This is an archive of the discontinued LLVM Phabricator instance.

Differential D144455

[RISCV] Enable ordered reduction.
ClosedPublic

Authored by Mel-Chen on Feb 21 2023, 1:01 AM.

Download Raw Diff

Details

Reviewers

craig.topper
reames
rogfer01
frasercrmck

Commits

rGd5c404d1b9ab: [RISCV] Enable ordered reduction.

Diff Detail

Repository: rG LLVM Github Monorepo

Event Timeline

Mel-Chen created this revision.Feb 21 2023, 1:01 AM

Herald added a project: Restricted Project. · View Herald TranscriptFeb 21 2023, 1:01 AM

Herald added subscribers: luke, VincentWu, vkmr and 28 others. · View Herald Transcript

Mel-Chen requested review of this revision.Feb 21 2023, 1:01 AM

Herald added a project: Restricted Project. · View Herald TranscriptFeb 21 2023, 1:01 AM

Herald added subscribers: llvm-commits, • pcwang-thead, eopXD, MaskRay. · View Herald Transcript

Harbormaster completed remote builds in B214945: Diff 499064.Feb 21 2023, 1:02 AM

Add the test case.

Create another NFC revision.

Mel-Chen added a parent revision: D144458: [RISCV] Pre-commit test case for ordered reduction, NFC.Feb 21 2023, 1:36 AM

Harbormaster completed remote builds in B214955: Diff 499076.Feb 21 2023, 1:36 AM

Mel-Chen added reviewers: craig.topper, reames.Feb 21 2023, 1:41 AM

Mel-Chen added reviewers: rogfer01, frasercrmck.Feb 21 2023, 1:52 AM

LGTM, but please remember to rebase over the change I asked for the underlying revision. After you do so, your test change should reduce to the default configuration using a different check prefix.

Update since parent revision.

Harbormaster completed remote builds in B215750: Diff 500188.Feb 24 2023, 10:19 AM

Rebase.

Harbormaster completed remote builds in B216709: Diff 501491.Mar 1 2023, 7:26 AM

LGTM

This revision is now accepted and ready to land.Mar 1 2023, 11:06 AM

Rebase and land.

This revision was landed with ongoing or failed builds.Mar 7 2023, 4:37 AM

Closed by commit rGd5c404d1b9ab: [RISCV] Enable ordered reduction. (authored by Mel-Chen). · Explain Why

This revision was automatically updated to reflect the committed changes.

Mel-Chen added a commit: rGd5c404d1b9ab: [RISCV] Enable ordered reduction..

Harbormaster completed remote builds in B217841: Diff 502987.Mar 7 2023, 5:47 AM

Revision Contents

Path

Size

llvm/

lib/

Target/

RISCV/

RISCVTargetTransformInfo.h

1 line

test/

Transforms/

LoopVectorize/

RISCV/

ordered-reduction.ll

2 lines

Diff 502988

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h

	Show First 20 Lines • Show All 69 Lines • ▼ Show 20 Lines		public:
	InstructionCost getIntImmCostIntrin(Intrinsic::ID IID, unsigned Idx,			InstructionCost getIntImmCostIntrin(Intrinsic::ID IID, unsigned Idx,
	const APInt &Imm, Type *Ty,			const APInt &Imm, Type *Ty,
	TTI::TargetCostKind CostKind);			TTI::TargetCostKind CostKind);

	TargetTransformInfo::PopcntSupportKind getPopcntSupport(unsigned TyWidth);			TargetTransformInfo::PopcntSupportKind getPopcntSupport(unsigned TyWidth);

	bool shouldExpandReduction(const IntrinsicInst *II) const;			bool shouldExpandReduction(const IntrinsicInst *II) const;
	bool supportsScalableVectors() const { return ST->hasVInstructions(); }			bool supportsScalableVectors() const { return ST->hasVInstructions(); }
				bool enableOrderedReductions() const { return true; }
	bool enableScalableVectorization() const { return ST->hasVInstructions(); }			bool enableScalableVectorization() const { return ST->hasVInstructions(); }
	TailFoldingStyle			TailFoldingStyle
	getPreferredTailFoldingStyle(bool IVUpdateMayOverflow) const {			getPreferredTailFoldingStyle(bool IVUpdateMayOverflow) const {
	return ST->hasVInstructions() ? TailFoldingStyle::Data			return ST->hasVInstructions() ? TailFoldingStyle::Data
	: TailFoldingStyle::DataWithoutLaneMask;			: TailFoldingStyle::DataWithoutLaneMask;
	}			}
	std::optional<unsigned> getMaxVScale() const;			std::optional<unsigned> getMaxVScale() const;
	std::optional<unsigned> getVScaleForTuning() const;			std::optional<unsigned> getVScaleForTuning() const;
	▲ Show 20 Lines • Show All 253 Lines • Show Last 20 Lines

llvm/test/Transforms/LoopVectorize/RISCV/ordered-reduction.ll

	; NOTE: Assertions have been autogenerated by utils/update_test_checks.py			; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
	; RUN: opt -passes=loop-vectorize -hints-allow-reordering=false -S < %s \| FileCheck %s --check-prefix=CHECK-NOT-VECTORIZED			; RUN: opt -passes=loop-vectorize -hints-allow-reordering=false -S < %s \| FileCheck %s --check-prefix=CHECK-ORDERED
	; RUN: opt -passes=loop-vectorize -S < %s \| FileCheck %s --check-prefix=CHECK-UNORDERED			; RUN: opt -passes=loop-vectorize -S < %s \| FileCheck %s --check-prefix=CHECK-UNORDERED
	; RUN: opt -passes=loop-vectorize -force-ordered-reductions=false -hints-allow-reordering=false -S < %s \| FileCheck %s --check-prefix=CHECK-NOT-VECTORIZED			; RUN: opt -passes=loop-vectorize -force-ordered-reductions=false -hints-allow-reordering=false -S < %s \| FileCheck %s --check-prefix=CHECK-NOT-VECTORIZED
	; RUN: opt -passes=loop-vectorize -force-ordered-reductions=true -hints-allow-reordering=false -S < %s \| FileCheck %s --check-prefix=CHECK-ORDERED			; RUN: opt -passes=loop-vectorize -force-ordered-reductions=true -hints-allow-reordering=false -S < %s \| FileCheck %s --check-prefix=CHECK-ORDERED


	target triple = "riscv64"			target triple = "riscv64"

	define float @fadd(ptr noalias nocapture readonly %a, i64 %n) #0 {			define float @fadd(ptr noalias nocapture readonly %a, i64 %n) #0 {
	▲ Show 20 Lines • Show All 116 Lines • Show Last 20 Lines