maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[ScalarizeMaskedMemIntrin] Use IRBuilder functions that take uint32_t/uint64_t for getelementptr, extractelement, and insertelement. 

This saves needing to call getInt32 ourselves. Making the code a little shorter.

The test changes are because insert/extract use getInt64 internally. Shouldn't be a functional issue.

This cleanup because I plan to write similar code for expandload/compressstore.

llvm-svn: 355767
This commit is contained in:
Craig Topper 2019-03-09 02:08:41 +00:00
parent 08ae3e0f0b
commit 69f8c1653d
5 changed files with 68 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

72
llvm/lib/CodeGen/ScalarizeMaskedMemIntrin.cpp
View File

@ -151,9 +151,9 @@ static void scalarizeMaskedLoad(CallInst *CI, bool &ModifiedDT) {
// Adjust alignment for the scalar instruction. // Adjust alignment for the scalar instruction.
AlignVal = MinAlign(AlignVal, EltTy->getPrimitiveSizeInBits() / 8); AlignVal = MinAlign(AlignVal, EltTy->getPrimitiveSizeInBits() / 8);
// Bitcast %addr fron i8* to EltTy* // Bitcast %addr from i8* to EltTy*
Type *NewPtrType = Type *NewPtrType =
EltTy->getPointerTo(cast<PointerType>(Ptr->getType())->getAddressSpace()); EltTy->getPointerTo(Ptr->getType()->getPointerAddressSpace());
Value *FirstEltPtr = Builder.CreateBitCast(Ptr, NewPtrType); Value *FirstEltPtr = Builder.CreateBitCast(Ptr, NewPtrType);
unsigned VectorWidth = VecType->getNumElements(); unsigned VectorWidth = VecType->getNumElements();
@ -164,11 +164,9 @@ static void scalarizeMaskedLoad(CallInst *CI, bool &ModifiedDT) {
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) { for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
if (cast<Constant>(Mask)->getAggregateElement(Idx)->isNullValue()) if (cast<Constant>(Mask)->getAggregateElement(Idx)->isNullValue())
continue; continue;
Value *Gep = Value *Gep = Builder.CreateConstInBoundsGEP1_32(EltTy, FirstEltPtr, Idx);
Builder.CreateInBoundsGEP(EltTy, FirstEltPtr, Builder.getInt32(Idx));
LoadInst *Load = Builder.CreateAlignedLoad(EltTy, Gep, AlignVal); LoadInst *Load = Builder.CreateAlignedLoad(EltTy, Gep, AlignVal);
VResult = VResult = Builder.CreateInsertElement(VResult, Load, Idx);
Builder.CreateInsertElement(VResult, Load, Builder.getInt32(Idx));
} }
CI->replaceAllUsesWith(VResult); CI->replaceAllUsesWith(VResult);
CI->eraseFromParent(); CI->eraseFromParent();
@ -183,8 +181,7 @@ static void scalarizeMaskedLoad(CallInst *CI, bool &ModifiedDT) {
// br i1 %mask_1, label %cond.load, label %else // br i1 %mask_1, label %cond.load, label %else
// //
Value *Predicate = Value *Predicate = Builder.CreateExtractElement(Mask, Idx);
Builder.CreateExtractElement(Mask, Builder.getInt32(Idx));
// Create "cond" block // Create "cond" block
// //
@ -196,11 +193,9 @@ static void scalarizeMaskedLoad(CallInst *CI, bool &ModifiedDT) {
"cond.load"); "cond.load");
Builder.SetInsertPoint(InsertPt); Builder.SetInsertPoint(InsertPt);
Value *Gep = Value *Gep = Builder.CreateConstInBoundsGEP1_32(EltTy, FirstEltPtr, Idx);
Builder.CreateInBoundsGEP(EltTy, FirstEltPtr, Builder.getInt32(Idx));
LoadInst *Load = Builder.CreateAlignedLoad(EltTy, Gep, AlignVal); LoadInst *Load = Builder.CreateAlignedLoad(EltTy, Gep, AlignVal);
Value *NewVResult = Builder.CreateInsertElement(VResult, Load, Value *NewVResult = Builder.CreateInsertElement(VResult, Load, Idx);
Builder.getInt32(Idx));
// Create "else" block, fill it in the next iteration // Create "else" block, fill it in the next iteration
BasicBlock *NewIfBlock = BasicBlock *NewIfBlock =
@ -277,9 +272,9 @@ static void scalarizeMaskedStore(CallInst *CI, bool &ModifiedDT) {
// Adjust alignment for the scalar instruction. // Adjust alignment for the scalar instruction.
AlignVal = MinAlign(AlignVal, EltTy->getPrimitiveSizeInBits() / 8); AlignVal = MinAlign(AlignVal, EltTy->getPrimitiveSizeInBits() / 8);
// Bitcast %addr fron i8* to EltTy* // Bitcast %addr from i8* to EltTy*
Type *NewPtrType = Type *NewPtrType =
EltTy->getPointerTo(cast<PointerType>(Ptr->getType())->getAddressSpace()); EltTy->getPointerTo(Ptr->getType()->getPointerAddressSpace());
Value *FirstEltPtr = Builder.CreateBitCast(Ptr, NewPtrType); Value *FirstEltPtr = Builder.CreateBitCast(Ptr, NewPtrType);
unsigned VectorWidth = VecType->getNumElements(); unsigned VectorWidth = VecType->getNumElements();
@ -287,9 +282,8 @@ static void scalarizeMaskedStore(CallInst *CI, bool &ModifiedDT) {
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) { for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
if (cast<Constant>(Mask)->getAggregateElement(Idx)->isNullValue()) if (cast<Constant>(Mask)->getAggregateElement(Idx)->isNullValue())
continue; continue;
Value *OneElt = Builder.CreateExtractElement(Src, Builder.getInt32(Idx)); Value *OneElt = Builder.CreateExtractElement(Src, Idx);
Value *Gep = Value *Gep = Builder.CreateConstInBoundsGEP1_32(EltTy, FirstEltPtr, Idx);
Builder.CreateInBoundsGEP(EltTy, FirstEltPtr, Builder.getInt32(Idx));
Builder.CreateAlignedStore(OneElt, Gep, AlignVal); Builder.CreateAlignedStore(OneElt, Gep, AlignVal);
} }
CI->eraseFromParent(); CI->eraseFromParent();
@ -302,8 +296,7 @@ static void scalarizeMaskedStore(CallInst *CI, bool &ModifiedDT) {
// %mask_1 = extractelement <16 x i1> %mask, i32 Idx // %mask_1 = extractelement <16 x i1> %mask, i32 Idx
// br i1 %mask_1, label %cond.store, label %else // br i1 %mask_1, label %cond.store, label %else
// //
Value *Predicate = Value *Predicate = Builder.CreateExtractElement(Mask, Idx);
Builder.CreateExtractElement(Mask, Builder.getInt32(Idx));
// Create "cond" block // Create "cond" block
// //
@ -315,9 +308,8 @@ static void scalarizeMaskedStore(CallInst *CI, bool &ModifiedDT) {
IfBlock->splitBasicBlock(InsertPt->getIterator(), "cond.store"); IfBlock->splitBasicBlock(InsertPt->getIterator(), "cond.store");
Builder.SetInsertPoint(InsertPt); Builder.SetInsertPoint(InsertPt);
Value *OneElt = Builder.CreateExtractElement(Src, Builder.getInt32(Idx)); Value *OneElt = Builder.CreateExtractElement(Src, Idx);
Value *Gep = Value *Gep = Builder.CreateConstInBoundsGEP1_32(EltTy, FirstEltPtr, Idx);
Builder.CreateInBoundsGEP(EltTy, FirstEltPtr, Builder.getInt32(Idx));
Builder.CreateAlignedStore(OneElt, Gep, AlignVal); Builder.CreateAlignedStore(OneElt, Gep, AlignVal);
// Create "else" block, fill it in the next iteration // Create "else" block, fill it in the next iteration
@ -389,12 +381,11 @@ static void scalarizeMaskedGather(CallInst *CI, bool &ModifiedDT) {
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) { for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
if (cast<Constant>(Mask)->getAggregateElement(Idx)->isNullValue()) if (cast<Constant>(Mask)->getAggregateElement(Idx)->isNullValue())
continue; continue;
Value *Ptr = Builder.CreateExtractElement(Ptrs, Builder.getInt32(Idx), Value *Ptr = Builder.CreateExtractElement(Ptrs, Idx, "Ptr" + Twine(Idx));
"Ptr" + Twine(Idx));
LoadInst *Load = LoadInst *Load =
Builder.CreateAlignedLoad(EltTy, Ptr, AlignVal, "Load" + Twine(Idx)); Builder.CreateAlignedLoad(EltTy, Ptr, AlignVal, "Load" + Twine(Idx));
VResult = Builder.CreateInsertElement( VResult =
VResult, Load, Builder.getInt32(Idx), "Res" + Twine(Idx)); Builder.CreateInsertElement(VResult, Load, Idx, "Res" + Twine(Idx));
} }
CI->replaceAllUsesWith(VResult); CI->replaceAllUsesWith(VResult);
CI->eraseFromParent(); CI->eraseFromParent();
@ -408,8 +399,8 @@ static void scalarizeMaskedGather(CallInst *CI, bool &ModifiedDT) {
// br i1 %Mask1, label %cond.load, label %else // br i1 %Mask1, label %cond.load, label %else
// //
Value *Predicate = Builder.CreateExtractElement(Mask, Builder.getInt32(Idx), Value *Predicate =
"Mask" + Twine(Idx)); Builder.CreateExtractElement(Mask, Idx, "Mask" + Twine(Idx));
// Create "cond" block // Create "cond" block
// //
@ -420,13 +411,11 @@ static void scalarizeMaskedGather(CallInst *CI, bool &ModifiedDT) {
BasicBlock *CondBlock = IfBlock->splitBasicBlock(InsertPt, "cond.load"); BasicBlock *CondBlock = IfBlock->splitBasicBlock(InsertPt, "cond.load");
Builder.SetInsertPoint(InsertPt); Builder.SetInsertPoint(InsertPt);
Value *Ptr = Builder.CreateExtractElement(Ptrs, Builder.getInt32(Idx), Value *Ptr = Builder.CreateExtractElement(Ptrs, Idx, "Ptr" + Twine(Idx));
"Ptr" + Twine(Idx));
LoadInst *Load = LoadInst *Load =
Builder.CreateAlignedLoad(EltTy, Ptr, AlignVal, "Load" + Twine(Idx)); Builder.CreateAlignedLoad(EltTy, Ptr, AlignVal, "Load" + Twine(Idx));
Value *NewVResult = Builder.CreateInsertElement(VResult, Load, Value *NewVResult =
Builder.getInt32(Idx), Builder.CreateInsertElement(VResult, Load, Idx, "Res" + Twine(Idx));
"Res" + Twine(Idx));
// Create "else" block, fill it in the next iteration // Create "else" block, fill it in the next iteration
BasicBlock *NewIfBlock = CondBlock->splitBasicBlock(InsertPt, "else"); BasicBlock *NewIfBlock = CondBlock->splitBasicBlock(InsertPt, "else");
@ -501,10 +490,9 @@ static void scalarizeMaskedScatter(CallInst *CI, bool &ModifiedDT) {
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) { for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
if (cast<ConstantVector>(Mask)->getAggregateElement(Idx)->isNullValue()) if (cast<ConstantVector>(Mask)->getAggregateElement(Idx)->isNullValue())
continue; continue;
Value *OneElt = Builder.CreateExtractElement(Src, Builder.getInt32(Idx), Value *OneElt =
"Elt" + Twine(Idx)); Builder.CreateExtractElement(Src, Idx, "Elt" + Twine(Idx));
Value *Ptr = Builder.CreateExtractElement(Ptrs, Builder.getInt32(Idx), Value *Ptr = Builder.CreateExtractElement(Ptrs, Idx, "Ptr" + Twine(Idx));
"Ptr" + Twine(Idx));
Builder.CreateAlignedStore(OneElt, Ptr, AlignVal); Builder.CreateAlignedStore(OneElt, Ptr, AlignVal);
} }
CI->eraseFromParent(); CI->eraseFromParent();
@ -517,8 +505,8 @@ static void scalarizeMaskedScatter(CallInst *CI, bool &ModifiedDT) {
// %Mask1 = extractelement <16 x i1> %Mask, i32 Idx // %Mask1 = extractelement <16 x i1> %Mask, i32 Idx
// br i1 %Mask1, label %cond.store, label %else // br i1 %Mask1, label %cond.store, label %else
// //
Value *Predicate = Builder.CreateExtractElement(Mask, Builder.getInt32(Idx), Value *Predicate =
"Mask" + Twine(Idx)); Builder.CreateExtractElement(Mask, Idx, "Mask" + Twine(Idx));
// Create "cond" block // Create "cond" block
// //
@ -529,10 +517,8 @@ static void scalarizeMaskedScatter(CallInst *CI, bool &ModifiedDT) {
BasicBlock *CondBlock = IfBlock->splitBasicBlock(InsertPt, "cond.store"); BasicBlock *CondBlock = IfBlock->splitBasicBlock(InsertPt, "cond.store");
Builder.SetInsertPoint(InsertPt); Builder.SetInsertPoint(InsertPt);
Value *OneElt = Builder.CreateExtractElement(Src, Builder.getInt32(Idx), Value *OneElt = Builder.CreateExtractElement(Src, Idx, "Elt" + Twine(Idx));
"Elt" + Twine(Idx)); Value *Ptr = Builder.CreateExtractElement(Ptrs, Idx, "Ptr" + Twine(Idx));
Value *Ptr = Builder.CreateExtractElement(Ptrs, Builder.getInt32(Idx),
"Ptr" + Twine(Idx));
Builder.CreateAlignedStore(OneElt, Ptr, AlignVal); Builder.CreateAlignedStore(OneElt, Ptr, AlignVal);
// Create "else" block, fill it in the next iteration // Create "else" block, fill it in the next iteration

18
llvm/test/CodeGen/X86/masked_gather_scatter.ll
View File

@ -69,7 +69,7 @@ declare <8 x i32> @llvm.masked.gather.v8i32.v8p0i32(<8 x i32*> , i32, <8 x i1> ,
; SCALAR-NEXT: br label %else ; SCALAR-NEXT: br label %else
; SCALAR: else: ; SCALAR: else:
; SCALAR-NEXT: %res.phi.else = phi ; SCALAR-NEXT: %res.phi.else = phi
; SCALAR-NEXT: %Mask1 = extractelement <16 x i1> %imask, i32 1 ; SCALAR-NEXT: %Mask1 = extractelement <16 x i1> %imask, i64 1
; SCALAR-NEXT: br i1 %Mask1, label %cond.load1, label %else2 ; SCALAR-NEXT: br i1 %Mask1, label %cond.load1, label %else2
define <16 x float> @test2(float* %base, <16 x i32> %ind, i16 %mask) { define <16 x float> @test2(float* %base, <16 x i32> %ind, i16 %mask) {
@ -211,15 +211,15 @@ define <16 x i32> @test4(i32* %base, <16 x i32> %ind, i16 %mask) {
; SCALAR-LABEL: test5 ; SCALAR-LABEL: test5
; SCALAR: %Mask0 = extractelement <16 x i1> %imask, i32 0 ; SCALAR: %Mask0 = extractelement <16 x i1> %imask, i64 0
; SCALAR-NEXT: br i1 %Mask0, label %cond.store, label %else ; SCALAR-NEXT: br i1 %Mask0, label %cond.store, label %else
; SCALAR: cond.store: ; SCALAR: cond.store:
; SCALAR-NEXT: %Elt0 = extractelement <16 x i32> %val, i32 0 ; SCALAR-NEXT: %Elt0 = extractelement <16 x i32> %val, i64 0
; SCALAR-NEXT: %Ptr0 = extractelement <16 x i32*> %gep.random, i32 0 ; SCALAR-NEXT: %Ptr0 = extractelement <16 x i32*> %gep.random, i64 0
; SCALAR-NEXT: store i32 %Elt0, i32* %Ptr0, align 4 ; SCALAR-NEXT: store i32 %Elt0, i32* %Ptr0, align 4
; SCALAR-NEXT: br label %else ; SCALAR-NEXT: br label %else
; SCALAR: else: ; SCALAR: else:
; SCALAR-NEXT: %Mask1 = extractelement <16 x i1> %imask, i32 1 ; SCALAR-NEXT: %Mask1 = extractelement <16 x i1> %imask, i64 1
; SCALAR-NEXT: br i1 %Mask1, label %cond.store1, label %else2 ; SCALAR-NEXT: br i1 %Mask1, label %cond.store1, label %else2
define void @test5(i32* %base, <16 x i32> %ind, i16 %mask, <16 x i32>%val) { define void @test5(i32* %base, <16 x i32> %ind, i16 %mask, <16 x i32>%val) {
@ -277,11 +277,11 @@ declare void @llvm.masked.scatter.v16i32.v16p0i32(<16 x i32> , <16 x i32*> , i32
; SCALAR-LABEL: test6 ; SCALAR-LABEL: test6
; SCALAR: store i32 %Elt0, i32* %Ptr01, align 4 ; SCALAR: store i32 %Elt0, i32* %Ptr01, align 4
; SCALAR-NEXT: %Elt1 = extractelement <8 x i32> %a1, i32 1 ; SCALAR-NEXT: %Elt1 = extractelement <8 x i32> %a1, i64 1
; SCALAR-NEXT: %Ptr12 = extractelement <8 x i32*> %ptr, i32 1 ; SCALAR-NEXT: %Ptr12 = extractelement <8 x i32*> %ptr, i64 1
; SCALAR-NEXT: store i32 %Elt1, i32* %Ptr12, align 4 ; SCALAR-NEXT: store i32 %Elt1, i32* %Ptr12, align 4
; SCALAR-NEXT: %Elt2 = extractelement <8 x i32> %a1, i32 2 ; SCALAR-NEXT: %Elt2 = extractelement <8 x i32> %a1, i64 2
; SCALAR-NEXT: %Ptr23 = extractelement <8 x i32*> %ptr, i32 2 ; SCALAR-NEXT: %Ptr23 = extractelement <8 x i32*> %ptr, i64 2
; SCALAR-NEXT: store i32 %Elt2, i32* %Ptr23, align 4 ; SCALAR-NEXT: store i32 %Elt2, i32* %Ptr23, align 4
define <8 x i32> @test6(<8 x i32>%a1, <8 x i32*> %ptr) { define <8 x i32> @test6(<8 x i32>%a1, <8 x i32*> %ptr) {

24
llvm/test/Transforms/ScalarizeMaskedMemIntrin/X86/expand-masked-gather.ll
View File

@ -3,21 +3,21 @@
define <2 x i64> @scalarize_v2i64(<2 x i64*> %p, <2 x i1> %mask, <2 x i64> %passthru) { define <2 x i64> @scalarize_v2i64(<2 x i64*> %p, <2 x i1> %mask, <2 x i64> %passthru) {
; CHECK-LABEL: @scalarize_v2i64( ; CHECK-LABEL: @scalarize_v2i64(
; CHECK-NEXT: [[MASK0:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i32 0 ; CHECK-NEXT: [[MASK0:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
; CHECK-NEXT: br i1 [[MASK0]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]] ; CHECK-NEXT: br i1 [[MASK0]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
; CHECK: cond.load: ; CHECK: cond.load:
; CHECK-NEXT: [[PTR0:%.*]] = extractelement <2 x i64*> [[P:%.*]], i32 0 ; CHECK-NEXT: [[PTR0:%.*]] = extractelement <2 x i64*> [[P:%.*]], i64 0
; CHECK-NEXT: [[LOAD0:%.*]] = load i64, i64* [[PTR0]], align 8 ; CHECK-NEXT: [[LOAD0:%.*]] = load i64, i64* [[PTR0]], align 8
; CHECK-NEXT: [[RES0:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[LOAD0]], i32 0 ; CHECK-NEXT: [[RES0:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[LOAD0]], i64 0
; CHECK-NEXT: br label [[ELSE]] ; CHECK-NEXT: br label [[ELSE]]
; CHECK: else: ; CHECK: else:
; CHECK-NEXT: [[RES_PHI_ELSE:%.*]] = phi <2 x i64> [ [[RES0]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ] ; CHECK-NEXT: [[RES_PHI_ELSE:%.*]] = phi <2 x i64> [ [[RES0]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
; CHECK-NEXT: [[MASK1:%.*]] = extractelement <2 x i1> [[MASK]], i32 1 ; CHECK-NEXT: [[MASK1:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
; CHECK-NEXT: br i1 [[MASK1]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]] ; CHECK-NEXT: br i1 [[MASK1]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
; CHECK: cond.load1: ; CHECK: cond.load1:
; CHECK-NEXT: [[PTR1:%.*]] = extractelement <2 x i64*> [[P]], i32 1 ; CHECK-NEXT: [[PTR1:%.*]] = extractelement <2 x i64*> [[P]], i64 1
; CHECK-NEXT: [[LOAD1:%.*]] = load i64, i64* [[PTR1]], align 8 ; CHECK-NEXT: [[LOAD1:%.*]] = load i64, i64* [[PTR1]], align 8
; CHECK-NEXT: [[RES1:%.*]] = insertelement <2 x i64> [[RES_PHI_ELSE]], i64 [[LOAD1]], i32 1 ; CHECK-NEXT: [[RES1:%.*]] = insertelement <2 x i64> [[RES_PHI_ELSE]], i64 [[LOAD1]], i64 1
; CHECK-NEXT: br label [[ELSE2]] ; CHECK-NEXT: br label [[ELSE2]]
; CHECK: else2: ; CHECK: else2:
; CHECK-NEXT: [[RES_PHI_ELSE3:%.*]] = phi <2 x i64> [ [[RES1]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ] ; CHECK-NEXT: [[RES_PHI_ELSE3:%.*]] = phi <2 x i64> [ [[RES1]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
@ -29,12 +29,12 @@ define <2 x i64> @scalarize_v2i64(<2 x i64*> %p, <2 x i1> %mask, <2 x i64> %pass
define <2 x i64> @scalarize_v2i64_ones_mask(<2 x i64*> %p, <2 x i64> %passthru) { define <2 x i64> @scalarize_v2i64_ones_mask(<2 x i64*> %p, <2 x i64> %passthru) {
; CHECK-LABEL: @scalarize_v2i64_ones_mask( ; CHECK-LABEL: @scalarize_v2i64_ones_mask(
; CHECK-NEXT: [[PTR0:%.*]] = extractelement <2 x i64*> [[P:%.*]], i32 0 ; CHECK-NEXT: [[PTR0:%.*]] = extractelement <2 x i64*> [[P:%.*]], i64 0
; CHECK-NEXT: [[LOAD0:%.*]] = load i64, i64* [[PTR0]], align 8 ; CHECK-NEXT: [[LOAD0:%.*]] = load i64, i64* [[PTR0]], align 8
; CHECK-NEXT: [[RES0:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[LOAD0]], i32 0 ; CHECK-NEXT: [[RES0:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[LOAD0]], i64 0
; CHECK-NEXT: [[PTR1:%.*]] = extractelement <2 x i64*> [[P]], i32 1 ; CHECK-NEXT: [[PTR1:%.*]] = extractelement <2 x i64*> [[P]], i64 1
; CHECK-NEXT: [[LOAD1:%.*]] = load i64, i64* [[PTR1]], align 8 ; CHECK-NEXT: [[LOAD1:%.*]] = load i64, i64* [[PTR1]], align 8
; CHECK-NEXT: [[RES1:%.*]] = insertelement <2 x i64> [[RES0]], i64 [[LOAD1]], i32 1 ; CHECK-NEXT: [[RES1:%.*]] = insertelement <2 x i64> [[RES0]], i64 [[LOAD1]], i64 1
; CHECK-NEXT: ret <2 x i64> [[RES1]] ; CHECK-NEXT: ret <2 x i64> [[RES1]]
; ;
%ret = call <2 x i64> @llvm.masked.gather.v2i64.v2p0i64(<2 x i64*> %p, i32 8, <2 x i1> <i1 true, i1 true>, <2 x i64> %passthru) %ret = call <2 x i64> @llvm.masked.gather.v2i64.v2p0i64(<2 x i64*> %p, i32 8, <2 x i1> <i1 true, i1 true>, <2 x i64> %passthru)
@ -51,9 +51,9 @@ define <2 x i64> @scalarize_v2i64_zero_mask(<2 x i64*> %p, <2 x i64> %passthru)
define <2 x i64> @scalarize_v2i64_const_mask(<2 x i64*> %p, <2 x i64> %passthru) { define <2 x i64> @scalarize_v2i64_const_mask(<2 x i64*> %p, <2 x i64> %passthru) {
; CHECK-LABEL: @scalarize_v2i64_const_mask( ; CHECK-LABEL: @scalarize_v2i64_const_mask(
; CHECK-NEXT: [[PTR1:%.*]] = extractelement <2 x i64*> [[P:%.*]], i32 1 ; CHECK-NEXT: [[PTR1:%.*]] = extractelement <2 x i64*> [[P:%.*]], i64 1
; CHECK-NEXT: [[LOAD1:%.*]] = load i64, i64* [[PTR1]], align 8 ; CHECK-NEXT: [[LOAD1:%.*]] = load i64, i64* [[PTR1]], align 8
; CHECK-NEXT: [[RES1:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[LOAD1]], i32 1 ; CHECK-NEXT: [[RES1:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[LOAD1]], i64 1
; CHECK-NEXT: ret <2 x i64> [[RES1]] ; CHECK-NEXT: ret <2 x i64> [[RES1]]
; ;
%ret = call <2 x i64> @llvm.masked.gather.v2i64.v2p0i64(<2 x i64*> %p, i32 8, <2 x i1> <i1 false, i1 true>, <2 x i64> %passthru) %ret = call <2 x i64> @llvm.masked.gather.v2i64.v2p0i64(<2 x i64*> %p, i32 8, <2 x i1> <i1 false, i1 true>, <2 x i64> %passthru)

26
llvm/test/Transforms/ScalarizeMaskedMemIntrin/X86/expand-masked-load.ll
View File

@ -4,21 +4,21 @@
define <2 x i64> @scalarize_v2i64(<2 x i64>* %p, <2 x i1> %mask, <2 x i64> %passthru) { define <2 x i64> @scalarize_v2i64(<2 x i64>* %p, <2 x i1> %mask, <2 x i64> %passthru) {
; CHECK-LABEL: @scalarize_v2i64( ; CHECK-LABEL: @scalarize_v2i64(
; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64* ; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64*
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i32 0 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
; CHECK-NEXT: br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]] ; CHECK-NEXT: br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
; CHECK: cond.load: ; CHECK: cond.load:
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0
; CHECK-NEXT: [[TMP4:%.*]] = load i64, i64* [[TMP3]], align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i64, i64* [[TMP3]], align 8
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[TMP4]], i32 0 ; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[TMP4]], i64 0
; CHECK-NEXT: br label [[ELSE]] ; CHECK-NEXT: br label [[ELSE]]
; CHECK: else: ; CHECK: else:
; CHECK-NEXT: [[RES_PHI_ELSE:%.*]] = phi <2 x i64> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ] ; CHECK-NEXT: [[RES_PHI_ELSE:%.*]] = phi <2 x i64> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i32 1 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
; CHECK-NEXT: br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]] ; CHECK-NEXT: br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
; CHECK: cond.load1: ; CHECK: cond.load1:
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1 ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1
; CHECK-NEXT: [[TMP8:%.*]] = load i64, i64* [[TMP7]], align 8 ; CHECK-NEXT: [[TMP8:%.*]] = load i64, i64* [[TMP7]], align 8
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x i64> [[RES_PHI_ELSE]], i64 [[TMP8]], i32 1 ; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x i64> [[RES_PHI_ELSE]], i64 [[TMP8]], i64 1
; CHECK-NEXT: br label [[ELSE2]] ; CHECK-NEXT: br label [[ELSE2]]
; CHECK: else2: ; CHECK: else2:
; CHECK-NEXT: [[RES_PHI_ELSE3:%.*]] = phi <2 x i64> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ] ; CHECK-NEXT: [[RES_PHI_ELSE3:%.*]] = phi <2 x i64> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
@ -51,7 +51,7 @@ define <2 x i64> @scalarize_v2i64_const_mask(<2 x i64>* %p, <2 x i64> %passthru)
; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64* ; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64*
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1 ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1
; CHECK-NEXT: [[TMP3:%.*]] = load i64, i64* [[TMP2]], align 8 ; CHECK-NEXT: [[TMP3:%.*]] = load i64, i64* [[TMP2]], align 8
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[TMP3]], i32 1 ; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i64> [[PASSTHRU:%.*]], i64 [[TMP3]], i64 1
; CHECK-NEXT: ret <2 x i64> [[TMP4]] ; CHECK-NEXT: ret <2 x i64> [[TMP4]]
; ;
%ret = call <2 x i64> @llvm.masked.load.v2i64.p0v2i64(<2 x i64>* %p, i32 8, <2 x i1> <i1 false, i1 true>, <2 x i64> %passthru) %ret = call <2 x i64> @llvm.masked.load.v2i64.p0v2i64(<2 x i64>* %p, i32 8, <2 x i1> <i1 false, i1 true>, <2 x i64> %passthru)
@ -62,21 +62,21 @@ define <2 x i64> @scalarize_v2i64_const_mask(<2 x i64>* %p, <2 x i64> %passthru)
define <2 x i24> @scalarize_v2i24(<2 x i24>* %p, <2 x i1> %mask, <2 x i24> %passthru) { define <2 x i24> @scalarize_v2i24(<2 x i24>* %p, <2 x i1> %mask, <2 x i24> %passthru) {
; CHECK-LABEL: @scalarize_v2i24( ; CHECK-LABEL: @scalarize_v2i24(
; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i24>* [[P:%.*]] to i24* ; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i24>* [[P:%.*]] to i24*
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i32 0 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
; CHECK-NEXT: br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]] ; CHECK-NEXT: br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
; CHECK: cond.load: ; CHECK: cond.load:
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i24, i24* [[TMP1]], i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i24, i24* [[TMP1]], i32 0
; CHECK-NEXT: [[TMP4:%.*]] = load i24, i24* [[TMP3]], align 1 ; CHECK-NEXT: [[TMP4:%.*]] = load i24, i24* [[TMP3]], align 1
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i24> [[PASSTHRU:%.*]], i24 [[TMP4]], i32 0 ; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i24> [[PASSTHRU:%.*]], i24 [[TMP4]], i64 0
; CHECK-NEXT: br label [[ELSE]] ; CHECK-NEXT: br label [[ELSE]]
; CHECK: else: ; CHECK: else:
; CHECK-NEXT: [[RES_PHI_ELSE:%.*]] = phi <2 x i24> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ] ; CHECK-NEXT: [[RES_PHI_ELSE:%.*]] = phi <2 x i24> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i32 1 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
; CHECK-NEXT: br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]] ; CHECK-NEXT: br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
; CHECK: cond.load1: ; CHECK: cond.load1:
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i24, i24* [[TMP1]], i32 1 ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i24, i24* [[TMP1]], i32 1
; CHECK-NEXT: [[TMP8:%.*]] = load i24, i24* [[TMP7]], align 1 ; CHECK-NEXT: [[TMP8:%.*]] = load i24, i24* [[TMP7]], align 1
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x i24> [[RES_PHI_ELSE]], i24 [[TMP8]], i32 1 ; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x i24> [[RES_PHI_ELSE]], i24 [[TMP8]], i64 1
; CHECK-NEXT: br label [[ELSE2]] ; CHECK-NEXT: br label [[ELSE2]]
; CHECK: else2: ; CHECK: else2:
; CHECK-NEXT: [[RES_PHI_ELSE3:%.*]] = phi <2 x i24> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ] ; CHECK-NEXT: [[RES_PHI_ELSE3:%.*]] = phi <2 x i24> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]
@ -90,21 +90,21 @@ define <2 x i24> @scalarize_v2i24(<2 x i24>* %p, <2 x i1> %mask, <2 x i24> %pass
define <2 x i48> @scalarize_v2i48(<2 x i48>* %p, <2 x i1> %mask, <2 x i48> %passthru) { define <2 x i48> @scalarize_v2i48(<2 x i48>* %p, <2 x i1> %mask, <2 x i48> %passthru) {
; CHECK-LABEL: @scalarize_v2i48( ; CHECK-LABEL: @scalarize_v2i48(
; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i48>* [[P:%.*]] to i48* ; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i48>* [[P:%.*]] to i48*
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i32 0 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
; CHECK-NEXT: br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]] ; CHECK-NEXT: br i1 [[TMP2]], label [[COND_LOAD:%.*]], label [[ELSE:%.*]]
; CHECK: cond.load: ; CHECK: cond.load:
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i48, i48* [[TMP1]], i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i48, i48* [[TMP1]], i32 0
; CHECK-NEXT: [[TMP4:%.*]] = load i48, i48* [[TMP3]], align 2 ; CHECK-NEXT: [[TMP4:%.*]] = load i48, i48* [[TMP3]], align 2
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i48> [[PASSTHRU:%.*]], i48 [[TMP4]], i32 0 ; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i48> [[PASSTHRU:%.*]], i48 [[TMP4]], i64 0
; CHECK-NEXT: br label [[ELSE]] ; CHECK-NEXT: br label [[ELSE]]
; CHECK: else: ; CHECK: else:
; CHECK-NEXT: [[RES_PHI_ELSE:%.*]] = phi <2 x i48> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ] ; CHECK-NEXT: [[RES_PHI_ELSE:%.*]] = phi <2 x i48> [ [[TMP5]], [[COND_LOAD]] ], [ [[PASSTHRU]], [[TMP0:%.*]] ]
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i32 1 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
; CHECK-NEXT: br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]] ; CHECK-NEXT: br i1 [[TMP6]], label [[COND_LOAD1:%.*]], label [[ELSE2:%.*]]
; CHECK: cond.load1: ; CHECK: cond.load1:
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i48, i48* [[TMP1]], i32 1 ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i48, i48* [[TMP1]], i32 1
; CHECK-NEXT: [[TMP8:%.*]] = load i48, i48* [[TMP7]], align 2 ; CHECK-NEXT: [[TMP8:%.*]] = load i48, i48* [[TMP7]], align 2
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x i48> [[RES_PHI_ELSE]], i48 [[TMP8]], i32 1 ; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x i48> [[RES_PHI_ELSE]], i48 [[TMP8]], i64 1
; CHECK-NEXT: br label [[ELSE2]] ; CHECK-NEXT: br label [[ELSE2]]
; CHECK: else2: ; CHECK: else2:
; CHECK-NEXT: [[RES_PHI_ELSE3:%.*]] = phi <2 x i48> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ] ; CHECK-NEXT: [[RES_PHI_ELSE3:%.*]] = phi <2 x i48> [ [[TMP9]], [[COND_LOAD1]] ], [ [[RES_PHI_ELSE]], [[ELSE]] ]

10
llvm/test/Transforms/ScalarizeMaskedMemIntrin/X86/expand-masked-store.ll
View File

@ -4,18 +4,18 @@
define void @scalarize_v2i64(<2 x i64>* %p, <2 x i1> %mask, <2 x i64> %data) { define void @scalarize_v2i64(<2 x i64>* %p, <2 x i1> %mask, <2 x i64> %data) {
; CHECK-LABEL: @scalarize_v2i64( ; CHECK-LABEL: @scalarize_v2i64(
; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64* ; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64*
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i32 0 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[MASK:%.*]], i64 0
; CHECK-NEXT: br i1 [[TMP2]], label [[COND_STORE:%.*]], label [[ELSE:%.*]] ; CHECK-NEXT: br i1 [[TMP2]], label [[COND_STORE:%.*]], label [[ELSE:%.*]]
; CHECK: cond.store: ; CHECK: cond.store:
; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i64> [[DATA:%.*]], i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i64> [[DATA:%.*]], i64 0
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0 ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0
; CHECK-NEXT: store i64 [[TMP3]], i64* [[TMP4]], align 8 ; CHECK-NEXT: store i64 [[TMP3]], i64* [[TMP4]], align 8
; CHECK-NEXT: br label [[ELSE]] ; CHECK-NEXT: br label [[ELSE]]
; CHECK: else: ; CHECK: else:
; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[MASK]], i32 1 ; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[MASK]], i64 1
; CHECK-NEXT: br i1 [[TMP5]], label [[COND_STORE1:%.*]], label [[ELSE2:%.*]] ; CHECK-NEXT: br i1 [[TMP5]], label [[COND_STORE1:%.*]], label [[ELSE2:%.*]]
; CHECK: cond.store1: ; CHECK: cond.store1:
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i64> [[DATA]], i32 1 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i64> [[DATA]], i64 1
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1 ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1
; CHECK-NEXT: store i64 [[TMP6]], i64* [[TMP7]], align 8 ; CHECK-NEXT: store i64 [[TMP6]], i64* [[TMP7]], align 8
; CHECK-NEXT: br label [[ELSE2]] ; CHECK-NEXT: br label [[ELSE2]]
@ -47,7 +47,7 @@ define void @scalarize_v2i64_zero_mask(<2 x i64>* %p, <2 x i64> %data) {
define void @scalarize_v2i64_const_mask(<2 x i64>* %p, <2 x i64> %data) { define void @scalarize_v2i64_const_mask(<2 x i64>* %p, <2 x i64> %data) {
; CHECK-LABEL: @scalarize_v2i64_const_mask( ; CHECK-LABEL: @scalarize_v2i64_const_mask(
; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64* ; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64>* [[P:%.*]] to i64*
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[DATA:%.*]], i32 1 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[DATA:%.*]], i64 1
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 1
; CHECK-NEXT: store i64 [[TMP2]], i64* [[TMP3]], align 8 ; CHECK-NEXT: store i64 [[TMP2]], i64* [[TMP3]], align 8
; CHECK-NEXT: ret void ; CHECK-NEXT: ret void