[VPlan] Populate and use VPIRFlags from initial VPInstruction. #168450
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
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Florian Hahn (fhahn) ChangesUpdate VPlan to populate VPIRFlags during VPInstruction construction and use it when creating widened recipes, instead of constructing VPIRFlags from the underlying IR instruction each time. The VPRecipeWithIRFlags constructor taking an underlying instruction and setting the flags based on it has been removed. This centralizes initial VPIRFlags creation and ensures flags are consistently available throughout VPlan transformations and makes sure we don't accidentally re-add flags from the underlying instruction that already got dropped during transformations. Follow-up to #167253, which did the same for VPIRMetadata. Patch is 33.47 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/168450.diff 9 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h b/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
index f533a47150a7b..741392247c0d6 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
@@ -152,11 +152,12 @@ class VPBuilder {
/// its underlying Instruction.
VPInstruction *createNaryOp(unsigned Opcode, ArrayRef<VPValue *> Operands,
Instruction *Inst = nullptr,
+ const VPIRFlags &Flags = {},
const VPIRMetadata &MD = {},
DebugLoc DL = DebugLoc::getUnknown(),
const Twine &Name = "") {
VPInstruction *NewVPInst = tryInsertInstruction(
- new VPInstruction(Opcode, Operands, {}, MD, DL, Name));
+ new VPInstruction(Opcode, Operands, Flags, MD, DL, Name));
NewVPInst->setUnderlyingValue(Inst);
return NewVPInst;
}
@@ -329,7 +330,7 @@ class VPBuilder {
else if (Opcode == Instruction::ZExt)
Flags = VPIRFlags::NonNegFlagsTy(false);
return tryInsertInstruction(
- new VPWidenCastRecipe(Opcode, Op, ResultTy, Flags));
+ new VPWidenCastRecipe(Opcode, Op, ResultTy, nullptr, Flags));
}
VPScalarIVStepsRecipe *
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 356d759b94799..c680b6fca84cd 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7750,7 +7750,7 @@ VPSingleDefRecipe *VPRecipeBuilder::tryToWidenCall(VPInstruction *VPI,
},
Range);
if (ShouldUseVectorIntrinsic)
- return new VPWidenIntrinsicRecipe(*CI, ID, Ops, CI->getType(), *VPI,
+ return new VPWidenIntrinsicRecipe(*CI, ID, Ops, CI->getType(), *VPI, *VPI,
VPI->getDebugLoc());
Function *Variant = nullptr;
@@ -7804,7 +7804,8 @@ VPSingleDefRecipe *VPRecipeBuilder::tryToWidenCall(VPInstruction *VPI,
}
Ops.push_back(VPI->getOperand(VPI->getNumOperands() - 1));
- return new VPWidenCallRecipe(CI, Variant, Ops, VPI->getDebugLoc());
+ return new VPWidenCallRecipe(CI, Variant, Ops, *VPI, *VPI,
+ VPI->getDebugLoc());
}
return nullptr;
@@ -7842,7 +7843,7 @@ VPWidenRecipe *VPRecipeBuilder::tryToWiden(VPInstruction *VPI) {
auto *SafeRHS =
Builder.createSelect(Mask, Ops[1], One, VPI->getDebugLoc());
Ops[1] = SafeRHS;
- return new VPWidenRecipe(*I, Ops, *VPI, VPI->getDebugLoc());
+ return new VPWidenRecipe(*I, Ops, *VPI, *VPI, VPI->getDebugLoc());
}
[[fallthrough]];
}
@@ -7888,7 +7889,7 @@ VPWidenRecipe *VPRecipeBuilder::tryToWiden(VPInstruction *VPI) {
// For other binops, the legacy cost model only checks the second operand.
NewOps[1] = GetConstantViaSCEV(NewOps[1]);
}
- return new VPWidenRecipe(*I, NewOps, *VPI, VPI->getDebugLoc());
+ return new VPWidenRecipe(*I, NewOps, *VPI, *VPI, VPI->getDebugLoc());
}
case Instruction::ExtractValue: {
SmallVector<VPValue *> NewOps(VPI->operands());
@@ -7896,7 +7897,7 @@ VPWidenRecipe *VPRecipeBuilder::tryToWiden(VPInstruction *VPI) {
assert(EVI->getNumIndices() == 1 && "Expected one extractvalue index");
unsigned Idx = EVI->getIndices()[0];
NewOps.push_back(Plan.getConstantInt(32, Idx));
- return new VPWidenRecipe(*I, NewOps, *VPI, VPI->getDebugLoc());
+ return new VPWidenRecipe(*I, NewOps, *VPI, *VPI, VPI->getDebugLoc());
}
};
}
@@ -7981,7 +7982,8 @@ VPReplicateRecipe *VPRecipeBuilder::handleReplication(VPInstruction *VPI,
(Range.Start.isScalable() && isa<IntrinsicInst>(I))) &&
"Should not predicate a uniform recipe");
auto *Recipe =
- new VPReplicateRecipe(I, VPI->operands(), IsUniform, BlockInMask, *VPI);
+ new VPReplicateRecipe(I, VPI->operands(), IsUniform, BlockInMask, *VPI,
+ *VPI, VPI->getDebugLoc());
return Recipe;
}
@@ -8231,17 +8233,19 @@ VPRecipeBase *VPRecipeBuilder::tryToCreateWidenRecipe(VPSingleDefRecipe *R,
return nullptr;
if (VPI->getOpcode() == Instruction::GetElementPtr)
- return new VPWidenGEPRecipe(cast<GetElementPtrInst>(Instr), R->operands());
+ return new VPWidenGEPRecipe(cast<GetElementPtrInst>(Instr), R->operands(),
+ *VPI, VPI->getDebugLoc());
if (VPI->getOpcode() == Instruction::Select)
- return new VPWidenSelectRecipe(*cast<SelectInst>(Instr), R->operands(),
- *VPI);
+ return new VPWidenSelectRecipe(cast<SelectInst>(Instr), R->operands(), *VPI,
+ *VPI, VPI->getDebugLoc());
if (Instruction::isCast(VPI->getOpcode())) {
- auto *CastR = cast<VPInstructionWithType>(R);
auto *CI = cast<CastInst>(Instr);
+ auto *CastR = cast<VPInstructionWithType>(VPI);
return new VPWidenCastRecipe(CI->getOpcode(), VPI->getOperand(0),
- CastR->getResultType(), *CI, *VPI);
+ CastR->getResultType(), CI, *VPI, *VPI,
+ VPI->getDebugLoc());
}
return tryToWiden(VPI);
@@ -8269,8 +8273,8 @@ VPRecipeBuilder::tryToCreatePartialReduction(VPInstruction *Reduction,
SmallVector<VPValue *, 2> Ops;
Ops.push_back(Plan.getOrAddLiveIn(Zero));
Ops.push_back(BinOp);
- BinOp = new VPWidenRecipe(*ReductionI, Ops, VPIRMetadata(),
- ReductionI->getDebugLoc());
+ BinOp = new VPWidenRecipe(*ReductionI, Ops, VPIRFlags(*ReductionI),
+ VPIRMetadata(), ReductionI->getDebugLoc());
Builder.insert(BinOp->getDefiningRecipe());
ReductionOpcode = Instruction::Add;
}
@@ -8454,9 +8458,10 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
Legal->isInvariantAddressOfReduction(SI->getPointerOperand())) {
// Only create recipe for the final invariant store of the reduction.
if (Legal->isInvariantStoreOfReduction(SI)) {
+ auto *VPI = cast<VPInstruction>(SingleDef);
auto *Recipe = new VPReplicateRecipe(
- SI, R.operands(), true /* IsUniform */, nullptr /*Mask*/,
- *cast<VPInstruction>(SingleDef));
+ SI, R.operands(), true /* IsUniform */, nullptr /*Mask*/, *VPI,
+ *VPI, VPI->getDebugLoc());
Recipe->insertBefore(*MiddleVPBB, MBIP);
}
R.eraseFromParent();
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index c81834e401726..094c9668b54ef 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -881,14 +881,6 @@ class VPIRFlags {
/// A pure-virtual common base class for recipes defining a single VPValue and
/// using IR flags.
struct VPRecipeWithIRFlags : public VPSingleDefRecipe, public VPIRFlags {
- VPRecipeWithIRFlags(const unsigned char SC, ArrayRef<VPValue *> Operands,
- DebugLoc DL = DebugLoc::getUnknown())
- : VPSingleDefRecipe(SC, Operands, DL), VPIRFlags() {}
-
- VPRecipeWithIRFlags(const unsigned char SC, ArrayRef<VPValue *> Operands,
- Instruction &I)
- : VPSingleDefRecipe(SC, Operands, &I, I.getDebugLoc()), VPIRFlags(I) {}
-
VPRecipeWithIRFlags(const unsigned char SC, ArrayRef<VPValue *> Operands,
const VPIRFlags &Flags,
DebugLoc DL = DebugLoc::getUnknown())
@@ -1466,9 +1458,12 @@ class LLVM_ABI_FOR_TEST VPWidenRecipe : public VPRecipeWithIRFlags,
VPIRMetadata(Metadata), Opcode(Opcode) {}
VPWidenRecipe(Instruction &I, ArrayRef<VPValue *> Operands,
- const VPIRMetadata &Metadata, DebugLoc DL)
- : VPRecipeWithIRFlags(VPDef::VPWidenSC, Operands, I),
- VPIRMetadata(Metadata), Opcode(I.getOpcode()) {}
+ const VPIRFlags &Flags = {}, const VPIRMetadata &Metadata = {},
+ DebugLoc DL = {})
+ : VPRecipeWithIRFlags(VPDef::VPWidenSC, Operands, Flags, DL),
+ VPIRMetadata(Metadata), Opcode(I.getOpcode()) {
+ setUnderlyingValue(&I);
+ }
~VPWidenRecipe() override = default;
@@ -1509,30 +1504,23 @@ class VPWidenCastRecipe : public VPRecipeWithIRFlags, public VPIRMetadata {
public:
VPWidenCastRecipe(Instruction::CastOps Opcode, VPValue *Op, Type *ResultTy,
- CastInst &UI, const VPIRMetadata &Metadata)
- : VPRecipeWithIRFlags(VPDef::VPWidenCastSC, Op, UI),
- VPIRMetadata(Metadata), Opcode(Opcode), ResultTy(ResultTy) {
- assert(UI.getOpcode() == Opcode &&
- "opcode of underlying cast doesn't match");
- }
- VPWidenCastRecipe(Instruction::CastOps Opcode, VPValue *Op, Type *ResultTy,
- const VPIRFlags &Flags = {},
+ CastInst *CI = nullptr, const VPIRFlags &Flags = {},
const VPIRMetadata &Metadata = {},
DebugLoc DL = DebugLoc::getUnknown())
: VPRecipeWithIRFlags(VPDef::VPWidenCastSC, Op, Flags, DL),
VPIRMetadata(Metadata), Opcode(Opcode), ResultTy(ResultTy) {
assert(flagsValidForOpcode(Opcode) &&
"Set flags not supported for the provided opcode");
+ setUnderlyingValue(CI);
}
~VPWidenCastRecipe() override = default;
VPWidenCastRecipe *clone() override {
- auto *New = new VPWidenCastRecipe(Opcode, getOperand(0), ResultTy, *this,
- *this, getDebugLoc());
- if (auto *UV = getUnderlyingValue())
- New->setUnderlyingValue(UV);
- return New;
+ return new VPWidenCastRecipe(
+ Opcode, getOperand(0), ResultTy,
+ dyn_cast_or_null<CastInst>(getUnderlyingValue()), *this, *this,
+ getDebugLoc());
}
VP_CLASSOF_IMPL(VPDef::VPWidenCastSC)
@@ -1577,13 +1565,17 @@ class VPWidenIntrinsicRecipe : public VPRecipeWithIRFlags, public VPIRMetadata {
public:
VPWidenIntrinsicRecipe(CallInst &CI, Intrinsic::ID VectorIntrinsicID,
ArrayRef<VPValue *> CallArguments, Type *Ty,
+ const VPIRFlags &Flags = {},
const VPIRMetadata &MD = {},
DebugLoc DL = DebugLoc::getUnknown())
- : VPRecipeWithIRFlags(VPDef::VPWidenIntrinsicSC, CallArguments, CI),
+ : VPRecipeWithIRFlags(VPDef::VPWidenIntrinsicSC, CallArguments, Flags,
+ DL),
VPIRMetadata(MD), VectorIntrinsicID(VectorIntrinsicID), ResultTy(Ty),
MayReadFromMemory(CI.mayReadFromMemory()),
MayWriteToMemory(CI.mayWriteToMemory()),
- MayHaveSideEffects(CI.mayHaveSideEffects()) {}
+ MayHaveSideEffects(CI.mayHaveSideEffects()) {
+ setUnderlyingValue(&CI);
+ }
VPWidenIntrinsicRecipe(Intrinsic::ID VectorIntrinsicID,
ArrayRef<VPValue *> CallArguments, Type *Ty,
@@ -1609,7 +1601,7 @@ class VPWidenIntrinsicRecipe : public VPRecipeWithIRFlags, public VPIRMetadata {
VPWidenIntrinsicRecipe *clone() override {
if (Value *CI = getUnderlyingValue())
return new VPWidenIntrinsicRecipe(*cast<CallInst>(CI), VectorIntrinsicID,
- operands(), ResultTy, *this,
+ operands(), ResultTy, *this, *this,
getDebugLoc());
return new VPWidenIntrinsicRecipe(VectorIntrinsicID, operands(), ResultTy,
*this, *this, getDebugLoc());
@@ -1663,10 +1655,11 @@ class LLVM_ABI_FOR_TEST VPWidenCallRecipe : public VPRecipeWithIRFlags,
public:
VPWidenCallRecipe(Value *UV, Function *Variant,
ArrayRef<VPValue *> CallArguments,
- DebugLoc DL = DebugLoc::getUnknown())
- : VPRecipeWithIRFlags(VPDef::VPWidenCallSC, CallArguments,
- *cast<Instruction>(UV)),
- VPIRMetadata(*cast<Instruction>(UV)), Variant(Variant) {
+ const VPIRFlags &Flags = {},
+ const VPIRMetadata &Metadata = {}, DebugLoc DL = {})
+ : VPRecipeWithIRFlags(VPDef::VPWidenCallSC, CallArguments, Flags, DL),
+ VPIRMetadata(Metadata), Variant(Variant) {
+ setUnderlyingValue(UV);
assert(
isa<Function>(getOperand(getNumOperands() - 1)->getLiveInIRValue()) &&
"last operand must be the called function");
@@ -1676,7 +1669,7 @@ class LLVM_ABI_FOR_TEST VPWidenCallRecipe : public VPRecipeWithIRFlags,
VPWidenCallRecipe *clone() override {
return new VPWidenCallRecipe(getUnderlyingValue(), Variant, operands(),
- getDebugLoc());
+ *this, *this, getDebugLoc());
}
VP_CLASSOF_IMPL(VPDef::VPWidenCallSC)
@@ -1753,16 +1746,19 @@ class VPHistogramRecipe : public VPRecipeBase {
/// instruction.
struct LLVM_ABI_FOR_TEST VPWidenSelectRecipe : public VPRecipeWithIRFlags,
public VPIRMetadata {
- VPWidenSelectRecipe(SelectInst &I, ArrayRef<VPValue *> Operands,
- const VPIRMetadata &MD = {})
- : VPRecipeWithIRFlags(VPDef::VPWidenSelectSC, Operands, I),
- VPIRMetadata(MD) {}
+ VPWidenSelectRecipe(SelectInst *SI, ArrayRef<VPValue *> Operands,
+ const VPIRFlags &Flags = {}, const VPIRMetadata &MD = {},
+ DebugLoc DL = {})
+ : VPRecipeWithIRFlags(VPDef::VPWidenSelectSC, Operands, Flags, DL),
+ VPIRMetadata(MD) {
+ setUnderlyingValue(SI);
+ }
~VPWidenSelectRecipe() override = default;
VPWidenSelectRecipe *clone() override {
- return new VPWidenSelectRecipe(*cast<SelectInst>(getUnderlyingInstr()),
- operands(), *this);
+ return new VPWidenSelectRecipe(cast<SelectInst>(getUnderlyingInstr()),
+ operands(), *this, *this, getDebugLoc());
}
VP_CLASSOF_IMPL(VPDef::VPWidenSelectSC)
@@ -1814,9 +1810,12 @@ class LLVM_ABI_FOR_TEST VPWidenGEPRecipe : public VPRecipeWithIRFlags {
}
public:
- VPWidenGEPRecipe(GetElementPtrInst *GEP, ArrayRef<VPValue *> Operands)
- : VPRecipeWithIRFlags(VPDef::VPWidenGEPSC, Operands, *GEP),
+ VPWidenGEPRecipe(GetElementPtrInst *GEP, ArrayRef<VPValue *> Operands,
+ const VPIRFlags &Flags = {},
+ DebugLoc DL = DebugLoc::getUnknown())
+ : VPRecipeWithIRFlags(VPDef::VPWidenGEPSC, Operands, Flags, DL),
SourceElementTy(GEP->getSourceElementType()) {
+ setUnderlyingValue(GEP);
SmallVector<std::pair<unsigned, MDNode *>> Metadata;
(void)Metadata;
getMetadataToPropagate(GEP, Metadata);
@@ -1827,7 +1826,7 @@ class LLVM_ABI_FOR_TEST VPWidenGEPRecipe : public VPRecipeWithIRFlags {
VPWidenGEPRecipe *clone() override {
return new VPWidenGEPRecipe(cast<GetElementPtrInst>(getUnderlyingInstr()),
- operands());
+ operands(), *this, getDebugLoc());
}
VP_CLASSOF_IMPL(VPDef::VPWidenGEPSC)
@@ -2921,10 +2920,12 @@ class LLVM_ABI_FOR_TEST VPReplicateRecipe : public VPRecipeWithIRFlags,
public:
VPReplicateRecipe(Instruction *I, ArrayRef<VPValue *> Operands,
bool IsSingleScalar, VPValue *Mask = nullptr,
- VPIRMetadata Metadata = {})
- : VPRecipeWithIRFlags(VPDef::VPReplicateSC, Operands, *I),
+ const VPIRFlags &Flags = {}, VPIRMetadata Metadata = {},
+ DebugLoc DL = DebugLoc::getUnknown())
+ : VPRecipeWithIRFlags(VPDef::VPReplicateSC, Operands, Flags, DL),
VPIRMetadata(Metadata), IsSingleScalar(IsSingleScalar),
IsPredicated(Mask) {
+ setUnderlyingValue(I);
if (Mask)
addOperand(Mask);
}
@@ -2932,9 +2933,9 @@ class LLVM_ABI_FOR_TEST VPReplicateRecipe : public VPRecipeWithIRFlags,
~VPReplicateRecipe() override = default;
VPReplicateRecipe *clone() override {
- auto *Copy =
- new VPReplicateRecipe(getUnderlyingInstr(), operands(), IsSingleScalar,
- isPredicated() ? getMask() : nullptr, *this);
+ auto *Copy = new VPReplicateRecipe(
+ getUnderlyingInstr(), operands(), IsSingleScalar,
+ isPredicated() ? getMask() : nullptr, *this, *this, getDebugLoc());
Copy->transferFlags(*this);
return Copy;
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
index 612202d049774..dbbde1cafa9f2 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
@@ -190,7 +190,7 @@ void PlainCFGBuilder::createVPInstructionsForVPBB(VPBasicBlock *VPBB,
// recipes.
if (Br->isConditional()) {
VPValue *Cond = getOrCreateVPOperand(Br->getCondition());
- VPIRBuilder.createNaryOp(VPInstruction::BranchOnCond, {Cond}, Inst,
+ VPIRBuilder.createNaryOp(VPInstruction::BranchOnCond, {Cond}, Inst, {},
VPIRMetadata(*Inst), Inst->getDebugLoc());
}
@@ -205,7 +205,7 @@ void PlainCFGBuilder::createVPInstructionsForVPBB(VPBasicBlock *VPBB,
SmallVector<VPValue *> Ops = {getOrCreateVPOperand(SI->getCondition())};
for (auto Case : SI->cases())
Ops.push_back(getOrCreateVPOperand(Case.getCaseValue()));
- VPIRBuilder.createNaryOp(Instruction::Switch, Ops, Inst,
+ VPIRBuilder.createNaryOp(Instruction::Switch, Ops, Inst, {},
VPIRMetadata(*Inst), Inst->getDebugLoc());
continue;
}
@@ -255,13 +255,14 @@ void PlainCFGBuilder::createVPInstructionsForVPBB(VPBasicBlock *VPBB,
if (auto *CI = dyn_cast<CastInst>(Inst)) {
NewR = VPIRBuilder.createScalarCast(CI->getOpcode(), VPOperands[0],
CI->getType(), CI->getDebugLoc(),
- {}, MD);
+ VPIRFlags(*CI), MD);
NewR->setUnderlyingValue(CI);
} else {
// Build VPInstruction for any arbitrary Instruction without specific
// representation in VPlan.
- NewR = VPIRBuilder.createNaryOp(Inst->getOpcode(), VPOperands, Inst, MD,
- Inst->getDebugLoc());
+ NewR =
+ VPIRBuilder.createNaryOp(Inst->getOpcode(), VPOperands, Inst,
+ VPIRFlags(*Inst), MD, Inst->getDebugLoc());
}
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index fca6554ad77c6..ef36e29aaa5c4 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -2056,24 +2056,26 @@ bool VPIRFlags::flagsValidForOpcode(unsigned Opcode) const {
switch (OpType) {
case OperationType::OverflowingBinOp:
return Opcode == Instruction::Add || Opcode == Instruction::Sub ||
- Opcode == Instruction::Mul ||
+ Opcode == Instruction::Mul || Opcode == Instruction::Shl ||
Opcode == VPInstruction::VPInstruction::CanonicalIVIncrementForPart;
case OperationType::Trunc:
return Opcode == Instruction::Trunc;
case OperationType::DisjointOp:
return Opcode == Instruction::Or;
case OperationType::PossiblyExactOp:
- return Opcode == Instruction::AShr;
+ return Opcode == Instruction::AShr || Opcode == Instruction::LShr ||
+ Opcode == Instruction::UDiv || Opcode == Instruction::SDiv;
case OperationType::GEPOp:
return Opcode == Instruction::GetElementPtr ||
Opcode == VPInstruction::PtrAdd ||
Opcode == VPInstruction::WidePtrAdd;
case OperationType::FPMathOp:
- return Opcode == Instruction::FAdd || Opcode == Instruction::FMul ||
- Opcode == Instruction::FSub || Opcode == Instruction::FNeg ||
- Opcode == Instruction::FDiv || Opcode == Instruction::FRem ||
- Opcode == Instruction::FPExt || Opcode == Instruction::FPTrunc ||
- Opcode == Instruction::FCmp || Opcode == Instruction::Select ||
+ return Opcode == Instruction::Call || Opcode == Instruction::FAdd ||
+ ...
[truncated]
|
Member
|
@llvm/pr-subscribers-vectorizers Author: Florian Hahn (fhahn) ChangesUpdate VPlan to populate VPIRFlags during VPInstruction construction and use it when creating widened recipes, instead of constructing VPIRFlags from the underlying IR instruction each time. The VPRecipeWithIRFlags constructor taking an underlying instruction and setting the flags based on it has been removed. This centralizes initial VPIRFlags creation and ensures flags are consistently available throughout VPlan transformations and makes sure we don't accidentally re-add flags from the underlying instruction that already got dropped during transformations. Follow-up to #167253, which did the same for VPIRMetadata. Patch is 33.47 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/168450.diff 9 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h b/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
index f533a47150a7b..741392247c0d6 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
@@ -152,11 +152,12 @@ class VPBuilder {
/// its underlying Instruction.
VPInstruction *createNaryOp(unsigned Opcode, ArrayRef<VPValue *> Operands,
Instruction *Inst = nullptr,
+ const VPIRFlags &Flags = {},
const VPIRMetadata &MD = {},
DebugLoc DL = DebugLoc::getUnknown(),
const Twine &Name = "") {
VPInstruction *NewVPInst = tryInsertInstruction(
- new VPInstruction(Opcode, Operands, {}, MD, DL, Name));
+ new VPInstruction(Opcode, Operands, Flags, MD, DL, Name));
NewVPInst->setUnderlyingValue(Inst);
return NewVPInst;
}
@@ -329,7 +330,7 @@ class VPBuilder {
else if (Opcode == Instruction::ZExt)
Flags = VPIRFlags::NonNegFlagsTy(false);
return tryInsertInstruction(
- new VPWidenCastRecipe(Opcode, Op, ResultTy, Flags));
+ new VPWidenCastRecipe(Opcode, Op, ResultTy, nullptr, Flags));
}
VPScalarIVStepsRecipe *
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 356d759b94799..c680b6fca84cd 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7750,7 +7750,7 @@ VPSingleDefRecipe *VPRecipeBuilder::tryToWidenCall(VPInstruction *VPI,
},
Range);
if (ShouldUseVectorIntrinsic)
- return new VPWidenIntrinsicRecipe(*CI, ID, Ops, CI->getType(), *VPI,
+ return new VPWidenIntrinsicRecipe(*CI, ID, Ops, CI->getType(), *VPI, *VPI,
VPI->getDebugLoc());
Function *Variant = nullptr;
@@ -7804,7 +7804,8 @@ VPSingleDefRecipe *VPRecipeBuilder::tryToWidenCall(VPInstruction *VPI,
}
Ops.push_back(VPI->getOperand(VPI->getNumOperands() - 1));
- return new VPWidenCallRecipe(CI, Variant, Ops, VPI->getDebugLoc());
+ return new VPWidenCallRecipe(CI, Variant, Ops, *VPI, *VPI,
+ VPI->getDebugLoc());
}
return nullptr;
@@ -7842,7 +7843,7 @@ VPWidenRecipe *VPRecipeBuilder::tryToWiden(VPInstruction *VPI) {
auto *SafeRHS =
Builder.createSelect(Mask, Ops[1], One, VPI->getDebugLoc());
Ops[1] = SafeRHS;
- return new VPWidenRecipe(*I, Ops, *VPI, VPI->getDebugLoc());
+ return new VPWidenRecipe(*I, Ops, *VPI, *VPI, VPI->getDebugLoc());
}
[[fallthrough]];
}
@@ -7888,7 +7889,7 @@ VPWidenRecipe *VPRecipeBuilder::tryToWiden(VPInstruction *VPI) {
// For other binops, the legacy cost model only checks the second operand.
NewOps[1] = GetConstantViaSCEV(NewOps[1]);
}
- return new VPWidenRecipe(*I, NewOps, *VPI, VPI->getDebugLoc());
+ return new VPWidenRecipe(*I, NewOps, *VPI, *VPI, VPI->getDebugLoc());
}
case Instruction::ExtractValue: {
SmallVector<VPValue *> NewOps(VPI->operands());
@@ -7896,7 +7897,7 @@ VPWidenRecipe *VPRecipeBuilder::tryToWiden(VPInstruction *VPI) {
assert(EVI->getNumIndices() == 1 && "Expected one extractvalue index");
unsigned Idx = EVI->getIndices()[0];
NewOps.push_back(Plan.getConstantInt(32, Idx));
- return new VPWidenRecipe(*I, NewOps, *VPI, VPI->getDebugLoc());
+ return new VPWidenRecipe(*I, NewOps, *VPI, *VPI, VPI->getDebugLoc());
}
};
}
@@ -7981,7 +7982,8 @@ VPReplicateRecipe *VPRecipeBuilder::handleReplication(VPInstruction *VPI,
(Range.Start.isScalable() && isa<IntrinsicInst>(I))) &&
"Should not predicate a uniform recipe");
auto *Recipe =
- new VPReplicateRecipe(I, VPI->operands(), IsUniform, BlockInMask, *VPI);
+ new VPReplicateRecipe(I, VPI->operands(), IsUniform, BlockInMask, *VPI,
+ *VPI, VPI->getDebugLoc());
return Recipe;
}
@@ -8231,17 +8233,19 @@ VPRecipeBase *VPRecipeBuilder::tryToCreateWidenRecipe(VPSingleDefRecipe *R,
return nullptr;
if (VPI->getOpcode() == Instruction::GetElementPtr)
- return new VPWidenGEPRecipe(cast<GetElementPtrInst>(Instr), R->operands());
+ return new VPWidenGEPRecipe(cast<GetElementPtrInst>(Instr), R->operands(),
+ *VPI, VPI->getDebugLoc());
if (VPI->getOpcode() == Instruction::Select)
- return new VPWidenSelectRecipe(*cast<SelectInst>(Instr), R->operands(),
- *VPI);
+ return new VPWidenSelectRecipe(cast<SelectInst>(Instr), R->operands(), *VPI,
+ *VPI, VPI->getDebugLoc());
if (Instruction::isCast(VPI->getOpcode())) {
- auto *CastR = cast<VPInstructionWithType>(R);
auto *CI = cast<CastInst>(Instr);
+ auto *CastR = cast<VPInstructionWithType>(VPI);
return new VPWidenCastRecipe(CI->getOpcode(), VPI->getOperand(0),
- CastR->getResultType(), *CI, *VPI);
+ CastR->getResultType(), CI, *VPI, *VPI,
+ VPI->getDebugLoc());
}
return tryToWiden(VPI);
@@ -8269,8 +8273,8 @@ VPRecipeBuilder::tryToCreatePartialReduction(VPInstruction *Reduction,
SmallVector<VPValue *, 2> Ops;
Ops.push_back(Plan.getOrAddLiveIn(Zero));
Ops.push_back(BinOp);
- BinOp = new VPWidenRecipe(*ReductionI, Ops, VPIRMetadata(),
- ReductionI->getDebugLoc());
+ BinOp = new VPWidenRecipe(*ReductionI, Ops, VPIRFlags(*ReductionI),
+ VPIRMetadata(), ReductionI->getDebugLoc());
Builder.insert(BinOp->getDefiningRecipe());
ReductionOpcode = Instruction::Add;
}
@@ -8454,9 +8458,10 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
Legal->isInvariantAddressOfReduction(SI->getPointerOperand())) {
// Only create recipe for the final invariant store of the reduction.
if (Legal->isInvariantStoreOfReduction(SI)) {
+ auto *VPI = cast<VPInstruction>(SingleDef);
auto *Recipe = new VPReplicateRecipe(
- SI, R.operands(), true /* IsUniform */, nullptr /*Mask*/,
- *cast<VPInstruction>(SingleDef));
+ SI, R.operands(), true /* IsUniform */, nullptr /*Mask*/, *VPI,
+ *VPI, VPI->getDebugLoc());
Recipe->insertBefore(*MiddleVPBB, MBIP);
}
R.eraseFromParent();
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index c81834e401726..094c9668b54ef 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -881,14 +881,6 @@ class VPIRFlags {
/// A pure-virtual common base class for recipes defining a single VPValue and
/// using IR flags.
struct VPRecipeWithIRFlags : public VPSingleDefRecipe, public VPIRFlags {
- VPRecipeWithIRFlags(const unsigned char SC, ArrayRef<VPValue *> Operands,
- DebugLoc DL = DebugLoc::getUnknown())
- : VPSingleDefRecipe(SC, Operands, DL), VPIRFlags() {}
-
- VPRecipeWithIRFlags(const unsigned char SC, ArrayRef<VPValue *> Operands,
- Instruction &I)
- : VPSingleDefRecipe(SC, Operands, &I, I.getDebugLoc()), VPIRFlags(I) {}
-
VPRecipeWithIRFlags(const unsigned char SC, ArrayRef<VPValue *> Operands,
const VPIRFlags &Flags,
DebugLoc DL = DebugLoc::getUnknown())
@@ -1466,9 +1458,12 @@ class LLVM_ABI_FOR_TEST VPWidenRecipe : public VPRecipeWithIRFlags,
VPIRMetadata(Metadata), Opcode(Opcode) {}
VPWidenRecipe(Instruction &I, ArrayRef<VPValue *> Operands,
- const VPIRMetadata &Metadata, DebugLoc DL)
- : VPRecipeWithIRFlags(VPDef::VPWidenSC, Operands, I),
- VPIRMetadata(Metadata), Opcode(I.getOpcode()) {}
+ const VPIRFlags &Flags = {}, const VPIRMetadata &Metadata = {},
+ DebugLoc DL = {})
+ : VPRecipeWithIRFlags(VPDef::VPWidenSC, Operands, Flags, DL),
+ VPIRMetadata(Metadata), Opcode(I.getOpcode()) {
+ setUnderlyingValue(&I);
+ }
~VPWidenRecipe() override = default;
@@ -1509,30 +1504,23 @@ class VPWidenCastRecipe : public VPRecipeWithIRFlags, public VPIRMetadata {
public:
VPWidenCastRecipe(Instruction::CastOps Opcode, VPValue *Op, Type *ResultTy,
- CastInst &UI, const VPIRMetadata &Metadata)
- : VPRecipeWithIRFlags(VPDef::VPWidenCastSC, Op, UI),
- VPIRMetadata(Metadata), Opcode(Opcode), ResultTy(ResultTy) {
- assert(UI.getOpcode() == Opcode &&
- "opcode of underlying cast doesn't match");
- }
- VPWidenCastRecipe(Instruction::CastOps Opcode, VPValue *Op, Type *ResultTy,
- const VPIRFlags &Flags = {},
+ CastInst *CI = nullptr, const VPIRFlags &Flags = {},
const VPIRMetadata &Metadata = {},
DebugLoc DL = DebugLoc::getUnknown())
: VPRecipeWithIRFlags(VPDef::VPWidenCastSC, Op, Flags, DL),
VPIRMetadata(Metadata), Opcode(Opcode), ResultTy(ResultTy) {
assert(flagsValidForOpcode(Opcode) &&
"Set flags not supported for the provided opcode");
+ setUnderlyingValue(CI);
}
~VPWidenCastRecipe() override = default;
VPWidenCastRecipe *clone() override {
- auto *New = new VPWidenCastRecipe(Opcode, getOperand(0), ResultTy, *this,
- *this, getDebugLoc());
- if (auto *UV = getUnderlyingValue())
- New->setUnderlyingValue(UV);
- return New;
+ return new VPWidenCastRecipe(
+ Opcode, getOperand(0), ResultTy,
+ dyn_cast_or_null<CastInst>(getUnderlyingValue()), *this, *this,
+ getDebugLoc());
}
VP_CLASSOF_IMPL(VPDef::VPWidenCastSC)
@@ -1577,13 +1565,17 @@ class VPWidenIntrinsicRecipe : public VPRecipeWithIRFlags, public VPIRMetadata {
public:
VPWidenIntrinsicRecipe(CallInst &CI, Intrinsic::ID VectorIntrinsicID,
ArrayRef<VPValue *> CallArguments, Type *Ty,
+ const VPIRFlags &Flags = {},
const VPIRMetadata &MD = {},
DebugLoc DL = DebugLoc::getUnknown())
- : VPRecipeWithIRFlags(VPDef::VPWidenIntrinsicSC, CallArguments, CI),
+ : VPRecipeWithIRFlags(VPDef::VPWidenIntrinsicSC, CallArguments, Flags,
+ DL),
VPIRMetadata(MD), VectorIntrinsicID(VectorIntrinsicID), ResultTy(Ty),
MayReadFromMemory(CI.mayReadFromMemory()),
MayWriteToMemory(CI.mayWriteToMemory()),
- MayHaveSideEffects(CI.mayHaveSideEffects()) {}
+ MayHaveSideEffects(CI.mayHaveSideEffects()) {
+ setUnderlyingValue(&CI);
+ }
VPWidenIntrinsicRecipe(Intrinsic::ID VectorIntrinsicID,
ArrayRef<VPValue *> CallArguments, Type *Ty,
@@ -1609,7 +1601,7 @@ class VPWidenIntrinsicRecipe : public VPRecipeWithIRFlags, public VPIRMetadata {
VPWidenIntrinsicRecipe *clone() override {
if (Value *CI = getUnderlyingValue())
return new VPWidenIntrinsicRecipe(*cast<CallInst>(CI), VectorIntrinsicID,
- operands(), ResultTy, *this,
+ operands(), ResultTy, *this, *this,
getDebugLoc());
return new VPWidenIntrinsicRecipe(VectorIntrinsicID, operands(), ResultTy,
*this, *this, getDebugLoc());
@@ -1663,10 +1655,11 @@ class LLVM_ABI_FOR_TEST VPWidenCallRecipe : public VPRecipeWithIRFlags,
public:
VPWidenCallRecipe(Value *UV, Function *Variant,
ArrayRef<VPValue *> CallArguments,
- DebugLoc DL = DebugLoc::getUnknown())
- : VPRecipeWithIRFlags(VPDef::VPWidenCallSC, CallArguments,
- *cast<Instruction>(UV)),
- VPIRMetadata(*cast<Instruction>(UV)), Variant(Variant) {
+ const VPIRFlags &Flags = {},
+ const VPIRMetadata &Metadata = {}, DebugLoc DL = {})
+ : VPRecipeWithIRFlags(VPDef::VPWidenCallSC, CallArguments, Flags, DL),
+ VPIRMetadata(Metadata), Variant(Variant) {
+ setUnderlyingValue(UV);
assert(
isa<Function>(getOperand(getNumOperands() - 1)->getLiveInIRValue()) &&
"last operand must be the called function");
@@ -1676,7 +1669,7 @@ class LLVM_ABI_FOR_TEST VPWidenCallRecipe : public VPRecipeWithIRFlags,
VPWidenCallRecipe *clone() override {
return new VPWidenCallRecipe(getUnderlyingValue(), Variant, operands(),
- getDebugLoc());
+ *this, *this, getDebugLoc());
}
VP_CLASSOF_IMPL(VPDef::VPWidenCallSC)
@@ -1753,16 +1746,19 @@ class VPHistogramRecipe : public VPRecipeBase {
/// instruction.
struct LLVM_ABI_FOR_TEST VPWidenSelectRecipe : public VPRecipeWithIRFlags,
public VPIRMetadata {
- VPWidenSelectRecipe(SelectInst &I, ArrayRef<VPValue *> Operands,
- const VPIRMetadata &MD = {})
- : VPRecipeWithIRFlags(VPDef::VPWidenSelectSC, Operands, I),
- VPIRMetadata(MD) {}
+ VPWidenSelectRecipe(SelectInst *SI, ArrayRef<VPValue *> Operands,
+ const VPIRFlags &Flags = {}, const VPIRMetadata &MD = {},
+ DebugLoc DL = {})
+ : VPRecipeWithIRFlags(VPDef::VPWidenSelectSC, Operands, Flags, DL),
+ VPIRMetadata(MD) {
+ setUnderlyingValue(SI);
+ }
~VPWidenSelectRecipe() override = default;
VPWidenSelectRecipe *clone() override {
- return new VPWidenSelectRecipe(*cast<SelectInst>(getUnderlyingInstr()),
- operands(), *this);
+ return new VPWidenSelectRecipe(cast<SelectInst>(getUnderlyingInstr()),
+ operands(), *this, *this, getDebugLoc());
}
VP_CLASSOF_IMPL(VPDef::VPWidenSelectSC)
@@ -1814,9 +1810,12 @@ class LLVM_ABI_FOR_TEST VPWidenGEPRecipe : public VPRecipeWithIRFlags {
}
public:
- VPWidenGEPRecipe(GetElementPtrInst *GEP, ArrayRef<VPValue *> Operands)
- : VPRecipeWithIRFlags(VPDef::VPWidenGEPSC, Operands, *GEP),
+ VPWidenGEPRecipe(GetElementPtrInst *GEP, ArrayRef<VPValue *> Operands,
+ const VPIRFlags &Flags = {},
+ DebugLoc DL = DebugLoc::getUnknown())
+ : VPRecipeWithIRFlags(VPDef::VPWidenGEPSC, Operands, Flags, DL),
SourceElementTy(GEP->getSourceElementType()) {
+ setUnderlyingValue(GEP);
SmallVector<std::pair<unsigned, MDNode *>> Metadata;
(void)Metadata;
getMetadataToPropagate(GEP, Metadata);
@@ -1827,7 +1826,7 @@ class LLVM_ABI_FOR_TEST VPWidenGEPRecipe : public VPRecipeWithIRFlags {
VPWidenGEPRecipe *clone() override {
return new VPWidenGEPRecipe(cast<GetElementPtrInst>(getUnderlyingInstr()),
- operands());
+ operands(), *this, getDebugLoc());
}
VP_CLASSOF_IMPL(VPDef::VPWidenGEPSC)
@@ -2921,10 +2920,12 @@ class LLVM_ABI_FOR_TEST VPReplicateRecipe : public VPRecipeWithIRFlags,
public:
VPReplicateRecipe(Instruction *I, ArrayRef<VPValue *> Operands,
bool IsSingleScalar, VPValue *Mask = nullptr,
- VPIRMetadata Metadata = {})
- : VPRecipeWithIRFlags(VPDef::VPReplicateSC, Operands, *I),
+ const VPIRFlags &Flags = {}, VPIRMetadata Metadata = {},
+ DebugLoc DL = DebugLoc::getUnknown())
+ : VPRecipeWithIRFlags(VPDef::VPReplicateSC, Operands, Flags, DL),
VPIRMetadata(Metadata), IsSingleScalar(IsSingleScalar),
IsPredicated(Mask) {
+ setUnderlyingValue(I);
if (Mask)
addOperand(Mask);
}
@@ -2932,9 +2933,9 @@ class LLVM_ABI_FOR_TEST VPReplicateRecipe : public VPRecipeWithIRFlags,
~VPReplicateRecipe() override = default;
VPReplicateRecipe *clone() override {
- auto *Copy =
- new VPReplicateRecipe(getUnderlyingInstr(), operands(), IsSingleScalar,
- isPredicated() ? getMask() : nullptr, *this);
+ auto *Copy = new VPReplicateRecipe(
+ getUnderlyingInstr(), operands(), IsSingleScalar,
+ isPredicated() ? getMask() : nullptr, *this, *this, getDebugLoc());
Copy->transferFlags(*this);
return Copy;
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
index 612202d049774..dbbde1cafa9f2 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
@@ -190,7 +190,7 @@ void PlainCFGBuilder::createVPInstructionsForVPBB(VPBasicBlock *VPBB,
// recipes.
if (Br->isConditional()) {
VPValue *Cond = getOrCreateVPOperand(Br->getCondition());
- VPIRBuilder.createNaryOp(VPInstruction::BranchOnCond, {Cond}, Inst,
+ VPIRBuilder.createNaryOp(VPInstruction::BranchOnCond, {Cond}, Inst, {},
VPIRMetadata(*Inst), Inst->getDebugLoc());
}
@@ -205,7 +205,7 @@ void PlainCFGBuilder::createVPInstructionsForVPBB(VPBasicBlock *VPBB,
SmallVector<VPValue *> Ops = {getOrCreateVPOperand(SI->getCondition())};
for (auto Case : SI->cases())
Ops.push_back(getOrCreateVPOperand(Case.getCaseValue()));
- VPIRBuilder.createNaryOp(Instruction::Switch, Ops, Inst,
+ VPIRBuilder.createNaryOp(Instruction::Switch, Ops, Inst, {},
VPIRMetadata(*Inst), Inst->getDebugLoc());
continue;
}
@@ -255,13 +255,14 @@ void PlainCFGBuilder::createVPInstructionsForVPBB(VPBasicBlock *VPBB,
if (auto *CI = dyn_cast<CastInst>(Inst)) {
NewR = VPIRBuilder.createScalarCast(CI->getOpcode(), VPOperands[0],
CI->getType(), CI->getDebugLoc(),
- {}, MD);
+ VPIRFlags(*CI), MD);
NewR->setUnderlyingValue(CI);
} else {
// Build VPInstruction for any arbitrary Instruction without specific
// representation in VPlan.
- NewR = VPIRBuilder.createNaryOp(Inst->getOpcode(), VPOperands, Inst, MD,
- Inst->getDebugLoc());
+ NewR =
+ VPIRBuilder.createNaryOp(Inst->getOpcode(), VPOperands, Inst,
+ VPIRFlags(*Inst), MD, Inst->getDebugLoc());
}
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index fca6554ad77c6..ef36e29aaa5c4 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -2056,24 +2056,26 @@ bool VPIRFlags::flagsValidForOpcode(unsigned Opcode) const {
switch (OpType) {
case OperationType::OverflowingBinOp:
return Opcode == Instruction::Add || Opcode == Instruction::Sub ||
- Opcode == Instruction::Mul ||
+ Opcode == Instruction::Mul || Opcode == Instruction::Shl ||
Opcode == VPInstruction::VPInstruction::CanonicalIVIncrementForPart;
case OperationType::Trunc:
return Opcode == Instruction::Trunc;
case OperationType::DisjointOp:
return Opcode == Instruction::Or;
case OperationType::PossiblyExactOp:
- return Opcode == Instruction::AShr;
+ return Opcode == Instruction::AShr || Opcode == Instruction::LShr ||
+ Opcode == Instruction::UDiv || Opcode == Instruction::SDiv;
case OperationType::GEPOp:
return Opcode == Instruction::GetElementPtr ||
Opcode == VPInstruction::PtrAdd ||
Opcode == VPInstruction::WidePtrAdd;
case OperationType::FPMathOp:
- return Opcode == Instruction::FAdd || Opcode == Instruction::FMul ||
- Opcode == Instruction::FSub || Opcode == Instruction::FNeg ||
- Opcode == Instruction::FDiv || Opcode == Instruction::FRem ||
- Opcode == Instruction::FPExt || Opcode == Instruction::FPTrunc ||
- Opcode == Instruction::FCmp || Opcode == Instruction::Select ||
+ return Opcode == Instruction::Call || Opcode == Instruction::FAdd ||
+ ...
[truncated]
|
🐧 Linux x64 Test Results
|
fhahn
changed the title
Update VPlan to populate VPIRFlags during VPInstruction construction and use it when creating widened recipes, instead of constructing VPIRFlags from the underlying IR instruction each time. The VPRecipeWithIRFlags constructor taking an underlying instruction and setting the flags based on it has been removed. This centralizes initial VPIRFlags creation and ensures flags are consistently available throughout VPlan transformations and makes sure we don't accidentally re-add flags from the underlying instruction that already got dropped during transformations. Follow-up to llvm#167253, which did the same for VPIRMetadata.
fhahn
force-pushed
the
vplan-recipe-irflags-early
branch
from
b80ad4d to
fb3bb39
Compare
lukel97
reviewed
Nov 18, 2025
| NewR = VPIRBuilder.createScalarCast(CI->getOpcode(), VPOperands[0], | ||
| CI->getType(), CI->getDebugLoc(), | ||
| {}, MD); | ||
| VPIRFlags(*CI), MD); |
Contributor
There was a problem hiding this comment.
If we're propagating more flags that weren't previously do we need to drop NFC from the PR title?
Contributor
Author
There was a problem hiding this comment.
It should be NFC w.r.t. to the generated IR, but it impacts the debug output. I removed NFC.
Contributor
There was a problem hiding this comment.
Oh I see that e.g. VPInstructionWithType::execute doesn't set the IR flags etc. I presume we should eventually do that in a separate PR?
fhahn
changed the title
lukel97
approved these changes
Nov 18, 2025
| NewR = VPIRBuilder.createScalarCast(CI->getOpcode(), VPOperands[0], | ||
| CI->getType(), CI->getDebugLoc(), | ||
| {}, MD); | ||
| VPIRFlags(*CI), MD); |
Contributor
There was a problem hiding this comment.
Oh I see that e.g. VPInstructionWithType::execute doesn't set the IR flags etc. I presume we should eventually do that in a separate PR?
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Nov 18, 2025
…ion. (#168450) Update VPlan to populate VPIRFlags during VPInstruction construction and use it when creating widened recipes, instead of constructing VPIRFlags from the underlying IR instruction each time. The VPRecipeWithIRFlags constructor taking an underlying instruction and setting the flags based on it has been removed. This centralizes initial VPIRFlags creation and ensures flags are consistently available throughout VPlan transformations and makes sure we don't accidentally re-add flags from the underlying instruction that already got dropped during transformations. Follow-up to llvm/llvm-project#167253, which did the same for VPIRMetadata. Should be NFC w.r.t. to the generated IR. PR: llvm/llvm-project#168450
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/92/builds/27667 Here is the relevant piece of the build log for the reference |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Update VPlan to populate VPIRFlags during VPInstruction construction and use it when creating widened recipes, instead of constructing VPIRFlags from the underlying IR instruction each time. The VPRecipeWithIRFlags constructor taking an underlying instruction and setting the flags based on it has been removed.
This centralizes initial VPIRFlags creation and ensures flags are consistently available throughout VPlan transformations and makes sure we don't accidentally re-add flags from the underlying instruction that already got dropped during transformations.
Follow-up to #167253, which did the same for VPIRMetadata.
Should be NFC w.r.t. to the generated IR.