This is an archive of the discontinued LLVM Phabricator instance.

Differential D103627

[Matrix] Fix transpose-multiply folding if transpose has multiple uses
ClosedPublic

Authored by anemet on Jun 3 2021, 9:10 AM.

Details

Reviewers
fhahn
Commits
rGffde966cd9bf: [Matrix] Fix transpose-multiply folding if transpose has multiple uses
Summary

Don't add it to FusedInsts in this case.

Diff Detail

Unit TestsFailed

TimeTest
140 msx64 debian > LLVM.Examples/OrcV2Examples::lljit-with-remote-debugging.test

Event Timeline

anemet created this revision.Jun 3 2021, 9:10 AM
Herald added subscribers: tschuett, hiraditya. · View Herald TranscriptJun 3 2021, 9:10 AM
anemet requested review of this revision.Jun 3 2021, 9:10 AM
Herald added a project: Restricted Project. · View Herald TranscriptJun 3 2021, 9:10 AM
Herald added a subscriber: llvm-commits. · View Herald Transcript
Harbormaster completed remote builds in B107473: Diff 349576.Jun 3 2021, 10:00 AM
fhahn accepted this revision.Jun 4 2021, 10:12 AM

LGTM, thanks!

This revision is now accepted and ready to land.Jun 4 2021, 10:12 AM
This revision was landed with ongoing or failed builds.Jun 4 2021, 10:56 AM
Closed by commit rGffde966cd9bf: [Matrix] Fix transpose-multiply folding if transpose has multiple uses (authored by anemet). · Explain Why
This revision was automatically updated to reflect the committed changes.
anemet added a commit: rGffde966cd9bf: [Matrix] Fix transpose-multiply folding if transpose has multiple uses.

Revision Contents

PathSize
llvm/
lib/
Transforms/
Scalar/
5 lines
test/
Transforms/
LowerMatrixIntrinsics/
120 lines

Diff 349576

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

Show First 20 Lines • Show All 1,531 Lines • ▼ Show 20 Lines if (MatrixLayout == MatrixLayoutTy::ColumnMajor
// Initialize the output // Initialize the output
MatrixTy Result(R, C, EltType); MatrixTy Result(R, C, EltType);
emitMatrixMultiply(Result, MA, MB, Builder, false, true, emitMatrixMultiply(Result, MA, MB, Builder, false, true,
getFastMathFlags(MatMul)); getFastMathFlags(MatMul));
FusedInsts.insert(MatMul); FusedInsts.insert(MatMul);
if (Transpose->hasOneUse()) {
FusedInsts.insert(cast<Instruction>(Transpose)); FusedInsts.insert(cast<Instruction>(Transpose));
if (Transpose->hasOneUse())
ToRemove.push_back(cast<Instruction>(Transpose)); ToRemove.push_back(cast<Instruction>(Transpose));
}
finalizeLowering(MatMul, Result, Builder); finalizeLowering(MatMul, Result, Builder);
// TODO: add a fake entry for the folded instruction so that this is // TODO: add a fake entry for the folded instruction so that this is
// included in the expression in the remark. // included in the expression in the remark.
Inst2ColumnMatrix[Transpose] = MatrixTy(M, C, EltType); Inst2ColumnMatrix[Transpose] = MatrixTy(M, C, EltType);
return; return;
} }
if (!MatMul->hasOneUse() || MatrixLayout != MatrixLayoutTy::ColumnMajor) if (!MatMul->hasOneUse() || MatrixLayout != MatrixLayoutTy::ColumnMajor)
▲ Show 20 Lines • Show All 749 Lines • Show Last 20 Lines

llvm/test/Transforms/LowerMatrixIntrinsics/multiply-right-transpose.ll

Show First 20 Lines • Show All 64 Lines • ▼ Show 20 Lines
; CHECK-NEXT: ret <4 x double> [[TMP28]] ; CHECK-NEXT: ret <4 x double> [[TMP28]]
; ;
entry: entry:
%bt = call <4 x double> @llvm.matrix.transpose.v4f64(<4 x double> %b, i32 2, i32 2) %bt = call <4 x double> @llvm.matrix.transpose.v4f64(<4 x double> %b, i32 2, i32 2)
%c = call <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double> %a, <4 x double> %bt, i32 2, i32 2, i32 2) %c = call <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double> %a, <4 x double> %bt, i32 2, i32 2, i32 2)
ret <4 x double> %c ret <4 x double> %c
} }
; Lower transpose if not all uses are folded.
define <4 x double> @multiply_right_transpose_2x3x2(<6 x double> %a, <6 x double> %b, <6 x double>* %p) {
; CHECK-LABEL: @multiply_right_transpose_2x3x2(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[SPLIT39:%.*]] = shufflevector <6 x double> [[B:%.*]], <6 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[SPLIT40:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 2, i32 3>
; CHECK-NEXT: [[SPLIT41:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 4, i32 5>
; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT39]], i64 0
; CHECK-NEXT: [[TMP1:%.*]] = insertelement <3 x double> undef, double [[TMP0]], i64 0
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT40]], i64 0
; CHECK-NEXT: [[TMP3:%.*]] = insertelement <3 x double> [[TMP1]], double [[TMP2]], i64 1
; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT41]], i64 0
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <3 x double> [[TMP3]], double [[TMP4]], i64 2
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT39]], i64 1
; CHECK-NEXT: [[TMP7:%.*]] = insertelement <3 x double> undef, double [[TMP6]], i64 0
; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT40]], i64 1
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <3 x double> [[TMP7]], double [[TMP8]], i64 1
; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT41]], i64 1
; CHECK-NEXT: [[TMP11:%.*]] = insertelement <3 x double> [[TMP9]], double [[TMP10]], i64 2
; CHECK-NEXT: [[TMP12:%.*]] = bitcast <6 x double>* [[P:%.*]] to double*
; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast double* [[TMP12]] to <3 x double>*
; CHECK-NEXT: store <3 x double> undef, <3 x double>* [[VEC_CAST]], align 16
; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, double* [[TMP12]], i64 3
; CHECK-NEXT: [[VEC_CAST42:%.*]] = bitcast double* [[VEC_GEP]] to <3 x double>*
; CHECK-NEXT: store <3 x double> undef, <3 x double>* [[VEC_CAST42]], align 8
; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <6 x double> [[A:%.*]], <6 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <6 x double> [[A]], <6 x double> poison, <2 x i32> <i32 2, i32 3>
; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <6 x double> [[A]], <6 x double> poison, <2 x i32> <i32 4, i32 5>
; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 0, i32 1>
; CHECK-NEXT: [[SPLIT4:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 2, i32 3>
; CHECK-NEXT: [[SPLIT5:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 4, i32 5>
; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x double> poison, double [[TMP13]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP14:%.*]] = fmul <1 x double> [[BLOCK]], [[SPLAT_SPLAT]]
; CHECK-NEXT: [[BLOCK6:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT7:%.*]] = insertelement <1 x double> poison, double [[TMP15]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT8:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT7]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP16:%.*]] = fmul <1 x double> [[BLOCK6]], [[SPLAT_SPLAT8]]
; CHECK-NEXT: [[TMP17:%.*]] = fadd <1 x double> [[TMP14]], [[TMP16]]
; CHECK-NEXT: [[BLOCK9:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP18:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT10:%.*]] = insertelement <1 x double> poison, double [[TMP18]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT11:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT10]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP19:%.*]] = fmul <1 x double> [[BLOCK9]], [[SPLAT_SPLAT11]]
; CHECK-NEXT: [[TMP20:%.*]] = fadd <1 x double> [[TMP17]], [[TMP19]]
; CHECK-NEXT: [[TMP21:%.*]] = shufflevector <1 x double> [[TMP20]], <1 x double> poison, <2 x i32> <i32 0, i32 undef>
; CHECK-NEXT: [[TMP22:%.*]] = shufflevector <2 x double> undef, <2 x double> [[TMP21]], <2 x i32> <i32 2, i32 1>
; CHECK-NEXT: [[BLOCK12:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> <i32 1>
; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT13:%.*]] = insertelement <1 x double> poison, double [[TMP23]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT14:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT13]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP24:%.*]] = fmul <1 x double> [[BLOCK12]], [[SPLAT_SPLAT14]]
; CHECK-NEXT: [[BLOCK15:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> <i32 1>
; CHECK-NEXT: [[TMP25:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT16:%.*]] = insertelement <1 x double> poison, double [[TMP25]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT17:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT16]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP26:%.*]] = fmul <1 x double> [[BLOCK15]], [[SPLAT_SPLAT17]]
; CHECK-NEXT: [[TMP27:%.*]] = fadd <1 x double> [[TMP24]], [[TMP26]]
; CHECK-NEXT: [[BLOCK18:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> <i32 1>
; CHECK-NEXT: [[TMP28:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 0
; CHECK-NEXT: [[SPLAT_SPLATINSERT19:%.*]] = insertelement <1 x double> poison, double [[TMP28]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT20:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT19]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP29:%.*]] = fmul <1 x double> [[BLOCK18]], [[SPLAT_SPLAT20]]
; CHECK-NEXT: [[TMP30:%.*]] = fadd <1 x double> [[TMP27]], [[TMP29]]
; CHECK-NEXT: [[TMP31:%.*]] = shufflevector <1 x double> [[TMP30]], <1 x double> poison, <2 x i32> <i32 0, i32 undef>
; CHECK-NEXT: [[TMP32:%.*]] = shufflevector <2 x double> [[TMP22]], <2 x double> [[TMP31]], <2 x i32> <i32 0, i32 2>
; CHECK-NEXT: [[BLOCK21:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP33:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT22:%.*]] = insertelement <1 x double> poison, double [[TMP33]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT23:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT22]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP34:%.*]] = fmul <1 x double> [[BLOCK21]], [[SPLAT_SPLAT23]]
; CHECK-NEXT: [[BLOCK24:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP35:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT25:%.*]] = insertelement <1 x double> poison, double [[TMP35]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT26:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT25]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP36:%.*]] = fmul <1 x double> [[BLOCK24]], [[SPLAT_SPLAT26]]
; CHECK-NEXT: [[TMP37:%.*]] = fadd <1 x double> [[TMP34]], [[TMP36]]
; CHECK-NEXT: [[BLOCK27:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP38:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT28:%.*]] = insertelement <1 x double> poison, double [[TMP38]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT29:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT28]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP39:%.*]] = fmul <1 x double> [[BLOCK27]], [[SPLAT_SPLAT29]]
; CHECK-NEXT: [[TMP40:%.*]] = fadd <1 x double> [[TMP37]], [[TMP39]]
; CHECK-NEXT: [[TMP41:%.*]] = shufflevector <1 x double> [[TMP40]], <1 x double> poison, <2 x i32> <i32 0, i32 undef>
; CHECK-NEXT: [[TMP42:%.*]] = shufflevector <2 x double> undef, <2 x double> [[TMP41]], <2 x i32> <i32 2, i32 1>
; CHECK-NEXT: [[BLOCK30:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> <i32 1>
; CHECK-NEXT: [[TMP43:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT31:%.*]] = insertelement <1 x double> poison, double [[TMP43]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT32:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT31]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP44:%.*]] = fmul <1 x double> [[BLOCK30]], [[SPLAT_SPLAT32]]
; CHECK-NEXT: [[BLOCK33:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> <i32 1>
; CHECK-NEXT: [[TMP45:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT34:%.*]] = insertelement <1 x double> poison, double [[TMP45]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT35:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT34]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP46:%.*]] = fmul <1 x double> [[BLOCK33]], [[SPLAT_SPLAT35]]
; CHECK-NEXT: [[TMP47:%.*]] = fadd <1 x double> [[TMP44]], [[TMP46]]
; CHECK-NEXT: [[BLOCK36:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> <i32 1>
; CHECK-NEXT: [[TMP48:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 1
; CHECK-NEXT: [[SPLAT_SPLATINSERT37:%.*]] = insertelement <1 x double> poison, double [[TMP48]], i32 0
; CHECK-NEXT: [[SPLAT_SPLAT38:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT37]], <1 x double> poison, <1 x i32> zeroinitializer
; CHECK-NEXT: [[TMP49:%.*]] = fmul <1 x double> [[BLOCK36]], [[SPLAT_SPLAT38]]
; CHECK-NEXT: [[TMP50:%.*]] = fadd <1 x double> [[TMP47]], [[TMP49]]
; CHECK-NEXT: [[TMP51:%.*]] = shufflevector <1 x double> [[TMP50]], <1 x double> poison, <2 x i32> <i32 0, i32 undef>
; CHECK-NEXT: [[TMP52:%.*]] = shufflevector <2 x double> [[TMP42]], <2 x double> [[TMP51]], <2 x i32> <i32 0, i32 2>
; CHECK-NEXT: [[TMP53:%.*]] = shufflevector <2 x double> [[TMP32]], <2 x double> [[TMP52]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; CHECK-NEXT: ret <4 x double> [[TMP53]]
;
entry:
%bt = call <6 x double> @llvm.matrix.transpose.v6f64(<6 x double> %b, i32 2, i32 3)
store <6 x double> %bt, <6 x double>* %p, align 16
%c = call <4 x double> @llvm.matrix.multiply.v4f64.v6f64.v6f64(<6 x double> %a, <6 x double> %bt, i32 2, i32 3, i32 2)
ret <4 x double> %c
}
declare <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double>, <4 x double>, i32, i32, i32) declare <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double>, <4 x double>, i32, i32, i32)
declare <4 x double> @llvm.matrix.multiply.v4f64.v6f64.v6f64(<6 x double>, <6 x double>, i32, i32, i32)
declare <4 x double> @llvm.matrix.transpose.v4f64(<4 x double>, i32, i32) declare <4 x double> @llvm.matrix.transpose.v4f64(<4 x double>, i32, i32)
declare <6 x double> @llvm.matrix.transpose.v6f64(<6 x double>, i32, i32)