This is an archive of the discontinued LLVM Phabricator instance.

Differential D118239

[Matrix] Use ArrayType for allocas instead of VectorType.
ClosedPublic

Authored by fhahn on Jan 26 2022, 4:18 AM.

Details

Reviewers
anemet
thegameg
Commits
rGb339bbdb1970: [Matrix] Use ArrayType for allocas instead of VectorType.
Summary

When creating an alloca to copy a matrix due to memory conflicts, those
allocas used to use VectorTypes, which forced them to have huge
alignments for large vectors.

This patch updates LowerMatrixIntrinsics to use a corresponding array
type, like Clang already does, to get more manageable alignments.

Diff Detail

Event Timeline

fhahn created this revision.Jan 26 2022, 4:18 AM
Herald added subscribers: tschuett, hiraditya. · View Herald TranscriptJan 26 2022, 4:18 AM
fhahn updated this revision to Diff 403218.Jan 26 2022, 4:32 AM

Update to remove accidentially set 'Draft' status

fhahn published this revision for review.Jan 26 2022, 4:33 AM
Herald added a project: Restricted Project. · View Herald TranscriptJan 26 2022, 4:33 AM
Harbormaster completed remote builds in B145721: Diff 403218.Jan 27 2022, 1:38 AM
anemet accepted this revision.Jan 27 2022, 9:05 AM

LGTM

This revision is now accepted and ready to land.Jan 27 2022, 9:05 AM
thegameg accepted this revision.Jan 27 2022, 1:06 PM

LGTM

This revision was landed with ongoing or failed builds.Jan 28 2022, 2:48 AM
Closed by commit rGb339bbdb1970: [Matrix] Use ArrayType for allocas instead of VectorType. (authored by fhahn). · Explain Why
This revision was automatically updated to reflect the committed changes.
fhahn added a commit: rGb339bbdb1970: [Matrix] Use ArrayType for allocas instead of VectorType..

Revision Contents

PathSize
llvm/
lib/
Transforms/
Scalar/
17 lines
test/
Transforms/
LowerMatrixIntrinsics/
12 lines
8 lines
36 lines
8 lines

Diff 403941

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

Show First 20 Lines • Show All 1,333 Lines • ▼ Show 20 Lines Value *getNonAliasingPointer(LoadInst *Load, StoreInst *Store,
Value *LoadEnd = Builder.CreateAdd( Value *LoadEnd = Builder.CreateAdd(
LoadBegin, ConstantInt::get(IntPtrTy, LoadLoc.Size.getValue()), LoadBegin, ConstantInt::get(IntPtrTy, LoadLoc.Size.getValue()),
"load.end", true, true); "load.end", true, true);
Builder.CreateCondBr(Builder.CreateICmpULT(StoreBegin, LoadEnd), Copy, Builder.CreateCondBr(Builder.CreateICmpULT(StoreBegin, LoadEnd), Copy,
Fusion); Fusion);
// Copy load operand to new alloca. // Copy load operand to new alloca.
Builder.SetInsertPoint(Copy, Copy->begin()); Builder.SetInsertPoint(Copy, Copy->begin());
AllocaInst *NewLd = auto *VT = cast<FixedVectorType>(Load->getType());
Builder.CreateAlloca(Load->getType(), Load->getPointerAddressSpace()); // Use an array type for the alloca, to avoid potentially huge alignment
Builder.CreateMemCpy(NewLd, NewLd->getAlign(), // requirements for large vector types.
Load->getPointerOperand(), Load->getAlign(), auto *ArrayTy = ArrayType::get(VT->getElementType(), VT->getNumElements());
LoadLoc.Size.getValue()); AllocaInst *Alloca =
Builder.CreateAlloca(ArrayTy, Load->getPointerAddressSpace());
Value *BC = Builder.CreateBitCast(Alloca, VT->getPointerTo());
Builder.CreateMemCpy(BC, Alloca->getAlign(), Load->getPointerOperand(),
Load->getAlign(), LoadLoc.Size.getValue());
Builder.SetInsertPoint(Fusion, Fusion->begin()); Builder.SetInsertPoint(Fusion, Fusion->begin());
PHINode *PHI = Builder.CreatePHI(Load->getPointerOperandType(), 3); PHINode *PHI = Builder.CreatePHI(Load->getPointerOperandType(), 3);
PHI->addIncoming(Load->getPointerOperand(), Check0); PHI->addIncoming(Load->getPointerOperand(), Check0);
PHI->addIncoming(Load->getPointerOperand(), Check1); PHI->addIncoming(Load->getPointerOperand(), Check1);
PHI->addIncoming(NewLd, Copy); PHI->addIncoming(BC, Copy);
// Adjust DT. // Adjust DT.
DTUpdates.push_back({DT->Insert, Check0, Check1}); DTUpdates.push_back({DT->Insert, Check0, Check1});
DTUpdates.push_back({DT->Insert, Check0, Fusion}); DTUpdates.push_back({DT->Insert, Check0, Fusion});
DTUpdates.push_back({DT->Insert, Check1, Copy}); DTUpdates.push_back({DT->Insert, Check1, Copy});
DTUpdates.push_back({DT->Insert, Check1, Fusion}); DTUpdates.push_back({DT->Insert, Check1, Fusion});
DT->applyUpdates(DTUpdates); DT->applyUpdates(DTUpdates);
return PHI; return PHI;
▲ Show 20 Lines • Show All 1,005 Lines • Show Last 20 Lines

llvm/test/Transforms/LowerMatrixIntrinsics/multiply-fused-dominance.ll

Show All 14 Lines
; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]] ; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]] ; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK: alias_cont: ; CHECK: alias_cont:
; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32 ; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]] ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]] ; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK: copy: ; CHECK: copy:
; CHECK-NEXT: [[TMP2:%.*]] = alloca <4 x double>, align 32 ; CHECK-NEXT: [[TMP2:%.*]] = alloca <4 x double>, align 8
; CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8* ; CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8*
; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8* ; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 32 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS]] ; CHECK-NEXT: br label [[NO_ALIAS]]
; CHECK: no_alias: ; CHECK: no_alias:
; CHECK-NEXT: [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ] ; CHECK-NEXT: [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ]
; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>* ; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>*
; CHECK-NEXT: [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT: [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>* ; CHECK-NEXT: [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT: [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8 ; CHECK-NEXT: [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8
; CHECK-NEXT: [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]] ; CHECK-NEXT: [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]]
▲ Show 20 Lines • Show All 75 Lines • ▼ Show 20 Lines
; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]] ; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]] ; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK: alias_cont: ; CHECK: alias_cont:
; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32 ; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]] ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]] ; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK: copy: ; CHECK: copy:
; CHECK-NEXT: [[TMP2:%.*]] = alloca <4 x double>, align 32 ; CHECK-NEXT: [[TMP2:%.*]] = alloca <4 x double>, align 8
; CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8* ; CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8*
; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8* ; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 32 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS]] ; CHECK-NEXT: br label [[NO_ALIAS]]
; CHECK: no_alias: ; CHECK: no_alias:
; CHECK-NEXT: [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ] ; CHECK-NEXT: [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ]
; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>* ; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>*
; CHECK-NEXT: [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT: [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>* ; CHECK-NEXT: [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT: [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8 ; CHECK-NEXT: [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8
; CHECK-NEXT: [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]] ; CHECK-NEXT: [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]]
▲ Show 20 Lines • Show All 77 Lines • ▼ Show 20 Lines
; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <4 x double>* [[B:%.*]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]] ; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]] ; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK: alias_cont: ; CHECK: alias_cont:
; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32 ; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 32
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]] ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]] ; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK: copy: ; CHECK: copy:
; CHECK-NEXT: [[TMP2:%.*]] = alloca <4 x double>, align 32 ; CHECK-NEXT: [[TMP2:%.*]] = alloca <4 x double>, align 8
; CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8* ; CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x double>* [[TMP2]] to i8*
; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8* ; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x double>* [[B]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 32 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(32) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(32) [[TMP4]], i64 32, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS]] ; CHECK-NEXT: br label [[NO_ALIAS]]
; CHECK: no_alias: ; CHECK: no_alias:
; CHECK-NEXT: [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ] ; CHECK-NEXT: [[TMP5:%.*]] = phi <4 x double>* [ [[B]], [[ENTRY:%.*]] ], [ [[B]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ]
; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>* ; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <4 x double>* [[A:%.*]] to <1 x double>*
; CHECK-NEXT: [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT: [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>* ; CHECK-NEXT: [[VEC_CAST2:%.*]] = bitcast <4 x double>* [[TMP5]] to <1 x double>*
; CHECK-NEXT: [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8 ; CHECK-NEXT: [[COL_LOAD3:%.*]] = load <1 x double>, <1 x double>* [[VEC_CAST2]], align 8
; CHECK-NEXT: [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]] ; CHECK-NEXT: [[TMP6:%.*]] = fmul contract <1 x double> [[COL_LOAD]], [[COL_LOAD3]]
▲ Show 20 Lines • Show All 245 Lines • Show Last 20 Lines

llvm/test/Transforms/LowerMatrixIntrinsics/multiply-fused-loops.ll

Show First 20 Lines • Show All 280 Lines • ▼ Show 20 Lines
; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <4 x float>* [[A:%.*]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <4 x float>* [[A:%.*]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]] ; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]] ; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK: alias_cont: ; CHECK: alias_cont:
; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 16 ; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 16
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]] ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]] ; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK: copy: ; CHECK: copy:
; CHECK-NEXT: [[TMP2:%.*]] = alloca <4 x float>, align 16 ; CHECK-NEXT: [[TMP2:%.*]] = alloca <4 x float>, align 4
; CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x float>* [[TMP2]] to i8* ; CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x float>* [[TMP2]] to i8*
; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x float>* [[A]] to i8* ; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x float>* [[A]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(16) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(16) [[TMP4]], i64 16, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 4 dereferenceable(16) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(16) [[TMP4]], i64 16, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS]] ; CHECK-NEXT: br label [[NO_ALIAS]]
; CHECK: no_alias: ; CHECK: no_alias:
; CHECK-NEXT: [[TMP5:%.*]] = phi <4 x float>* [ [[A]], [[ENTRY:%.*]] ], [ [[A]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ] ; CHECK-NEXT: [[TMP5:%.*]] = phi <4 x float>* [ [[A]], [[ENTRY:%.*]] ], [ [[A]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ]
; CHECK-NEXT: [[STORE_BEGIN4:%.*]] = ptrtoint <4 x float>* [[C]] to i64 ; CHECK-NEXT: [[STORE_BEGIN4:%.*]] = ptrtoint <4 x float>* [[C]] to i64
; CHECK-NEXT: [[STORE_END5:%.*]] = add nuw nsw i64 [[STORE_BEGIN4]], 16 ; CHECK-NEXT: [[STORE_END5:%.*]] = add nuw nsw i64 [[STORE_BEGIN4]], 16
; CHECK-NEXT: [[LOAD_BEGIN6:%.*]] = ptrtoint <4 x float>* [[B:%.*]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN6:%.*]] = ptrtoint <4 x float>* [[B:%.*]] to i64
; CHECK-NEXT: [[TMP6:%.*]] = icmp ugt i64 [[STORE_END5]], [[LOAD_BEGIN6]] ; CHECK-NEXT: [[TMP6:%.*]] = icmp ugt i64 [[STORE_END5]], [[LOAD_BEGIN6]]
; CHECK-NEXT: br i1 [[TMP6]], label [[ALIAS_CONT1:%.*]], label [[NO_ALIAS3:%.*]] ; CHECK-NEXT: br i1 [[TMP6]], label [[ALIAS_CONT1:%.*]], label [[NO_ALIAS3:%.*]]
; CHECK: alias_cont1: ; CHECK: alias_cont1:
; CHECK-NEXT: [[LOAD_END7:%.*]] = add nuw nsw i64 [[LOAD_BEGIN6]], 16 ; CHECK-NEXT: [[LOAD_END7:%.*]] = add nuw nsw i64 [[LOAD_BEGIN6]], 16
; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i64 [[LOAD_END7]], [[STORE_BEGIN4]] ; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i64 [[LOAD_END7]], [[STORE_BEGIN4]]
; CHECK-NEXT: br i1 [[TMP7]], label [[COPY2:%.*]], label [[NO_ALIAS3]] ; CHECK-NEXT: br i1 [[TMP7]], label [[COPY2:%.*]], label [[NO_ALIAS3]]
; CHECK: copy2: ; CHECK: copy2:
; CHECK-NEXT: [[TMP8:%.*]] = alloca <4 x float>, align 16 ; CHECK-NEXT: [[TMP8:%.*]] = alloca <4 x float>, align 4
; CHECK-NEXT: [[TMP9:%.*]] = bitcast <4 x float>* [[TMP8]] to i8* ; CHECK-NEXT: [[TMP9:%.*]] = bitcast <4 x float>* [[TMP8]] to i8*
; CHECK-NEXT: [[TMP10:%.*]] = bitcast <4 x float>* [[B]] to i8* ; CHECK-NEXT: [[TMP10:%.*]] = bitcast <4 x float>* [[B]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(16) [[TMP9]], i8* noundef nonnull align 8 dereferenceable(16) [[TMP10]], i64 16, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 4 dereferenceable(16) [[TMP9]], i8* noundef nonnull align 8 dereferenceable(16) [[TMP10]], i64 16, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS3]] ; CHECK-NEXT: br label [[NO_ALIAS3]]
; CHECK: no_alias3: ; CHECK: no_alias3:
; CHECK-NEXT: [[TMP11:%.*]] = phi <4 x float>* [ [[B]], [[NO_ALIAS]] ], [ [[B]], [[ALIAS_CONT1]] ], [ [[TMP8]], [[COPY2]] ] ; CHECK-NEXT: [[TMP11:%.*]] = phi <4 x float>* [ [[B]], [[NO_ALIAS]] ], [ [[B]], [[ALIAS_CONT1]] ], [ [[TMP8]], [[COPY2]] ]
; CHECK-NEXT: br label [[COLS_HEADER:%.*]] ; CHECK-NEXT: br label [[COLS_HEADER:%.*]]
; CHECK: cols.header: ; CHECK: cols.header:
; CHECK-NEXT: [[COLS_IV:%.*]] = phi i64 [ 0, [[NO_ALIAS3]] ], [ [[COLS_STEP:%.*]], [[COLS_LATCH:%.*]] ] ; CHECK-NEXT: [[COLS_IV:%.*]] = phi i64 [ 0, [[NO_ALIAS3]] ], [ [[COLS_STEP:%.*]], [[COLS_LATCH:%.*]] ]
; CHECK-NEXT: br label [[COLS_BODY:%.*]] ; CHECK-NEXT: br label [[COLS_BODY:%.*]]
; CHECK: cols.body: ; CHECK: cols.body:
▲ Show 20 Lines • Show All 80 Lines • Show Last 20 Lines

llvm/test/Transforms/LowerMatrixIntrinsics/multiply-fused-multiple-blocks.ll

Show All 26 Lines
; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <6 x double>* [[A]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <6 x double>* [[A]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]] ; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]] ; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK: alias_cont: ; CHECK: alias_cont:
; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 48 ; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 48
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]] ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]] ; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK: copy: ; CHECK: copy:
; CHECK-NEXT: [[TMP2:%.*]] = alloca <6 x double>, align 64 ; CHECK-NEXT: [[TMP2:%.*]] = alloca [6 x double], align 8
; CHECK-NEXT: [[TMP3:%.*]] = bitcast <6 x double>* [[TMP2]] to i8* ; CHECK-NEXT: [[BC:%.+]] = bitcast [6 x double]* [[TMP2]] to <6 x double>*
; CHECK-NEXT: [[TMP3:%.*]] = bitcast [6 x double]* [[TMP2]] to i8*
; CHECK-NEXT: [[TMP4:%.*]] = bitcast <6 x double>* [[A]] to i8* ; CHECK-NEXT: [[TMP4:%.*]] = bitcast <6 x double>* [[A]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 64 dereferenceable(48) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(48) [[TMP4]], i64 48, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(48) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(48) [[TMP4]], i64 48, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS]] ; CHECK-NEXT: br label [[NO_ALIAS]]
; CHECK: no_alias: ; CHECK: no_alias:
; CHECK-NEXT: [[TMP5:%.*]] = phi <6 x double>* [ [[A]], [[ENTRY:%.*]] ], [ [[A]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ] ; CHECK-NEXT: [[TMP5:%.*]] = phi <6 x double>* [ [[A]], [[ENTRY:%.*]] ], [ [[A]], [[ALIAS_CONT]] ], [ [[BC]], [[COPY]] ]
; CHECK-NEXT: [[STORE_BEGIN4:%.*]] = ptrtoint <9 x double>* [[C]] to i64 ; CHECK-NEXT: [[STORE_BEGIN4:%.*]] = ptrtoint <9 x double>* [[C]] to i64
; CHECK-NEXT: [[STORE_END5:%.*]] = add nuw nsw i64 [[STORE_BEGIN4]], 72 ; CHECK-NEXT: [[STORE_END5:%.*]] = add nuw nsw i64 [[STORE_BEGIN4]], 72
; CHECK-NEXT: [[LOAD_BEGIN6:%.*]] = ptrtoint <6 x double>* [[B]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN6:%.*]] = ptrtoint <6 x double>* [[B]] to i64
; CHECK-NEXT: [[TMP6:%.*]] = icmp ugt i64 [[STORE_END5]], [[LOAD_BEGIN6]] ; CHECK-NEXT: [[TMP6:%.*]] = icmp ugt i64 [[STORE_END5]], [[LOAD_BEGIN6]]
; CHECK-NEXT: br i1 [[TMP6]], label [[ALIAS_CONT1:%.*]], label [[NO_ALIAS3:%.*]] ; CHECK-NEXT: br i1 [[TMP6]], label [[ALIAS_CONT1:%.*]], label [[NO_ALIAS3:%.*]]
; CHECK: alias_cont1: ; CHECK: alias_cont1:
; CHECK-NEXT: [[LOAD_END7:%.*]] = add nuw nsw i64 [[LOAD_BEGIN6]], 48 ; CHECK-NEXT: [[LOAD_END7:%.*]] = add nuw nsw i64 [[LOAD_BEGIN6]], 48
; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i64 [[LOAD_END7]], [[STORE_BEGIN4]] ; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i64 [[LOAD_END7]], [[STORE_BEGIN4]]
; CHECK-NEXT: br i1 [[TMP7]], label [[COPY2:%.*]], label [[NO_ALIAS3]] ; CHECK-NEXT: br i1 [[TMP7]], label [[COPY2:%.*]], label [[NO_ALIAS3]]
; CHECK: copy2: ; CHECK: copy2:
; CHECK-NEXT: [[TMP8:%.*]] = alloca <6 x double>, align 64 ; CHECK-NEXT: [[TMP8:%.*]] = alloca [6 x double], align 8
; CHECK-NEXT: [[TMP9:%.*]] = bitcast <6 x double>* [[TMP8]] to i8* ; CHECK-NEXT: [[BC2:%.+]] = bitcast [6 x double]* [[TMP8]] to <6 x double>*
; CHECK-NEXT: [[TMP9:%.*]] = bitcast [6 x double]* [[TMP8]] to i8*
; CHECK-NEXT: [[TMP10:%.*]] = bitcast <6 x double>* [[B]] to i8* ; CHECK-NEXT: [[TMP10:%.*]] = bitcast <6 x double>* [[B]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 64 dereferenceable(48) [[TMP9]], i8* noundef nonnull align 8 dereferenceable(48) [[TMP10]], i64 48, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(48) [[TMP9]], i8* noundef nonnull align 8 dereferenceable(48) [[TMP10]], i64 48, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS3]] ; CHECK-NEXT: br label [[NO_ALIAS3]]
; CHECK: no_alias3: ; CHECK: no_alias3:
; CHECK-NEXT: [[TMP11:%.*]] = phi <6 x double>* [ [[B]], [[NO_ALIAS]] ], [ [[B]], [[ALIAS_CONT1]] ], [ [[TMP8]], [[COPY2]] ] ; CHECK-NEXT: [[TMP11:%.*]] = phi <6 x double>* [ [[B]], [[NO_ALIAS]] ], [ [[B]], [[ALIAS_CONT1]] ], [ [[BC2]], [[COPY2]] ]
; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <6 x double>* [[TMP5]] to <2 x double>* ; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <6 x double>* [[TMP5]] to <2 x double>*
; CHECK-NEXT: [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 8 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr <6 x double>, <6 x double>* [[TMP5]], i64 0, i64 3 ; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr <6 x double>, <6 x double>* [[TMP5]], i64 0, i64 3
; CHECK-NEXT: [[VEC_CAST8:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>* ; CHECK-NEXT: [[VEC_CAST8:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>*
; CHECK-NEXT: [[COL_LOAD9:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST8]], align 8 ; CHECK-NEXT: [[COL_LOAD9:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST8]], align 8
; CHECK-NEXT: [[VEC_CAST11:%.*]] = bitcast <6 x double>* [[TMP11]] to <2 x double>* ; CHECK-NEXT: [[VEC_CAST11:%.*]] = bitcast <6 x double>* [[TMP11]] to <2 x double>*
; CHECK-NEXT: [[COL_LOAD12:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST11]], align 8 ; CHECK-NEXT: [[COL_LOAD12:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST11]], align 8
; CHECK-NEXT: [[VEC_GEP13:%.*]] = getelementptr <6 x double>, <6 x double>* [[TMP11]], i64 0, i64 2 ; CHECK-NEXT: [[VEC_GEP13:%.*]] = getelementptr <6 x double>, <6 x double>* [[TMP11]], i64 0, i64 2
▲ Show 20 Lines • Show All 97 Lines • ▼ Show 20 Lines
; CHECK-NEXT: [[LOAD_BEGIN96:%.*]] = ptrtoint <6 x double>* [[A]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN96:%.*]] = ptrtoint <6 x double>* [[A]] to i64
; CHECK-NEXT: [[TMP36:%.*]] = icmp ugt i64 [[STORE_END95]], [[LOAD_BEGIN96]] ; CHECK-NEXT: [[TMP36:%.*]] = icmp ugt i64 [[STORE_END95]], [[LOAD_BEGIN96]]
; CHECK-NEXT: br i1 [[TMP36]], label [[ALIAS_CONT91:%.*]], label [[NO_ALIAS93:%.*]] ; CHECK-NEXT: br i1 [[TMP36]], label [[ALIAS_CONT91:%.*]], label [[NO_ALIAS93:%.*]]
; CHECK: alias_cont91: ; CHECK: alias_cont91:
; CHECK-NEXT: [[LOAD_END97:%.*]] = add nuw nsw i64 [[LOAD_BEGIN96]], 48 ; CHECK-NEXT: [[LOAD_END97:%.*]] = add nuw nsw i64 [[LOAD_BEGIN96]], 48
; CHECK-NEXT: [[TMP37:%.*]] = icmp ugt i64 [[LOAD_END97]], [[STORE_BEGIN94]] ; CHECK-NEXT: [[TMP37:%.*]] = icmp ugt i64 [[LOAD_END97]], [[STORE_BEGIN94]]
; CHECK-NEXT: br i1 [[TMP37]], label [[COPY92:%.*]], label [[NO_ALIAS93]] ; CHECK-NEXT: br i1 [[TMP37]], label [[COPY92:%.*]], label [[NO_ALIAS93]]
; CHECK: copy92: ; CHECK: copy92:
; CHECK-NEXT: [[TMP38:%.*]] = alloca <6 x double>, align 64 ; CHECK-NEXT: [[TMP38:%.*]] = alloca [6 x double], align 8
; CHECK-NEXT: [[TMP39:%.*]] = bitcast <6 x double>* [[TMP38]] to i8* ; CHECK-NEXT: [[BC3:%.+]] = bitcast [6 x double]* [[TMP38]] to <6 x double>*
; CHECK-NEXT: [[TMP39:%.*]] = bitcast [6 x double]* [[TMP38]] to i8*
; CHECK-NEXT: [[TMP40:%.*]] = bitcast <6 x double>* [[A]] to i8* ; CHECK-NEXT: [[TMP40:%.*]] = bitcast <6 x double>* [[A]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 64 dereferenceable(48) [[TMP39]], i8* noundef nonnull align 8 dereferenceable(48) [[TMP40]], i64 48, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(48) [[TMP39]], i8* noundef nonnull align 8 dereferenceable(48) [[TMP40]], i64 48, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS93]] ; CHECK-NEXT: br label [[NO_ALIAS93]]
; CHECK: no_alias93: ; CHECK: no_alias93:
; CHECK-NEXT: [[TMP41:%.*]] = phi <6 x double>* [ [[A]], [[END]] ], [ [[A]], [[ALIAS_CONT91]] ], [ [[TMP38]], [[COPY92]] ] ; CHECK-NEXT: [[TMP41:%.*]] = phi <6 x double>* [ [[A]], [[END]] ], [ [[A]], [[ALIAS_CONT91]] ], [ [[BC3]], [[COPY92]] ]
; CHECK-NEXT: [[STORE_BEGIN101:%.*]] = ptrtoint <9 x double>* [[C]] to i64 ; CHECK-NEXT: [[STORE_BEGIN101:%.*]] = ptrtoint <9 x double>* [[C]] to i64
; CHECK-NEXT: [[STORE_END102:%.*]] = add nuw nsw i64 [[STORE_BEGIN101]], 72 ; CHECK-NEXT: [[STORE_END102:%.*]] = add nuw nsw i64 [[STORE_BEGIN101]], 72
; CHECK-NEXT: [[LOAD_BEGIN103:%.*]] = ptrtoint <6 x double>* [[B]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN103:%.*]] = ptrtoint <6 x double>* [[B]] to i64
; CHECK-NEXT: [[TMP42:%.*]] = icmp ugt i64 [[STORE_END102]], [[LOAD_BEGIN103]] ; CHECK-NEXT: [[TMP42:%.*]] = icmp ugt i64 [[STORE_END102]], [[LOAD_BEGIN103]]
; CHECK-NEXT: br i1 [[TMP42]], label [[ALIAS_CONT98:%.*]], label [[NO_ALIAS100:%.*]] ; CHECK-NEXT: br i1 [[TMP42]], label [[ALIAS_CONT98:%.*]], label [[NO_ALIAS100:%.*]]
; CHECK: alias_cont98: ; CHECK: alias_cont98:
; CHECK-NEXT: [[LOAD_END104:%.*]] = add nuw nsw i64 [[LOAD_BEGIN103]], 48 ; CHECK-NEXT: [[LOAD_END104:%.*]] = add nuw nsw i64 [[LOAD_BEGIN103]], 48
; CHECK-NEXT: [[TMP43:%.*]] = icmp ugt i64 [[LOAD_END104]], [[STORE_BEGIN101]] ; CHECK-NEXT: [[TMP43:%.*]] = icmp ugt i64 [[LOAD_END104]], [[STORE_BEGIN101]]
; CHECK-NEXT: br i1 [[TMP43]], label [[COPY99:%.*]], label [[NO_ALIAS100]] ; CHECK-NEXT: br i1 [[TMP43]], label [[COPY99:%.*]], label [[NO_ALIAS100]]
; CHECK: copy99: ; CHECK: copy99:
; CHECK-NEXT: [[TMP44:%.*]] = alloca <6 x double>, align 64 ; CHECK-NEXT: [[TMP44:%.*]] = alloca [6 x double], align 8
; CHECK-NEXT: [[TMP45:%.*]] = bitcast <6 x double>* [[TMP44]] to i8* ; CHECK-NEXT: [[BC4:%.+]] = bitcast [6 x double]* [[TMP44]] to <6 x double>*
; CHECK-NEXT: [[TMP45:%.*]] = bitcast [6 x double]* [[TMP44]] to i8*
; CHECK-NEXT: [[TMP46:%.*]] = bitcast <6 x double>* [[B]] to i8* ; CHECK-NEXT: [[TMP46:%.*]] = bitcast <6 x double>* [[B]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 64 dereferenceable(48) [[TMP45]], i8* noundef nonnull align 8 dereferenceable(48) [[TMP46]], i64 48, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(48) [[TMP45]], i8* noundef nonnull align 8 dereferenceable(48) [[TMP46]], i64 48, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS100]] ; CHECK-NEXT: br label [[NO_ALIAS100]]
; CHECK: no_alias100: ; CHECK: no_alias100:
; CHECK-NEXT: [[TMP47:%.*]] = phi <6 x double>* [ [[B]], [[NO_ALIAS93]] ], [ [[B]], [[ALIAS_CONT98]] ], [ [[TMP44]], [[COPY99]] ] ; CHECK-NEXT: [[TMP47:%.*]] = phi <6 x double>* [ [[B]], [[NO_ALIAS93]] ], [ [[B]], [[ALIAS_CONT98]] ], [ [[BC4]], [[COPY99]] ]
; CHECK-NEXT: [[VEC_CAST106:%.*]] = bitcast <6 x double>* [[TMP41]] to <2 x double>* ; CHECK-NEXT: [[VEC_CAST106:%.*]] = bitcast <6 x double>* [[TMP41]] to <2 x double>*
; CHECK-NEXT: [[COL_LOAD107:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST106]], align 8 ; CHECK-NEXT: [[COL_LOAD107:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST106]], align 8
; CHECK-NEXT: [[VEC_GEP108:%.*]] = getelementptr <6 x double>, <6 x double>* [[TMP41]], i64 0, i64 3 ; CHECK-NEXT: [[VEC_GEP108:%.*]] = getelementptr <6 x double>, <6 x double>* [[TMP41]], i64 0, i64 3
; CHECK-NEXT: [[VEC_CAST109:%.*]] = bitcast double* [[VEC_GEP108]] to <2 x double>* ; CHECK-NEXT: [[VEC_CAST109:%.*]] = bitcast double* [[VEC_GEP108]] to <2 x double>*
; CHECK-NEXT: [[COL_LOAD110:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST109]], align 8 ; CHECK-NEXT: [[COL_LOAD110:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST109]], align 8
; CHECK-NEXT: [[VEC_CAST112:%.*]] = bitcast <6 x double>* [[TMP47]] to <2 x double>* ; CHECK-NEXT: [[VEC_CAST112:%.*]] = bitcast <6 x double>* [[TMP47]] to <2 x double>*
; CHECK-NEXT: [[COL_LOAD113:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST112]], align 8 ; CHECK-NEXT: [[COL_LOAD113:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST112]], align 8
; CHECK-NEXT: [[VEC_GEP114:%.*]] = getelementptr <6 x double>, <6 x double>* [[TMP47]], i64 0, i64 2 ; CHECK-NEXT: [[VEC_GEP114:%.*]] = getelementptr <6 x double>, <6 x double>* [[TMP47]], i64 0, i64 2
▲ Show 20 Lines • Show All 100 Lines • Show Last 20 Lines

llvm/test/Transforms/LowerMatrixIntrinsics/multiply-fused.ll

Show All 16 Lines
; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <16 x double>* [[A:%.*]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN:%.*]] = ptrtoint <16 x double>* [[A:%.*]] to i64
; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]] ; CHECK-NEXT: [[TMP0:%.*]] = icmp ugt i64 [[STORE_END]], [[LOAD_BEGIN]]
; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]] ; CHECK-NEXT: br i1 [[TMP0]], label [[ALIAS_CONT:%.*]], label [[NO_ALIAS:%.*]]
; CHECK: alias_cont: ; CHECK: alias_cont:
; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 128 ; CHECK-NEXT: [[LOAD_END:%.*]] = add nuw nsw i64 [[LOAD_BEGIN]], 128
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]] ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[LOAD_END]], [[STORE_BEGIN]]
; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]] ; CHECK-NEXT: br i1 [[TMP1]], label [[COPY:%.*]], label [[NO_ALIAS]]
; CHECK: copy: ; CHECK: copy:
; CHECK-NEXT: [[TMP2:%.*]] = alloca <16 x double>, align 128 ; CHECK-NEXT: [[TMP2:%.*]] = alloca <16 x double>, align 8
; CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x double>* [[TMP2]] to i8* ; CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x double>* [[TMP2]] to i8*
; CHECK-NEXT: [[TMP4:%.*]] = bitcast <16 x double>* [[A]] to i8* ; CHECK-NEXT: [[TMP4:%.*]] = bitcast <16 x double>* [[A]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 128 dereferenceable(128) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(128) [[TMP4]], i64 128, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(128) [[TMP3]], i8* noundef nonnull align 8 dereferenceable(128) [[TMP4]], i64 128, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS]] ; CHECK-NEXT: br label [[NO_ALIAS]]
; CHECK: no_alias: ; CHECK: no_alias:
; CHECK-NEXT: [[TMP5:%.*]] = phi <16 x double>* [ [[A]], [[ENTRY:%.*]] ], [ [[A]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ] ; CHECK-NEXT: [[TMP5:%.*]] = phi <16 x double>* [ [[A]], [[ENTRY:%.*]] ], [ [[A]], [[ALIAS_CONT]] ], [ [[TMP2]], [[COPY]] ]
; CHECK-NEXT: [[STORE_BEGIN4:%.*]] = ptrtoint <16 x double>* [[C]] to i64 ; CHECK-NEXT: [[STORE_BEGIN4:%.*]] = ptrtoint <16 x double>* [[C]] to i64
; CHECK-NEXT: [[STORE_END5:%.*]] = add nuw nsw i64 [[STORE_BEGIN4]], 128 ; CHECK-NEXT: [[STORE_END5:%.*]] = add nuw nsw i64 [[STORE_BEGIN4]], 128
; CHECK-NEXT: [[LOAD_BEGIN6:%.*]] = ptrtoint <16 x double>* [[B:%.*]] to i64 ; CHECK-NEXT: [[LOAD_BEGIN6:%.*]] = ptrtoint <16 x double>* [[B:%.*]] to i64
; CHECK-NEXT: [[TMP6:%.*]] = icmp ugt i64 [[STORE_END5]], [[LOAD_BEGIN6]] ; CHECK-NEXT: [[TMP6:%.*]] = icmp ugt i64 [[STORE_END5]], [[LOAD_BEGIN6]]
; CHECK-NEXT: br i1 [[TMP6]], label [[ALIAS_CONT1:%.*]], label [[NO_ALIAS3:%.*]] ; CHECK-NEXT: br i1 [[TMP6]], label [[ALIAS_CONT1:%.*]], label [[NO_ALIAS3:%.*]]
; CHECK: alias_cont1: ; CHECK: alias_cont1:
; CHECK-NEXT: [[LOAD_END7:%.*]] = add nuw nsw i64 [[LOAD_BEGIN6]], 128 ; CHECK-NEXT: [[LOAD_END7:%.*]] = add nuw nsw i64 [[LOAD_BEGIN6]], 128
; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i64 [[LOAD_END7]], [[STORE_BEGIN4]] ; CHECK-NEXT: [[TMP7:%.*]] = icmp ugt i64 [[LOAD_END7]], [[STORE_BEGIN4]]
; CHECK-NEXT: br i1 [[TMP7]], label [[COPY2:%.*]], label [[NO_ALIAS3]] ; CHECK-NEXT: br i1 [[TMP7]], label [[COPY2:%.*]], label [[NO_ALIAS3]]
; CHECK: copy2: ; CHECK: copy2:
; CHECK-NEXT: [[TMP8:%.*]] = alloca <16 x double>, align 128 ; CHECK-NEXT: [[TMP8:%.*]] = alloca <16 x double>, align 8
; CHECK-NEXT: [[TMP9:%.*]] = bitcast <16 x double>* [[TMP8]] to i8* ; CHECK-NEXT: [[TMP9:%.*]] = bitcast <16 x double>* [[TMP8]] to i8*
; CHECK-NEXT: [[TMP10:%.*]] = bitcast <16 x double>* [[B]] to i8* ; CHECK-NEXT: [[TMP10:%.*]] = bitcast <16 x double>* [[B]] to i8*
; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 128 dereferenceable(128) [[TMP9]], i8* noundef nonnull align 8 dereferenceable(128) [[TMP10]], i64 128, i1 false) ; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 8 dereferenceable(128) [[TMP9]], i8* noundef nonnull align 8 dereferenceable(128) [[TMP10]], i64 128, i1 false)
; CHECK-NEXT: br label [[NO_ALIAS3]] ; CHECK-NEXT: br label [[NO_ALIAS3]]
; CHECK: no_alias3: ; CHECK: no_alias3:
; CHECK-NEXT: [[TMP11:%.*]] = phi <16 x double>* [ [[B]], [[NO_ALIAS]] ], [ [[B]], [[ALIAS_CONT1]] ], [ [[TMP8]], [[COPY2]] ] ; CHECK-NEXT: [[TMP11:%.*]] = phi <16 x double>* [ [[B]], [[NO_ALIAS]] ], [ [[B]], [[ALIAS_CONT1]] ], [ [[TMP8]], [[COPY2]] ]
; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <16 x double>* [[TMP5]] to <2 x double>* ; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast <16 x double>* [[TMP5]] to <2 x double>*
; CHECK-NEXT: [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 8 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 8
; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr <16 x double>, <16 x double>* [[TMP5]], i64 0, i64 4 ; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr <16 x double>, <16 x double>* [[TMP5]], i64 0, i64 4
; CHECK-NEXT: [[VEC_CAST8:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>* ; CHECK-NEXT: [[VEC_CAST8:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>*
; CHECK-NEXT: [[COL_LOAD9:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST8]], align 8 ; CHECK-NEXT: [[COL_LOAD9:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST8]], align 8
▲ Show 20 Lines • Show All 405 Lines • Show Last 20 Lines