(1));
// Effectively we want to lower G_SDIV %lhs, %rhs, where %rhs is a power of 2,
// to the following version:
diff --git a/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp b/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
index 45b403bdd07658..93561b04150406 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
@@ -343,12 +343,11 @@ LegacyLegalizerInfo::findVectorLegalAction(const InstrAspect &Aspect) const {
ScalarInVectorActions[OpcodeIdx][TypeIdx];
LLT IntermediateType;
- auto ElementSizeAndAction =
+ auto &&[ElementSize, Action1] =
findAction(ElemSizeVec, Aspect.Type.getScalarSizeInBits());
- IntermediateType = LLT::fixed_vector(Aspect.Type.getNumElements(),
- ElementSizeAndAction.first);
- if (ElementSizeAndAction.second != Legal)
- return {ElementSizeAndAction.second, IntermediateType};
+ IntermediateType = Aspect.Type.changeElementType(LLT::scalar(ElementSize));
+ if (Action1 != Legal)
+ return {Action1, IntermediateType};
auto i = NumElements2Actions[OpcodeIdx].find(
IntermediateType.getScalarSizeInBits());
@@ -356,11 +355,10 @@ LegacyLegalizerInfo::findVectorLegalAction(const InstrAspect &Aspect) const {
return {NotFound, IntermediateType};
}
const SizeAndActionsVec &NumElementsVec = (*i).second[TypeIdx];
- auto NumElementsAndAction =
+ auto &&[NumElements, Action2] =
findAction(NumElementsVec, IntermediateType.getNumElements());
- return {NumElementsAndAction.second,
- LLT::fixed_vector(NumElementsAndAction.first,
- IntermediateType.getScalarSizeInBits())};
+ return {Action2,
+ LLT::fixed_vector(NumElements, IntermediateType.getScalarType())};
}
unsigned LegacyLegalizerInfo::getOpcodeIdxForOpcode(unsigned Opcode) const {
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
index 25c1db91b05d8e..b718300537ee21 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
@@ -80,7 +80,8 @@ LegalizeMutation LegalizeMutations::widenScalarOrEltToNextPow2(unsigned TypeIdx,
const LLT Ty = Query.Types[TypeIdx];
unsigned NewEltSizeInBits =
std::max(1u << Log2_32_Ceil(Ty.getScalarSizeInBits()), Min);
- return std::make_pair(TypeIdx, Ty.changeElementSize(NewEltSizeInBits));
+ return std::make_pair(TypeIdx,
+ Ty.changeElementType(LLT::scalar(NewEltSizeInBits)));
};
}
@@ -90,7 +91,8 @@ LegalizeMutations::widenScalarOrEltToNextMultipleOf(unsigned TypeIdx,
return [=](const LegalityQuery &Query) {
const LLT Ty = Query.Types[TypeIdx];
unsigned NewEltSizeInBits = alignTo(Ty.getScalarSizeInBits(), Size);
- return std::make_pair(TypeIdx, Ty.changeElementSize(NewEltSizeInBits));
+ return std::make_pair(TypeIdx,
+ Ty.changeElementType(LLT::scalar(NewEltSizeInBits)));
};
}
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index f56be39838ba7d..d5ec612b19cbd1 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -3081,9 +3081,8 @@ LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
LLT VecTy = MRI.getType(VecReg);
Observer.changingInstr(MI);
- widenScalarSrc(
- MI, LLT::vector(VecTy.getElementCount(), WideTy.getSizeInBits()), 1,
- TargetOpcode::G_ANYEXT);
+ widenScalarSrc(MI, VecTy.changeElementType(WideTy), 1,
+ TargetOpcode::G_ANYEXT);
widenScalarDst(MI, WideTy, 0);
Observer.changedInstr(MI);
@@ -3815,7 +3814,7 @@ LegalizerHelper::bitcastExtractSubvector(MachineInstr &MI, unsigned TypeIdx,
return UnableToLegalize;
Idx /= AdjustAmt;
- SrcTy = LLT::vector(SrcTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
+ SrcTy = SrcTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
auto CastVec = MIRBuilder.buildBitcast(SrcTy, Src);
auto PromotedES = MIRBuilder.buildExtractSubvector(CastTy, CastVec, Idx);
MIRBuilder.buildBitcast(Dst, PromotedES);
@@ -3883,8 +3882,10 @@ LegalizerHelper::bitcastInsertSubvector(MachineInstr &MI, unsigned TypeIdx,
return UnableToLegalize;
Idx /= AdjustAmt;
- BigVecTy = LLT::vector(BigVecTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
- SubVecTy = LLT::vector(SubVecTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
+ BigVecTy =
+ BigVecTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
+ SubVecTy =
+ SubVecTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
auto CastBigVec = MIRBuilder.buildBitcast(BigVecTy, BigVec);
auto CastSubVec = MIRBuilder.buildBitcast(SubVecTy, SubVec);
auto PromotedIS =
@@ -4632,7 +4633,8 @@ Register LegalizerHelper::getVectorElementPointer(Register VecPtr, LLT VecTy,
const DataLayout &DL = MIRBuilder.getDataLayout();
unsigned AS = MRI.getType(VecPtr).getAddressSpace();
unsigned IndexSizeInBits = DL.getIndexSize(AS) * 8;
- LLT IdxTy = MRI.getType(Index).changeElementSize(IndexSizeInBits);
+ LLT IdxTy =
+ MRI.getType(Index).changeElementType(LLT::scalar(IndexSizeInBits));
if (IdxTy != MRI.getType(Index))
Index = MIRBuilder.buildSExtOrTrunc(IdxTy, Index).getReg(0);
@@ -5348,8 +5350,8 @@ LegalizerHelper::fewerElementsBitcast(MachineInstr &MI, unsigned int TypeIdx,
auto [DstReg, DstTy, SrcReg, SrcTy] = MI.getFirst2RegLLTs();
unsigned SrcScalSize = SrcTy.getScalarSizeInBits();
- LLT SrcNarrowTy =
- LLT::fixed_vector(NarrowTy.getSizeInBits() / SrcScalSize, SrcScalSize);
+ LLT SrcNarrowTy = LLT::fixed_vector(NarrowTy.getSizeInBits() / SrcScalSize,
+ SrcTy.getScalarType());
// Split the Src and Dst Reg into smaller registers
SmallVector<Register> SrcVRegs, BitcastVRegs;
@@ -6883,8 +6885,9 @@ LegalizerHelper::lowerBitCount(MachineInstr &MI) {
// If CTLZ_ZERO_UNDEF is supported, emit that and a select for zero.
auto CtlzZU = MIRBuilder.buildCTLZ_ZERO_UNDEF(DstTy, SrcReg);
auto ZeroSrc = MIRBuilder.buildConstant(SrcTy, 0);
- auto ICmp = MIRBuilder.buildICmp(
- CmpInst::ICMP_EQ, SrcTy.changeElementSize(1), SrcReg, ZeroSrc);
+ auto ICmp = MIRBuilder.buildICmp(CmpInst::ICMP_EQ,
+ SrcTy.changeElementType(LLT::scalar(1)),
+ SrcReg, ZeroSrc);
auto LenConst = MIRBuilder.buildConstant(DstTy, Len);
MIRBuilder.buildSelect(DstReg, ICmp, LenConst, CtlzZU);
MI.eraseFromParent();
@@ -6931,8 +6934,9 @@ LegalizerHelper::lowerBitCount(MachineInstr &MI) {
// zero.
auto CttzZU = MIRBuilder.buildCTTZ_ZERO_UNDEF(DstTy, SrcReg);
auto Zero = MIRBuilder.buildConstant(SrcTy, 0);
- auto ICmp = MIRBuilder.buildICmp(
- CmpInst::ICMP_EQ, DstTy.changeElementSize(1), SrcReg, Zero);
+ auto ICmp = MIRBuilder.buildICmp(CmpInst::ICMP_EQ,
+ DstTy.changeElementType(LLT::scalar(1)),
+ SrcReg, Zero);
auto LenConst = MIRBuilder.buildConstant(DstTy, Len);
MIRBuilder.buildSelect(DstReg, ICmp, LenConst, CttzZU);
MI.eraseFromParent();
@@ -7177,7 +7181,7 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerEXT(MachineInstr &MI) {
// The step between extend is too large, split it by creating an intermediate
// extend instruction
if (SrcTyScalarSize * 2 < DstTyScalarSize) {
- LLT MidTy = SrcTy.changeElementSize(SrcTyScalarSize * 2);
+ LLT MidTy = SrcTy.changeElementType(LLT::scalar(SrcTyScalarSize * 2));
// If the destination type is illegal, split it into multiple statements
// zext x -> zext(merge(zext(unmerge), zext(unmerge)))
auto NewExt = MIRBuilder.buildInstr(MI.getOpcode(), {MidTy}, {Src});
@@ -7236,16 +7240,17 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerTRUNC(MachineInstr &MI) {
// Truncate the splits into intermediate narrower elements.
LLT InterTy;
if (DstTy.getScalarSizeInBits() * 2 < SrcTy.getScalarSizeInBits())
- InterTy = SplitSrcTy.changeElementSize(DstTy.getScalarSizeInBits() * 2);
+ InterTy = SplitSrcTy.changeElementType(
+ LLT::scalar(DstTy.getScalarSizeInBits() * 2));
else
- InterTy = SplitSrcTy.changeElementSize(DstTy.getScalarSizeInBits());
+ InterTy = SplitSrcTy.changeElementType(DstTy.getScalarType());
for (unsigned I = 0; I < SplitSrcs.size(); ++I) {
SplitSrcs[I] = MIRBuilder.buildTrunc(InterTy, SplitSrcs[I]).getReg(0);
}
// Combine the new truncates into one vector
auto Merge = MIRBuilder.buildMergeLikeInstr(
- DstTy.changeElementSize(InterTy.getScalarSizeInBits()), SplitSrcs);
+ DstTy.changeElementType(InterTy.getScalarType()), SplitSrcs);
// Truncate the new vector to the final result type
if (DstTy.getScalarSizeInBits() * 2 < SrcTy.getScalarSizeInBits())
@@ -7672,6 +7677,8 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
bool AreExactFloatBounds = !(MinStatus & APFloat::opStatus::opInexact) &&
!(MaxStatus & APFloat::opStatus::opInexact);
+ const LLT S1 = LLT::scalar(1);
+
// If the integer bounds are exactly representable as floats, emit a
// min+max+fptoi sequence. Otherwise we have to use a sequence of comparisons
// and selects.
@@ -7679,13 +7686,13 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// Clamp Src by MinFloat from below. If Src is NaN the result is MinFloat.
auto MaxC = MIRBuilder.buildFConstant(SrcTy, MinFloat);
auto MaxP = MIRBuilder.buildFCmp(CmpInst::FCMP_ULT,
- SrcTy.changeElementSize(1), Src, MaxC);
+ SrcTy.changeElementType(S1), Src, MaxC);
auto Max = MIRBuilder.buildSelect(SrcTy, MaxP, Src, MaxC);
// Clamp by MaxFloat from above. NaN cannot occur.
auto MinC = MIRBuilder.buildFConstant(SrcTy, MaxFloat);
auto MinP =
- MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementSize(1), Max,
- MinC, MachineInstr::FmNoNans);
+ MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementType(S1),
+ Max, MinC, MachineInstr::FmNoNans);
auto Min =
MIRBuilder.buildSelect(SrcTy, MinP, Max, MinC, MachineInstr::FmNoNans);
// Convert clamped value to integer. In the unsigned case we're done,
@@ -7699,7 +7706,7 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// Otherwise, select 0 if Src is NaN.
auto FpToInt = MIRBuilder.buildFPTOSI(DstTy, Min);
auto IsZero = MIRBuilder.buildFCmp(CmpInst::FCMP_UNO,
- DstTy.changeElementSize(1), Src, Src);
+ DstTy.changeElementType(S1), Src, Src);
MIRBuilder.buildSelect(Dst, IsZero, MIRBuilder.buildConstant(DstTy, 0),
FpToInt);
MI.eraseFromParent();
@@ -7715,13 +7722,13 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// If Src ULT MinFloat, select MinInt. In particular, this also selects
// MinInt if Src is NaN.
auto ULT =
- MIRBuilder.buildFCmp(CmpInst::FCMP_ULT, SrcTy.changeElementSize(1), Src,
+ MIRBuilder.buildFCmp(CmpInst::FCMP_ULT, SrcTy.changeElementType(S1), Src,
MIRBuilder.buildFConstant(SrcTy, MinFloat));
auto Max = MIRBuilder.buildSelect(
DstTy, ULT, MIRBuilder.buildConstant(DstTy, MinInt), FpToInt);
// If Src OGT MaxFloat, select MaxInt.
auto OGT =
- MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementSize(1), Src,
+ MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementType(S1), Src,
MIRBuilder.buildFConstant(SrcTy, MaxFloat));
// In the unsigned case we are done, because we mapped NaN to MinInt, which
@@ -7737,7 +7744,7 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
auto Min = MIRBuilder.buildSelect(
DstTy, OGT, MIRBuilder.buildConstant(DstTy, MaxInt), Max);
auto IsZero = MIRBuilder.buildFCmp(CmpInst::FCMP_UNO,
- DstTy.changeElementSize(1), Src, Src);
+ DstTy.changeElementType(S1), Src, Src);
MIRBuilder.buildSelect(Dst, IsZero, MIRBuilder.buildConstant(DstTy, 0), Min);
MI.eraseFromParent();
return Legalized;
@@ -7900,7 +7907,7 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerMinMax(MachineInstr &MI) {
auto [Dst, Src0, Src1] = MI.getFirst3Regs();
const CmpInst::Predicate Pred = minMaxToCompare(MI.getOpcode());
- LLT CmpType = MRI.getType(Dst).changeElementSize(1);
+ LLT CmpType = MRI.getType(Dst).changeElementType(LLT::scalar(1));
auto Cmp = MIRBuilder.buildICmp(Pred, CmpType, Src0, Src1);
MIRBuilder.buildSelect(Dst, Cmp, Src0, Src1);
@@ -7915,7 +7922,7 @@ LegalizerHelper::lowerThreewayCompare(MachineInstr &MI) {
Register Dst = Cmp->getReg(0);
LLT DstTy = MRI.getType(Dst);
- LLT CmpTy = DstTy.changeElementSize(1);
+ LLT CmpTy = DstTy.changeElementType(LLT::scalar(1));
CmpInst::Predicate LTPredicate = Cmp->isSigned()
? CmpInst::Predicate::ICMP_SLT
@@ -8028,7 +8035,7 @@ LegalizerHelper::lowerIntrinsicRound(MachineInstr &MI) {
auto [DstReg, X] = MI.getFirst2Regs();
const unsigned Flags = MI.getFlags();
const LLT Ty = MRI.getType(DstReg);
- const LLT CondTy = Ty.changeElementSize(1);
+ const LLT CondTy = Ty.changeElementType(LLT::scalar(1));
// round(x) =>
// t = trunc(x);
@@ -8061,7 +8068,7 @@ LegalizerHelp...
[truncated]
|
|
@llvm/pr-subscribers-tablegen Author: Tim Gymnich (tgymnich) ChangesRemove initializers taking a scalar size in favor of initializers taking a LLT. This avoids dropping additional type information that may be included in LLT in the future. progresses on #119667 Patch is 121.16 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/119725.diff 36 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/MachineFunction.h b/llvm/include/llvm/CodeGen/MachineFunction.h
index 547cc26eda2295..5fb397b99e0d77 100644
--- a/llvm/include/llvm/CodeGen/MachineFunction.h
+++ b/llvm/include/llvm/CodeGen/MachineFunction.h
@@ -1063,12 +1063,12 @@ class LLVM_ABI MachineFunction {
int64_t Offset, LLT Ty);
MachineMemOperand *getMachineMemOperand(const MachineMemOperand *MMO,
int64_t Offset, LocationSize Size) {
- return getMachineMemOperand(
- MMO, Offset,
- !Size.hasValue() ? LLT()
- : Size.isScalable()
- ? LLT::scalable_vector(1, 8 * Size.getValue().getKnownMinValue())
- : LLT::scalar(8 * Size.getValue().getKnownMinValue()));
+ if (!Size.hasValue())
+ return getMachineMemOperand(MMO, Offset, LLT());
+
+ ElementCount EC = ElementCount::get(1, Size.isScalable());
+ LLT Ty = LLT::scalar(8 * Size.getValue().getKnownMinValue());
+ return getMachineMemOperand(MMO, Offset, LLT::scalarOrVector(EC, Ty));
}
MachineMemOperand *getMachineMemOperand(const MachineMemOperand *MMO,
int64_t Offset, uint64_t Size) {
diff --git a/llvm/include/llvm/CodeGenTypes/LowLevelType.h b/llvm/include/llvm/CodeGenTypes/LowLevelType.h
index 62ee28cfac99c5..f139800025541d 100644
--- a/llvm/include/llvm/CodeGenTypes/LowLevelType.h
+++ b/llvm/include/llvm/CodeGenTypes/LowLevelType.h
@@ -60,13 +60,6 @@ class LLT {
ElementCount::getFixed(0), SizeInBits, AddressSpace};
}
- /// Get a low-level vector of some number of elements and element width.
- static constexpr LLT vector(ElementCount EC, unsigned ScalarSizeInBits) {
- assert(!EC.isScalar() && "invalid number of vector elements");
- return LLT{/*isPointer=*/false, /*isVector=*/true, /*isScalar=*/false,
- EC, ScalarSizeInBits, /*AddressSpace=*/0};
- }
-
/// Get a low-level vector of some number of elements and element type.
static constexpr LLT vector(ElementCount EC, LLT ScalarTy) {
assert(!EC.isScalar() && "invalid number of vector elements");
@@ -95,26 +88,12 @@ class LLT {
return scalar(64);
}
- /// Get a low-level fixed-width vector of some number of elements and element
- /// width.
- static constexpr LLT fixed_vector(unsigned NumElements,
- unsigned ScalarSizeInBits) {
- return vector(ElementCount::getFixed(NumElements), ScalarSizeInBits);
- }
-
/// Get a low-level fixed-width vector of some number of elements and element
/// type.
static constexpr LLT fixed_vector(unsigned NumElements, LLT ScalarTy) {
return vector(ElementCount::getFixed(NumElements), ScalarTy);
}
- /// Get a low-level scalable vector of some number of elements and element
- /// width.
- static constexpr LLT scalable_vector(unsigned MinNumElements,
- unsigned ScalarSizeInBits) {
- return vector(ElementCount::getScalable(MinNumElements), ScalarSizeInBits);
- }
-
/// Get a low-level scalable vector of some number of elements and element
/// type.
static constexpr LLT scalable_vector(unsigned MinNumElements, LLT ScalarTy) {
@@ -125,12 +104,6 @@ class LLT {
return EC.isScalar() ? ScalarTy : LLT::vector(EC, ScalarTy);
}
- static constexpr LLT scalarOrVector(ElementCount EC, uint64_t ScalarSize) {
- assert(ScalarSize <= std::numeric_limits<unsigned>::max() &&
- "Not enough bits in LLT to represent size");
- return scalarOrVector(EC, LLT::scalar(static_cast<unsigned>(ScalarSize)));
- }
-
explicit constexpr LLT(bool isPointer, bool isVector, bool isScalar,
ElementCount EC, uint64_t SizeInBits,
unsigned AddressSpace)
@@ -215,16 +188,6 @@ class LLT {
return isVector() ? LLT::vector(getElementCount(), NewEltTy) : NewEltTy;
}
- /// If this type is a vector, return a vector with the same number of elements
- /// but the new element size. Otherwise, return the new element type. Invalid
- /// for pointer types. For pointer types, use changeElementType.
- constexpr LLT changeElementSize(unsigned NewEltSize) const {
- assert(!isPointerOrPointerVector() &&
- "invalid to directly change element size for pointers");
- return isVector() ? LLT::vector(getElementCount(), NewEltSize)
- : LLT::scalar(NewEltSize);
- }
-
/// Return a vector or scalar with the same element type and the new element
/// count.
constexpr LLT changeElementCount(ElementCount EC) const {
diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
index a2737995446526..d53de6e6db0754 100644
--- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
@@ -2580,7 +2580,7 @@ static LLT getMidVTForTruncRightShiftCombine(LLT ShiftTy, LLT TruncTy) {
// ShiftTy > 32 > TruncTy -> 32
if (ShiftSize > 32 && TruncSize < 32)
- return ShiftTy.changeElementSize(32);
+ return ShiftTy.changeElementType(LLT::scalar(32));
// TODO: We could also reduce to 16 bits, but that's more target-dependent.
// Some targets like it, some don't, some only like it under certain
@@ -5362,9 +5362,8 @@ MachineInstr *CombinerHelper::buildUDivUsingMul(MachineInstr &MI) {
Q = MIB.buildLShr(Ty, Q, PostShift).getReg(0);
auto One = MIB.buildConstant(Ty, 1);
- auto IsOne = MIB.buildICmp(
- CmpInst::Predicate::ICMP_EQ,
- Ty.isScalar() ? LLT::scalar(1) : Ty.changeElementSize(1), RHS, One);
+ auto IsOne = MIB.buildICmp(CmpInst::Predicate::ICMP_EQ,
+ Ty.changeElementType(LLT::scalar(1)), RHS, One);
return MIB.buildSelect(Ty, IsOne, LHS, Q);
}
@@ -5404,8 +5403,7 @@ bool CombinerHelper::matchUDivByConst(MachineInstr &MI) {
return false;
if (!isLegalOrBeforeLegalizer(
{TargetOpcode::G_ICMP,
- {DstTy.isVector() ? DstTy.changeElementSize(1) : LLT::scalar(1),
- DstTy}}))
+ {DstTy.changeElementType(LLT::scalar(1)), DstTy}}))
return false;
}
@@ -5538,8 +5536,7 @@ void CombinerHelper::applySDivByPow2(MachineInstr &MI) {
Register RHS = SDiv.getReg(2);
LLT Ty = MRI.getType(Dst);
LLT ShiftAmtTy = getTargetLowering().getPreferredShiftAmountTy(Ty);
- LLT CCVT =
- Ty.isVector() ? LLT::vector(Ty.getElementCount(), 1) : LLT::scalar(1);
+ LLT CCVT = Ty.changeElementType(LLT::scalar(1));
// Effectively we want to lower G_SDIV %lhs, %rhs, where %rhs is a power of 2,
// to the following version:
diff --git a/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp b/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
index 45b403bdd07658..93561b04150406 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegacyLegalizerInfo.cpp
@@ -343,12 +343,11 @@ LegacyLegalizerInfo::findVectorLegalAction(const InstrAspect &Aspect) const {
ScalarInVectorActions[OpcodeIdx][TypeIdx];
LLT IntermediateType;
- auto ElementSizeAndAction =
+ auto &&[ElementSize, Action1] =
findAction(ElemSizeVec, Aspect.Type.getScalarSizeInBits());
- IntermediateType = LLT::fixed_vector(Aspect.Type.getNumElements(),
- ElementSizeAndAction.first);
- if (ElementSizeAndAction.second != Legal)
- return {ElementSizeAndAction.second, IntermediateType};
+ IntermediateType = Aspect.Type.changeElementType(LLT::scalar(ElementSize));
+ if (Action1 != Legal)
+ return {Action1, IntermediateType};
auto i = NumElements2Actions[OpcodeIdx].find(
IntermediateType.getScalarSizeInBits());
@@ -356,11 +355,10 @@ LegacyLegalizerInfo::findVectorLegalAction(const InstrAspect &Aspect) const {
return {NotFound, IntermediateType};
}
const SizeAndActionsVec &NumElementsVec = (*i).second[TypeIdx];
- auto NumElementsAndAction =
+ auto &&[NumElements, Action2] =
findAction(NumElementsVec, IntermediateType.getNumElements());
- return {NumElementsAndAction.second,
- LLT::fixed_vector(NumElementsAndAction.first,
- IntermediateType.getScalarSizeInBits())};
+ return {Action2,
+ LLT::fixed_vector(NumElements, IntermediateType.getScalarType())};
}
unsigned LegacyLegalizerInfo::getOpcodeIdxForOpcode(unsigned Opcode) const {
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
index 25c1db91b05d8e..b718300537ee21 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizeMutations.cpp
@@ -80,7 +80,8 @@ LegalizeMutation LegalizeMutations::widenScalarOrEltToNextPow2(unsigned TypeIdx,
const LLT Ty = Query.Types[TypeIdx];
unsigned NewEltSizeInBits =
std::max(1u << Log2_32_Ceil(Ty.getScalarSizeInBits()), Min);
- return std::make_pair(TypeIdx, Ty.changeElementSize(NewEltSizeInBits));
+ return std::make_pair(TypeIdx,
+ Ty.changeElementType(LLT::scalar(NewEltSizeInBits)));
};
}
@@ -90,7 +91,8 @@ LegalizeMutations::widenScalarOrEltToNextMultipleOf(unsigned TypeIdx,
return [=](const LegalityQuery &Query) {
const LLT Ty = Query.Types[TypeIdx];
unsigned NewEltSizeInBits = alignTo(Ty.getScalarSizeInBits(), Size);
- return std::make_pair(TypeIdx, Ty.changeElementSize(NewEltSizeInBits));
+ return std::make_pair(TypeIdx,
+ Ty.changeElementType(LLT::scalar(NewEltSizeInBits)));
};
}
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index f56be39838ba7d..d5ec612b19cbd1 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -3081,9 +3081,8 @@ LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
LLT VecTy = MRI.getType(VecReg);
Observer.changingInstr(MI);
- widenScalarSrc(
- MI, LLT::vector(VecTy.getElementCount(), WideTy.getSizeInBits()), 1,
- TargetOpcode::G_ANYEXT);
+ widenScalarSrc(MI, VecTy.changeElementType(WideTy), 1,
+ TargetOpcode::G_ANYEXT);
widenScalarDst(MI, WideTy, 0);
Observer.changedInstr(MI);
@@ -3815,7 +3814,7 @@ LegalizerHelper::bitcastExtractSubvector(MachineInstr &MI, unsigned TypeIdx,
return UnableToLegalize;
Idx /= AdjustAmt;
- SrcTy = LLT::vector(SrcTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
+ SrcTy = SrcTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
auto CastVec = MIRBuilder.buildBitcast(SrcTy, Src);
auto PromotedES = MIRBuilder.buildExtractSubvector(CastTy, CastVec, Idx);
MIRBuilder.buildBitcast(Dst, PromotedES);
@@ -3883,8 +3882,10 @@ LegalizerHelper::bitcastInsertSubvector(MachineInstr &MI, unsigned TypeIdx,
return UnableToLegalize;
Idx /= AdjustAmt;
- BigVecTy = LLT::vector(BigVecTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
- SubVecTy = LLT::vector(SubVecTyEC.divideCoefficientBy(AdjustAmt), AdjustAmt);
+ BigVecTy =
+ BigVecTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
+ SubVecTy =
+ SubVecTy.divide(AdjustAmt).changeElementType(LLT::scalar(AdjustAmt));
auto CastBigVec = MIRBuilder.buildBitcast(BigVecTy, BigVec);
auto CastSubVec = MIRBuilder.buildBitcast(SubVecTy, SubVec);
auto PromotedIS =
@@ -4632,7 +4633,8 @@ Register LegalizerHelper::getVectorElementPointer(Register VecPtr, LLT VecTy,
const DataLayout &DL = MIRBuilder.getDataLayout();
unsigned AS = MRI.getType(VecPtr).getAddressSpace();
unsigned IndexSizeInBits = DL.getIndexSize(AS) * 8;
- LLT IdxTy = MRI.getType(Index).changeElementSize(IndexSizeInBits);
+ LLT IdxTy =
+ MRI.getType(Index).changeElementType(LLT::scalar(IndexSizeInBits));
if (IdxTy != MRI.getType(Index))
Index = MIRBuilder.buildSExtOrTrunc(IdxTy, Index).getReg(0);
@@ -5348,8 +5350,8 @@ LegalizerHelper::fewerElementsBitcast(MachineInstr &MI, unsigned int TypeIdx,
auto [DstReg, DstTy, SrcReg, SrcTy] = MI.getFirst2RegLLTs();
unsigned SrcScalSize = SrcTy.getScalarSizeInBits();
- LLT SrcNarrowTy =
- LLT::fixed_vector(NarrowTy.getSizeInBits() / SrcScalSize, SrcScalSize);
+ LLT SrcNarrowTy = LLT::fixed_vector(NarrowTy.getSizeInBits() / SrcScalSize,
+ SrcTy.getScalarType());
// Split the Src and Dst Reg into smaller registers
SmallVector<Register> SrcVRegs, BitcastVRegs;
@@ -6883,8 +6885,9 @@ LegalizerHelper::lowerBitCount(MachineInstr &MI) {
// If CTLZ_ZERO_UNDEF is supported, emit that and a select for zero.
auto CtlzZU = MIRBuilder.buildCTLZ_ZERO_UNDEF(DstTy, SrcReg);
auto ZeroSrc = MIRBuilder.buildConstant(SrcTy, 0);
- auto ICmp = MIRBuilder.buildICmp(
- CmpInst::ICMP_EQ, SrcTy.changeElementSize(1), SrcReg, ZeroSrc);
+ auto ICmp = MIRBuilder.buildICmp(CmpInst::ICMP_EQ,
+ SrcTy.changeElementType(LLT::scalar(1)),
+ SrcReg, ZeroSrc);
auto LenConst = MIRBuilder.buildConstant(DstTy, Len);
MIRBuilder.buildSelect(DstReg, ICmp, LenConst, CtlzZU);
MI.eraseFromParent();
@@ -6931,8 +6934,9 @@ LegalizerHelper::lowerBitCount(MachineInstr &MI) {
// zero.
auto CttzZU = MIRBuilder.buildCTTZ_ZERO_UNDEF(DstTy, SrcReg);
auto Zero = MIRBuilder.buildConstant(SrcTy, 0);
- auto ICmp = MIRBuilder.buildICmp(
- CmpInst::ICMP_EQ, DstTy.changeElementSize(1), SrcReg, Zero);
+ auto ICmp = MIRBuilder.buildICmp(CmpInst::ICMP_EQ,
+ DstTy.changeElementType(LLT::scalar(1)),
+ SrcReg, Zero);
auto LenConst = MIRBuilder.buildConstant(DstTy, Len);
MIRBuilder.buildSelect(DstReg, ICmp, LenConst, CttzZU);
MI.eraseFromParent();
@@ -7177,7 +7181,7 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerEXT(MachineInstr &MI) {
// The step between extend is too large, split it by creating an intermediate
// extend instruction
if (SrcTyScalarSize * 2 < DstTyScalarSize) {
- LLT MidTy = SrcTy.changeElementSize(SrcTyScalarSize * 2);
+ LLT MidTy = SrcTy.changeElementType(LLT::scalar(SrcTyScalarSize * 2));
// If the destination type is illegal, split it into multiple statements
// zext x -> zext(merge(zext(unmerge), zext(unmerge)))
auto NewExt = MIRBuilder.buildInstr(MI.getOpcode(), {MidTy}, {Src});
@@ -7236,16 +7240,17 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerTRUNC(MachineInstr &MI) {
// Truncate the splits into intermediate narrower elements.
LLT InterTy;
if (DstTy.getScalarSizeInBits() * 2 < SrcTy.getScalarSizeInBits())
- InterTy = SplitSrcTy.changeElementSize(DstTy.getScalarSizeInBits() * 2);
+ InterTy = SplitSrcTy.changeElementType(
+ LLT::scalar(DstTy.getScalarSizeInBits() * 2));
else
- InterTy = SplitSrcTy.changeElementSize(DstTy.getScalarSizeInBits());
+ InterTy = SplitSrcTy.changeElementType(DstTy.getScalarType());
for (unsigned I = 0; I < SplitSrcs.size(); ++I) {
SplitSrcs[I] = MIRBuilder.buildTrunc(InterTy, SplitSrcs[I]).getReg(0);
}
// Combine the new truncates into one vector
auto Merge = MIRBuilder.buildMergeLikeInstr(
- DstTy.changeElementSize(InterTy.getScalarSizeInBits()), SplitSrcs);
+ DstTy.changeElementType(InterTy.getScalarType()), SplitSrcs);
// Truncate the new vector to the final result type
if (DstTy.getScalarSizeInBits() * 2 < SrcTy.getScalarSizeInBits())
@@ -7672,6 +7677,8 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
bool AreExactFloatBounds = !(MinStatus & APFloat::opStatus::opInexact) &&
!(MaxStatus & APFloat::opStatus::opInexact);
+ const LLT S1 = LLT::scalar(1);
+
// If the integer bounds are exactly representable as floats, emit a
// min+max+fptoi sequence. Otherwise we have to use a sequence of comparisons
// and selects.
@@ -7679,13 +7686,13 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// Clamp Src by MinFloat from below. If Src is NaN the result is MinFloat.
auto MaxC = MIRBuilder.buildFConstant(SrcTy, MinFloat);
auto MaxP = MIRBuilder.buildFCmp(CmpInst::FCMP_ULT,
- SrcTy.changeElementSize(1), Src, MaxC);
+ SrcTy.changeElementType(S1), Src, MaxC);
auto Max = MIRBuilder.buildSelect(SrcTy, MaxP, Src, MaxC);
// Clamp by MaxFloat from above. NaN cannot occur.
auto MinC = MIRBuilder.buildFConstant(SrcTy, MaxFloat);
auto MinP =
- MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementSize(1), Max,
- MinC, MachineInstr::FmNoNans);
+ MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementType(S1),
+ Max, MinC, MachineInstr::FmNoNans);
auto Min =
MIRBuilder.buildSelect(SrcTy, MinP, Max, MinC, MachineInstr::FmNoNans);
// Convert clamped value to integer. In the unsigned case we're done,
@@ -7699,7 +7706,7 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// Otherwise, select 0 if Src is NaN.
auto FpToInt = MIRBuilder.buildFPTOSI(DstTy, Min);
auto IsZero = MIRBuilder.buildFCmp(CmpInst::FCMP_UNO,
- DstTy.changeElementSize(1), Src, Src);
+ DstTy.changeElementType(S1), Src, Src);
MIRBuilder.buildSelect(Dst, IsZero, MIRBuilder.buildConstant(DstTy, 0),
FpToInt);
MI.eraseFromParent();
@@ -7715,13 +7722,13 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
// If Src ULT MinFloat, select MinInt. In particular, this also selects
// MinInt if Src is NaN.
auto ULT =
- MIRBuilder.buildFCmp(CmpInst::FCMP_ULT, SrcTy.changeElementSize(1), Src,
+ MIRBuilder.buildFCmp(CmpInst::FCMP_ULT, SrcTy.changeElementType(S1), Src,
MIRBuilder.buildFConstant(SrcTy, MinFloat));
auto Max = MIRBuilder.buildSelect(
DstTy, ULT, MIRBuilder.buildConstant(DstTy, MinInt), FpToInt);
// If Src OGT MaxFloat, select MaxInt.
auto OGT =
- MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementSize(1), Src,
+ MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, SrcTy.changeElementType(S1), Src,
MIRBuilder.buildFConstant(SrcTy, MaxFloat));
// In the unsigned case we are done, because we mapped NaN to MinInt, which
@@ -7737,7 +7744,7 @@ LegalizerHelper::lowerFPTOINT_SAT(MachineInstr &MI) {
auto Min = MIRBuilder.buildSelect(
DstTy, OGT, MIRBuilder.buildConstant(DstTy, MaxInt), Max);
auto IsZero = MIRBuilder.buildFCmp(CmpInst::FCMP_UNO,
- DstTy.changeElementSize(1), Src, Src);
+ DstTy.changeElementType(S1), Src, Src);
MIRBuilder.buildSelect(Dst, IsZero, MIRBuilder.buildConstant(DstTy, 0), Min);
MI.eraseFromParent();
return Legalized;
@@ -7900,7 +7907,7 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerMinMax(MachineInstr &MI) {
auto [Dst, Src0, Src1] = MI.getFirst3Regs();
const CmpInst::Predicate Pred = minMaxToCompare(MI.getOpcode());
- LLT CmpType = MRI.getType(Dst).changeElementSize(1);
+ LLT CmpType = MRI.getType(Dst).changeElementType(LLT::scalar(1));
auto Cmp = MIRBuilder.buildICmp(Pred, CmpType, Src0, Src1);
MIRBuilder.buildSelect(Dst, Cmp, Src0, Src1);
@@ -7915,7 +7922,7 @@ LegalizerHelper::lowerThreewayCompare(MachineInstr &MI) {
Register Dst = Cmp->getReg(0);
LLT DstTy = MRI.getType(Dst);
- LLT CmpTy = DstTy.changeElementSize(1);
+ LLT CmpTy = DstTy.changeElementType(LLT::scalar(1));
CmpInst::Predicate LTPredicate = Cmp->isSigned()
? CmpInst::Predicate::ICMP_SLT
@@ -8028,7 +8035,7 @@ LegalizerHelper::lowerIntrinsicRound(MachineInstr &MI) {
auto [DstReg, X] = MI.getFirst2Regs();
const unsigned Flags = MI.getFlags();
const LLT Ty = MRI.getType(DstReg);
- const LLT CondTy = Ty.changeElementSize(1);
+ const LLT CondTy = Ty.changeElementType(LLT::scalar(1));
// round(x) =>
// t = trunc(x);
@@ -8061,7 +8068,7 @@ LegalizerHelp...
[truncated]
|
arsenm
approved these changes
Dec 16, 2024
tgymnich
force-pushed
the
llt-remove-int-init
branch
from
8af2e4e to
4768857
Compare
tgymnich
force-pushed
the
llt-remove-int-init
branch
from
4768857 to
2ee10a5
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Remove initializers taking a scalar size in favor of initializers taking a LLT. This avoids dropping additional type information that may be included in LLT in the future.
progresses on #119667