From 9581b42589d72d1c931c4bcbc0101df7c507e27e Mon Sep 17 00:00:00 2001 From: Alexey Bataev Date: Mon, 7 Aug 2017 15:25:49 +0000 Subject: [PATCH] [SLP] General improvements of SLP vectorization process. Patch tries to improve two-pass vectorization analysis, existing in SLP vectorizer. What it does: 1. Defines key nodes, that are the vectorization roots. Previously vectorization started if StoreInst or ReturnInst is found. For now, the vectorization started for all Instructions with no users and void types (Terminators, StoreInst) + CallInsts. 2. CmpInsts, InsertElementInsts and InsertValueInsts are stored in the array. This array is processed only after the vectorization of the first-after-these instructions key node is finished. Vectorization goes in reverse order to try to vectorize as much code as possible. Reviewers: mzolotukhin, Ayal, mkuper, gilr, hfinkel, RKSimon Subscribers: ashahid, anemet, RKSimon, mssimpso, llvm-commits Differential Revision: https://reviews.llvm.org/D29826 llvm-svn: 310260 --- .../llvm/Transforms/Vectorize/SLPVectorizer.h | 13 ++ llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp | 213 +++++++++++---------- .../SLPVectorizer/AArch64/gather-root.ll | 100 +++++----- .../Transforms/SLPVectorizer/X86/horizontal.ll | 64 +++---- .../X86/insert-element-build-vector.ll | 40 ++-- 5 files changed, 224 insertions(+), 206 deletions(-) diff --git a/llvm/include/llvm/Transforms/Vectorize/SLPVectorizer.h b/llvm/include/llvm/Transforms/Vectorize/SLPVectorizer.h index f7c8cd4..a4fa3f3 100644 --- a/llvm/include/llvm/Transforms/Vectorize/SLPVectorizer.h +++ b/llvm/include/llvm/Transforms/Vectorize/SLPVectorizer.h @@ -100,6 +100,19 @@ private: slpvectorizer::BoUpSLP &R, TargetTransformInfo *TTI); + /// Try to vectorize trees that start at insertvalue instructions. + bool vectorizeInsertValueInst(InsertValueInst *IVI, BasicBlock *BB, + slpvectorizer::BoUpSLP &R); + /// Try to vectorize trees that start at insertelement instructions. + bool vectorizeInsertElementInst(InsertElementInst *IEI, BasicBlock *BB, + slpvectorizer::BoUpSLP &R); + /// Try to vectorize trees that start at compare instructions. + bool vectorizeCmpInst(CmpInst *CI, BasicBlock *BB, slpvectorizer::BoUpSLP &R); + /// Tries to vectorize constructs started from CmpInst, InsertValueInst or + /// InsertElementInst instructions. + bool vectorizeSimpleInstructions(SmallVectorImpl &Instructions, + BasicBlock *BB, slpvectorizer::BoUpSLP &R); + /// \brief Scan the basic block and look for patterns that are likely to start /// a vectorization chain. bool vectorizeChainsInBlock(BasicBlock *BB, slpvectorizer::BoUpSLP &R); diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp index 6e2865e..d5f4eae 100644 --- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp +++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp @@ -4387,7 +4387,7 @@ bool SLPVectorizerPass::tryToVectorize(Instruction *I, BoUpSLP &R) { if (!I) return false; - if (!isa(I)) + if (!isa(I) && !isa(I)) return false; Value *P = I->getParent(); @@ -4925,39 +4925,30 @@ private: /// %rb = insertelement <4 x float> %ra, float %s1, i32 1 /// %rc = insertelement <4 x float> %rb, float %s2, i32 2 /// %rd = insertelement <4 x float> %rc, float %s3, i32 3 +/// starting from the last insertelement instruction. /// /// Returns true if it matches /// -static bool findBuildVector(InsertElementInst *FirstInsertElem, +static bool findBuildVector(InsertElementInst *LastInsertElem, SmallVectorImpl &BuildVector, SmallVectorImpl &BuildVectorOpds) { - if (!isa(FirstInsertElem->getOperand(0))) - return false; - - InsertElementInst *IE = FirstInsertElem; - while (true) { - BuildVector.push_back(IE); - BuildVectorOpds.push_back(IE->getOperand(1)); - - if (IE->use_empty()) - return false; - - InsertElementInst *NextUse = dyn_cast(IE->user_back()); - if (!NextUse) - return true; - - // If this isn't the final use, make sure the next insertelement is the only - // use. It's OK if the final constructed vector is used multiple times - if (!IE->hasOneUse()) + Value *V = nullptr; + do { + BuildVector.push_back(LastInsertElem); + BuildVectorOpds.push_back(LastInsertElem->getOperand(1)); + V = LastInsertElem->getOperand(0); + if (isa(V)) + break; + LastInsertElem = dyn_cast(V); + if (!LastInsertElem || !LastInsertElem->hasOneUse()) return false; - - IE = NextUse; - } - - return false; + } while (true); + std::reverse(BuildVector.begin(), BuildVector.end()); + std::reverse(BuildVectorOpds.begin(), BuildVectorOpds.end()); + return true; } -/// \brief Like findBuildVector, but looks backwards for construction of aggregate. +/// \brief Like findBuildVector, but looks for construction of aggregate. /// /// \return true if it matches. static bool findBuildAggregate(InsertValueInst *IV, @@ -5142,6 +5133,64 @@ bool SLPVectorizerPass::vectorizeRootInstruction(PHINode *P, Value *V, ExtraVectorization); } +bool SLPVectorizerPass::vectorizeInsertValueInst(InsertValueInst *IVI, + BasicBlock *BB, BoUpSLP &R) { + const DataLayout &DL = BB->getModule()->getDataLayout(); + if (!R.canMapToVector(IVI->getType(), DL)) + return false; + + SmallVector BuildVector; + SmallVector BuildVectorOpds; + if (!findBuildAggregate(IVI, BuildVector, BuildVectorOpds)) + return false; + + DEBUG(dbgs() << "SLP: array mappable to vector: " << *IVI << "\n"); + return tryToVectorizeList(BuildVectorOpds, R, BuildVector, false); +} + +bool SLPVectorizerPass::vectorizeInsertElementInst(InsertElementInst *IEI, + BasicBlock *BB, BoUpSLP &R) { + SmallVector BuildVector; + SmallVector BuildVectorOpds; + if (!findBuildVector(IEI, BuildVector, BuildVectorOpds)) + return false; + + // Vectorize starting with the build vector operands ignoring the BuildVector + // instructions for the purpose of scheduling and user extraction. + return tryToVectorizeList(BuildVectorOpds, R, BuildVector); +} + +bool SLPVectorizerPass::vectorizeCmpInst(CmpInst *CI, BasicBlock *BB, + BoUpSLP &R) { + if (tryToVectorizePair(CI->getOperand(0), CI->getOperand(1), R)) + return true; + + bool OpsChanged = false; + for (int Idx = 0; Idx < 2; ++Idx) { + OpsChanged |= + vectorizeRootInstruction(nullptr, CI->getOperand(Idx), BB, R, TTI); + } + return OpsChanged; +} + +bool SLPVectorizerPass::vectorizeSimpleInstructions( + SmallVectorImpl &Instructions, BasicBlock *BB, BoUpSLP &R) { + bool OpsChanged = false; + for (auto &VH : reverse(Instructions)) { + auto *I = dyn_cast_or_null(VH); + if (!I) + continue; + if (auto *LastInsertValue = dyn_cast(I)) + OpsChanged |= vectorizeInsertValueInst(LastInsertValue, BB, R); + else if (auto *LastInsertElem = dyn_cast(I)) + OpsChanged |= vectorizeInsertElementInst(LastInsertElem, BB, R); + else if (auto *CI = dyn_cast(I)) + OpsChanged |= vectorizeCmpInst(CI, BB, R); + } + Instructions.clear(); + return OpsChanged; +} + bool SLPVectorizerPass::vectorizeChainsInBlock(BasicBlock *BB, BoUpSLP &R) { bool Changed = false; SmallVector Incoming; @@ -5201,10 +5250,21 @@ bool SLPVectorizerPass::vectorizeChainsInBlock(BasicBlock *BB, BoUpSLP &R) { VisitedInstrs.clear(); + SmallVector PostProcessInstructions; + SmallDenseSet KeyNodes; for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; it++) { // We may go through BB multiple times so skip the one we have checked. - if (!VisitedInstrs.insert(&*it).second) + if (!VisitedInstrs.insert(&*it).second) { + if (it->use_empty() && KeyNodes.count(&*it) > 0 && + vectorizeSimpleInstructions(PostProcessInstructions, BB, R)) { + // We would like to start over since some instructions are deleted + // and the iterator may become invalid value. + Changed = true; + it = BB->begin(); + e = BB->end(); + } continue; + } if (isa(it)) continue; @@ -5226,96 +5286,37 @@ bool SLPVectorizerPass::vectorizeChainsInBlock(BasicBlock *BB, BoUpSLP &R) { continue; } - if (ShouldStartVectorizeHorAtStore) { - if (StoreInst *SI = dyn_cast(it)) { - // Try to match and vectorize a horizontal reduction. - if (vectorizeRootInstruction(nullptr, SI->getValueOperand(), BB, R, - TTI)) { - Changed = true; - it = BB->begin(); - e = BB->end(); - continue; + // Ran into an instruction without users, like terminator, or function call + // with ignored return value, store. Ignore unused instructions (basing on + // instruction type, except for CallInst and InvokeInst). + if (it->use_empty() && (it->getType()->isVoidTy() || isa(it) || + isa(it))) { + KeyNodes.insert(&*it); + bool OpsChanged = false; + if (ShouldStartVectorizeHorAtStore || !isa(it)) { + for (auto *V : it->operand_values()) { + // Try to match and vectorize a horizontal reduction. + OpsChanged |= vectorizeRootInstruction(nullptr, V, BB, R, TTI); } } - } - - // Try to vectorize horizontal reductions feeding into a return. - if (ReturnInst *RI = dyn_cast(it)) { - if (RI->getNumOperands() != 0) { - // Try to match and vectorize a horizontal reduction. - if (vectorizeRootInstruction(nullptr, RI->getOperand(0), BB, R, TTI)) { - Changed = true; - it = BB->begin(); - e = BB->end(); - continue; - } - } - } - - // Try to vectorize trees that start at compare instructions. - if (CmpInst *CI = dyn_cast(it)) { - if (tryToVectorizePair(CI->getOperand(0), CI->getOperand(1), R)) { - Changed = true; + // Start vectorization of post-process list of instructions from the + // top-tree instructions to try to vectorize as many instructions as + // possible. + OpsChanged |= vectorizeSimpleInstructions(PostProcessInstructions, BB, R); + if (OpsChanged) { // We would like to start over since some instructions are deleted // and the iterator may become invalid value. - it = BB->begin(); - e = BB->end(); - continue; - } - - for (int I = 0; I < 2; ++I) { - if (vectorizeRootInstruction(nullptr, CI->getOperand(I), BB, R, TTI)) { - Changed = true; - // We would like to start over since some instructions are deleted - // and the iterator may become invalid value. - it = BB->begin(); - e = BB->end(); - break; - } - } - continue; - } - - // Try to vectorize trees that start at insertelement instructions. - if (InsertElementInst *FirstInsertElem = dyn_cast(it)) { - SmallVector BuildVector; - SmallVector BuildVectorOpds; - if (!findBuildVector(FirstInsertElem, BuildVector, BuildVectorOpds)) - continue; - - // Vectorize starting with the build vector operands ignoring the - // BuildVector instructions for the purpose of scheduling and user - // extraction. - if (tryToVectorizeList(BuildVectorOpds, R, BuildVector)) { Changed = true; it = BB->begin(); e = BB->end(); + continue; } - - continue; } - // Try to vectorize trees that start at insertvalue instructions feeding into - // a store. - if (StoreInst *SI = dyn_cast(it)) { - if (InsertValueInst *LastInsertValue = dyn_cast(SI->getValueOperand())) { - const DataLayout &DL = BB->getModule()->getDataLayout(); - if (R.canMapToVector(SI->getValueOperand()->getType(), DL)) { - SmallVector BuildVector; - SmallVector BuildVectorOpds; - if (!findBuildAggregate(LastInsertValue, BuildVector, BuildVectorOpds)) - continue; + if (isa(it) || isa(it) || + isa(it)) + PostProcessInstructions.push_back(&*it); - DEBUG(dbgs() << "SLP: store of array mappable to vector: " << *SI << "\n"); - if (tryToVectorizeList(BuildVectorOpds, R, BuildVector, false)) { - Changed = true; - it = BB->begin(); - e = BB->end(); - } - continue; - } - } - } } return Changed; diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/gather-root.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/gather-root.ll index 0ea08b7..dda8a0c 100644 --- a/llvm/test/Transforms/SLPVectorizer/AArch64/gather-root.ll +++ b/llvm/test/Transforms/SLPVectorizer/AArch64/gather-root.ll @@ -31,10 +31,8 @@ define void @PR28330(i32 %n) { ; ; GATHER-LABEL: @PR28330( ; GATHER-NEXT: entry: -; GATHER-NEXT: [[TMP0:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1), align 1 -; GATHER-NEXT: [[TMP1:%.*]] = icmp eq i8 [[TMP0]], 0 -; GATHER-NEXT: [[TMP2:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 2), align 2 -; GATHER-NEXT: [[TMP3:%.*]] = icmp eq i8 [[TMP2]], 0 +; GATHER-NEXT: [[TMP0:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <2 x i8>*), align 1 +; GATHER-NEXT: [[TMP1:%.*]] = icmp eq <2 x i8> [[TMP0]], zeroinitializer ; GATHER-NEXT: [[TMP4:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 3), align 1 ; GATHER-NEXT: [[TMP5:%.*]] = icmp eq i8 [[TMP4]], 0 ; GATHER-NEXT: [[TMP6:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 4), align 4 @@ -50,10 +48,11 @@ define void @PR28330(i32 %n) { ; GATHER-NEXT: br label [[FOR_BODY:%.*]] ; GATHER: for.body: ; GATHER-NEXT: [[TMP17:%.*]] = phi i32 [ [[BIN_EXTRA:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ] -; GATHER-NEXT: [[TMP19:%.*]] = select i1 [[TMP1]], i32 -720, i32 -80 -; GATHER-NEXT: [[TMP20:%.*]] = add i32 [[TMP17]], [[TMP19]] -; GATHER-NEXT: [[TMP21:%.*]] = select i1 [[TMP3]], i32 -720, i32 -80 -; GATHER-NEXT: [[TMP22:%.*]] = add i32 [[TMP20]], [[TMP21]] +; GATHER-NEXT: [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x i32> , <2 x i32> +; GATHER-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0 +; GATHER-NEXT: [[TMP20:%.*]] = add i32 [[TMP17]], [[TMP3]] +; GATHER-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1 +; GATHER-NEXT: [[TMP22:%.*]] = add i32 [[TMP20]], [[TMP4]] ; GATHER-NEXT: [[TMP23:%.*]] = select i1 [[TMP5]], i32 -720, i32 -80 ; GATHER-NEXT: [[TMP24:%.*]] = add i32 [[TMP22]], [[TMP23]] ; GATHER-NEXT: [[TMP25:%.*]] = select i1 [[TMP7]], i32 -720, i32 -80 @@ -65,16 +64,16 @@ define void @PR28330(i32 %n) { ; GATHER-NEXT: [[TMP31:%.*]] = select i1 [[TMP13]], i32 -720, i32 -80 ; GATHER-NEXT: [[TMP32:%.*]] = add i32 [[TMP30]], [[TMP31]] ; GATHER-NEXT: [[TMP33:%.*]] = select i1 [[TMP15]], i32 -720, i32 -80 -; GATHER-NEXT: [[TMP0:%.*]] = insertelement <8 x i32> undef, i32 [[TMP19]], i32 0 -; GATHER-NEXT: [[TMP1:%.*]] = insertelement <8 x i32> [[TMP0]], i32 [[TMP21]], i32 1 -; GATHER-NEXT: [[TMP2:%.*]] = insertelement <8 x i32> [[TMP1]], i32 [[TMP23]], i32 2 -; GATHER-NEXT: [[TMP3:%.*]] = insertelement <8 x i32> [[TMP2]], i32 [[TMP25]], i32 3 -; GATHER-NEXT: [[TMP4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[TMP27]], i32 4 -; GATHER-NEXT: [[TMP5:%.*]] = insertelement <8 x i32> [[TMP4]], i32 [[TMP29]], i32 5 -; GATHER-NEXT: [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[TMP31]], i32 6 -; GATHER-NEXT: [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[TMP33]], i32 7 -; GATHER-NEXT: [[TMP8:%.*]] = call i32 @llvm.experimental.vector.reduce.add.i32.v8i32(<8 x i32> [[TMP7]]) -; GATHER-NEXT: [[BIN_EXTRA]] = add i32 [[TMP8]], [[TMP17]] +; GATHER-NEXT: [[TMP5:%.*]] = insertelement <8 x i32> undef, i32 [[TMP3]], i32 0 +; GATHER-NEXT: [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[TMP4]], i32 1 +; GATHER-NEXT: [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[TMP23]], i32 2 +; GATHER-NEXT: [[TMP8:%.*]] = insertelement <8 x i32> [[TMP7]], i32 [[TMP25]], i32 3 +; GATHER-NEXT: [[TMP9:%.*]] = insertelement <8 x i32> [[TMP8]], i32 [[TMP27]], i32 4 +; GATHER-NEXT: [[TMP10:%.*]] = insertelement <8 x i32> [[TMP9]], i32 [[TMP29]], i32 5 +; GATHER-NEXT: [[TMP11:%.*]] = insertelement <8 x i32> [[TMP10]], i32 [[TMP31]], i32 6 +; GATHER-NEXT: [[TMP12:%.*]] = insertelement <8 x i32> [[TMP11]], i32 [[TMP33]], i32 7 +; GATHER-NEXT: [[TMP13:%.*]] = call i32 @llvm.experimental.vector.reduce.add.i32.v8i32(<8 x i32> [[TMP12]]) +; GATHER-NEXT: [[BIN_EXTRA]] = add i32 [[TMP13]], [[TMP17]] ; GATHER-NEXT: [[TMP34:%.*]] = add i32 [[TMP32]], [[TMP33]] ; GATHER-NEXT: br label [[FOR_BODY]] ; @@ -180,10 +179,8 @@ define void @PR32038(i32 %n) { ; ; GATHER-LABEL: @PR32038( ; GATHER-NEXT: entry: -; GATHER-NEXT: [[TMP0:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1), align 1 -; GATHER-NEXT: [[TMP1:%.*]] = icmp eq i8 [[TMP0]], 0 -; GATHER-NEXT: [[TMP2:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 2), align 2 -; GATHER-NEXT: [[TMP3:%.*]] = icmp eq i8 [[TMP2]], 0 +; GATHER-NEXT: [[TMP0:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <2 x i8>*), align 1 +; GATHER-NEXT: [[TMP1:%.*]] = icmp eq <2 x i8> [[TMP0]], zeroinitializer ; GATHER-NEXT: [[TMP4:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 3), align 1 ; GATHER-NEXT: [[TMP5:%.*]] = icmp eq i8 [[TMP4]], 0 ; GATHER-NEXT: [[TMP6:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 4), align 4 @@ -199,10 +196,11 @@ define void @PR32038(i32 %n) { ; GATHER-NEXT: br label [[FOR_BODY:%.*]] ; GATHER: for.body: ; GATHER-NEXT: [[TMP17:%.*]] = phi i32 [ [[BIN_EXTRA:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ] -; GATHER-NEXT: [[TMP19:%.*]] = select i1 [[TMP1]], i32 -720, i32 -80 -; GATHER-NEXT: [[TMP20:%.*]] = add i32 -5, [[TMP19]] -; GATHER-NEXT: [[TMP21:%.*]] = select i1 [[TMP3]], i32 -720, i32 -80 -; GATHER-NEXT: [[TMP22:%.*]] = add i32 [[TMP20]], [[TMP21]] +; GATHER-NEXT: [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x i32> , <2 x i32> +; GATHER-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0 +; GATHER-NEXT: [[TMP20:%.*]] = add i32 -5, [[TMP3]] +; GATHER-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1 +; GATHER-NEXT: [[TMP22:%.*]] = add i32 [[TMP20]], [[TMP4]] ; GATHER-NEXT: [[TMP23:%.*]] = select i1 [[TMP5]], i32 -720, i32 -80 ; GATHER-NEXT: [[TMP24:%.*]] = add i32 [[TMP22]], [[TMP23]] ; GATHER-NEXT: [[TMP25:%.*]] = select i1 [[TMP7]], i32 -720, i32 -80 @@ -214,29 +212,27 @@ define void @PR32038(i32 %n) { ; GATHER-NEXT: [[TMP31:%.*]] = select i1 [[TMP13]], i32 -720, i32 -80 ; GATHER-NEXT: [[TMP32:%.*]] = add i32 [[TMP30]], [[TMP31]] ; GATHER-NEXT: [[TMP33:%.*]] = select i1 [[TMP15]], i32 -720, i32 -80 -; GATHER-NEXT: [[TMP0:%.*]] = insertelement <8 x i32> undef, i32 [[TMP19]], i32 0 -; GATHER-NEXT: [[TMP1:%.*]] = insertelement <8 x i32> [[TMP0]], i32 [[TMP21]], i32 1 -; GATHER-NEXT: [[TMP2:%.*]] = insertelement <8 x i32> [[TMP1]], i32 [[TMP23]], i32 2 -; GATHER-NEXT: [[TMP3:%.*]] = insertelement <8 x i32> [[TMP2]], i32 [[TMP25]], i32 3 -; GATHER-NEXT: [[TMP4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[TMP27]], i32 4 -; GATHER-NEXT: [[TMP5:%.*]] = insertelement <8 x i32> [[TMP4]], i32 [[TMP29]], i32 5 -; GATHER-NEXT: [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[TMP31]], i32 6 -; GATHER-NEXT: [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[TMP33]], i32 7 -; GATHER-NEXT: [[TMP8:%.*]] = call i32 @llvm.experimental.vector.reduce.add.i32.v8i32(<8 x i32> [[TMP7]]) -; GATHER-NEXT: [[BIN_EXTRA]] = add i32 [[TMP8]], -5 +; GATHER-NEXT: [[TMP5:%.*]] = insertelement <8 x i32> undef, i32 [[TMP3]], i32 0 +; GATHER-NEXT: [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[TMP4]], i32 1 +; GATHER-NEXT: [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[TMP23]], i32 2 +; GATHER-NEXT: [[TMP8:%.*]] = insertelement <8 x i32> [[TMP7]], i32 [[TMP25]], i32 3 +; GATHER-NEXT: [[TMP9:%.*]] = insertelement <8 x i32> [[TMP8]], i32 [[TMP27]], i32 4 +; GATHER-NEXT: [[TMP10:%.*]] = insertelement <8 x i32> [[TMP9]], i32 [[TMP29]], i32 5 +; GATHER-NEXT: [[TMP11:%.*]] = insertelement <8 x i32> [[TMP10]], i32 [[TMP31]], i32 6 +; GATHER-NEXT: [[TMP12:%.*]] = insertelement <8 x i32> [[TMP11]], i32 [[TMP33]], i32 7 +; GATHER-NEXT: [[TMP13:%.*]] = call i32 @llvm.experimental.vector.reduce.add.i32.v8i32(<8 x i32> [[TMP12]]) +; GATHER-NEXT: [[BIN_EXTRA]] = add i32 [[TMP13]], -5 ; GATHER-NEXT: [[TMP34:%.*]] = add i32 [[TMP32]], [[TMP33]] ; GATHER-NEXT: br label [[FOR_BODY]] ; ; MAX-COST-LABEL: @PR32038( ; MAX-COST-NEXT: entry: -; MAX-COST-NEXT: [[TMP0:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1), align 1 -; MAX-COST-NEXT: [[TMP1:%.*]] = icmp eq i8 [[TMP0]], 0 -; MAX-COST-NEXT: [[TMP2:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 2), align 2 -; MAX-COST-NEXT: [[TMP3:%.*]] = icmp eq i8 [[TMP2]], 0 +; MAX-COST-NEXT: [[TMP0:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 1) to <2 x i8>*), align 1 +; MAX-COST-NEXT: [[TMP1:%.*]] = icmp eq <2 x i8> [[TMP0]], zeroinitializer ; MAX-COST-NEXT: [[TMP4:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 3), align 1 -; MAX-COST-NEXT: [[TMP5:%.*]] = icmp eq i8 [[TMP4]], 0 +; MAX-COST-NEXT: [[TMPP5:%.*]] = icmp eq i8 [[TMP4]], 0 ; MAX-COST-NEXT: [[TMP6:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 4), align 4 -; MAX-COST-NEXT: [[TMP7:%.*]] = icmp eq i8 [[TMP6]], 0 +; MAX-COST-NEXT: [[TMPP7:%.*]] = icmp eq i8 [[TMP6]], 0 ; MAX-COST-NEXT: [[TMP8:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 5), align 1 ; MAX-COST-NEXT: [[TMP9:%.*]] = icmp eq i8 [[TMP8]], 0 ; MAX-COST-NEXT: [[TMP10:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 6), align 2 @@ -245,14 +241,16 @@ define void @PR32038(i32 %n) { ; MAX-COST-NEXT: [[TMP13:%.*]] = icmp eq i8 [[TMP12]], 0 ; MAX-COST-NEXT: [[TMP14:%.*]] = load i8, i8* getelementptr inbounds ([80 x i8], [80 x i8]* @a, i64 0, i64 8), align 8 ; MAX-COST-NEXT: [[TMP15:%.*]] = icmp eq i8 [[TMP14]], 0 -; MAX-COST-NEXT: [[TMP0:%.*]] = insertelement <4 x i1> undef, i1 [[TMP1]], i32 0 -; MAX-COST-NEXT: [[TMP1:%.*]] = insertelement <4 x i1> [[TMP0]], i1 [[TMP3]], i32 1 -; MAX-COST-NEXT: [[TMP2:%.*]] = insertelement <4 x i1> [[TMP1]], i1 [[TMP5]], i32 2 -; MAX-COST-NEXT: [[TMP3:%.*]] = insertelement <4 x i1> [[TMP2]], i1 [[TMP7]], i32 3 ; MAX-COST-NEXT: br label [[FOR_BODY:%.*]] ; MAX-COST: for.body: ; MAX-COST-NEXT: [[TMP17:%.*]] = phi i32 [ [[TMP34:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ] -; MAX-COST-NEXT: [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x i32> , <4 x i32> +; MAX-COST-NEXT: [[TMP2:%.*]] = extractelement <2 x i1> [[TMP1]], i32 0 +; MAX-COST-NEXT: [[TMP3:%.*]] = insertelement <4 x i1> undef, i1 [[TMP2]], i32 0 +; MAX-COST-NEXT: [[TMP4:%.*]] = extractelement <2 x i1> [[TMP1]], i32 1 +; MAX-COST-NEXT: [[TMP5:%.*]] = insertelement <4 x i1> [[TMP3]], i1 [[TMP4]], i32 1 +; MAX-COST-NEXT: [[TMP6:%.*]] = insertelement <4 x i1> [[TMP5]], i1 [[TMPP5]], i32 2 +; MAX-COST-NEXT: [[TMP7:%.*]] = insertelement <4 x i1> [[TMP6]], i1 [[TMPP7]], i32 3 +; MAX-COST-NEXT: [[TMP8:%.*]] = select <4 x i1> [[TMP7]], <4 x i32> , <4 x i32> ; MAX-COST-NEXT: [[TMP20:%.*]] = add i32 -5, undef ; MAX-COST-NEXT: [[TMP22:%.*]] = add i32 [[TMP20]], undef ; MAX-COST-NEXT: [[TMP24:%.*]] = add i32 [[TMP22]], undef @@ -260,10 +258,10 @@ define void @PR32038(i32 %n) { ; MAX-COST-NEXT: [[TMP27:%.*]] = select i1 [[TMP9]], i32 -720, i32 -80 ; MAX-COST-NEXT: [[TMP28:%.*]] = add i32 [[TMP26]], [[TMP27]] ; MAX-COST-NEXT: [[TMP29:%.*]] = select i1 [[TMP11]], i32 -720, i32 -80 -; MAX-COST-NEXT: [[TMP5:%.*]] = call i32 @llvm.experimental.vector.reduce.add.i32.v4i32(<4 x i32> [[TMP4]]) -; MAX-COST-NEXT: [[TMP6:%.*]] = add i32 [[TMP5]], [[TMP27]] -; MAX-COST-NEXT: [[TMP7:%.*]] = add i32 [[TMP6]], [[TMP29]] -; MAX-COST-NEXT: [[BIN_EXTRA:%.*]] = add i32 [[TMP7]], -5 +; MAX-COST-NEXT: [[TMP9:%.*]] = call i32 @llvm.experimental.vector.reduce.add.i32.v4i32(<4 x i32> [[TMP8]]) +; MAX-COST-NEXT: [[TMP10:%.*]] = add i32 [[TMP9]], [[TMP27]] +; MAX-COST-NEXT: [[TMP11:%.*]] = add i32 [[TMP10]], [[TMP29]] +; MAX-COST-NEXT: [[BIN_EXTRA:%.*]] = add i32 [[TMP11]], -5 ; MAX-COST-NEXT: [[TMP30:%.*]] = add i32 [[TMP28]], [[TMP29]] ; MAX-COST-NEXT: [[TMP31:%.*]] = select i1 [[TMP13]], i32 -720, i32 -80 ; MAX-COST-NEXT: [[TMP32:%.*]] = add i32 [[BIN_EXTRA]], [[TMP31]] diff --git a/llvm/test/Transforms/SLPVectorizer/X86/horizontal.ll b/llvm/test/Transforms/SLPVectorizer/X86/horizontal.ll index 080f850..dd08ef28 100644 --- a/llvm/test/Transforms/SLPVectorizer/X86/horizontal.ll +++ b/llvm/test/Transforms/SLPVectorizer/X86/horizontal.ll @@ -817,22 +817,22 @@ declare i32 @foobar(i32) define void @i32_red_call(i32 %val) { ; CHECK-LABEL: @i32_red_call( ; CHECK-NEXT: entry: -; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 0), align 16 -; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 1), align 4 -; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP0]] -; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 2), align 8 -; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 [[TMP2]], [[ADD]] -; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 3), align 4 -; CHECK-NEXT: [[ADD_2:%.*]] = add nsw i32 [[TMP3]], [[ADD_1]] -; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 4), align 16 -; CHECK-NEXT: [[ADD_3:%.*]] = add nsw i32 [[TMP4]], [[ADD_2]] -; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 5), align 4 -; CHECK-NEXT: [[ADD_4:%.*]] = add nsw i32 [[TMP5]], [[ADD_3]] -; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 6), align 8 -; CHECK-NEXT: [[ADD_5:%.*]] = add nsw i32 [[TMP6]], [[ADD_4]] -; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 7), align 4 -; CHECK-NEXT: [[ADD_6:%.*]] = add nsw i32 [[TMP7]], [[ADD_5]] -; CHECK-NEXT: [[RES:%.*]] = call i32 @foobar(i32 [[ADD_6]]) +; CHECK-NEXT: [[TMP0:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr_i32 to <8 x i32>*), align 16 +; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 undef, undef +; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 undef, [[ADD]] +; CHECK-NEXT: [[ADD_2:%.*]] = add nsw i32 undef, [[ADD_1]] +; CHECK-NEXT: [[ADD_3:%.*]] = add nsw i32 undef, [[ADD_2]] +; CHECK-NEXT: [[ADD_4:%.*]] = add nsw i32 undef, [[ADD_3]] +; CHECK-NEXT: [[ADD_5:%.*]] = add nsw i32 undef, [[ADD_4]] +; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP0]], <8 x i32> undef, <8 x i32> +; CHECK-NEXT: [[BIN_RDX:%.*]] = add nsw <8 x i32> [[TMP0]], [[RDX_SHUF]] +; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> +; CHECK-NEXT: [[BIN_RDX2:%.*]] = add nsw <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]] +; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> +; CHECK-NEXT: [[BIN_RDX4:%.*]] = add nsw <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0 +; CHECK-NEXT: [[ADD_6:%.*]] = add nsw i32 undef, [[ADD_5]] +; CHECK-NEXT: [[RES:%.*]] = call i32 @foobar(i32 [[TMP1]]) ; CHECK-NEXT: ret void ; entry: @@ -858,22 +858,22 @@ entry: define void @i32_red_invoke(i32 %val) personality i32 (...)* @__gxx_personality_v0 { ; CHECK-LABEL: @i32_red_invoke( ; CHECK-NEXT: entry: -; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 0), align 16 -; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 1), align 4 -; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP0]] -; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 2), align 8 -; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 [[TMP2]], [[ADD]] -; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 3), align 4 -; CHECK-NEXT: [[ADD_2:%.*]] = add nsw i32 [[TMP3]], [[ADD_1]] -; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 4), align 16 -; CHECK-NEXT: [[ADD_3:%.*]] = add nsw i32 [[TMP4]], [[ADD_2]] -; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 5), align 4 -; CHECK-NEXT: [[ADD_4:%.*]] = add nsw i32 [[TMP5]], [[ADD_3]] -; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 6), align 8 -; CHECK-NEXT: [[ADD_5:%.*]] = add nsw i32 [[TMP6]], [[ADD_4]] -; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr_i32, i64 0, i64 7), align 4 -; CHECK-NEXT: [[ADD_6:%.*]] = add nsw i32 [[TMP7]], [[ADD_5]] -; CHECK-NEXT: [[RES:%.*]] = invoke i32 @foobar(i32 [[ADD_6]]) +; CHECK-NEXT: [[TMP0:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr_i32 to <8 x i32>*), align 16 +; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 undef, undef +; CHECK-NEXT: [[ADD_1:%.*]] = add nsw i32 undef, [[ADD]] +; CHECK-NEXT: [[ADD_2:%.*]] = add nsw i32 undef, [[ADD_1]] +; CHECK-NEXT: [[ADD_3:%.*]] = add nsw i32 undef, [[ADD_2]] +; CHECK-NEXT: [[ADD_4:%.*]] = add nsw i32 undef, [[ADD_3]] +; CHECK-NEXT: [[ADD_5:%.*]] = add nsw i32 undef, [[ADD_4]] +; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP0]], <8 x i32> undef, <8 x i32> +; CHECK-NEXT: [[BIN_RDX:%.*]] = add nsw <8 x i32> [[TMP0]], [[RDX_SHUF]] +; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> +; CHECK-NEXT: [[BIN_RDX2:%.*]] = add nsw <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]] +; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> +; CHECK-NEXT: [[BIN_RDX4:%.*]] = add nsw <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0 +; CHECK-NEXT: [[ADD_6:%.*]] = add nsw i32 undef, [[ADD_5]] +; CHECK-NEXT: [[RES:%.*]] = invoke i32 @foobar(i32 [[TMP1]]) ; CHECK-NEXT: to label [[NORMAL:%.*]] unwind label [[EXCEPTION:%.*]] ; CHECK: exception: ; CHECK-NEXT: [[CLEANUP:%.*]] = landingpad i8 diff --git a/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll b/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll index 9e4f503..46386e8 100644 --- a/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll +++ b/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll @@ -303,24 +303,30 @@ define <4 x float> @simple_select_no_users(<4 x float> %a, <4 x float> %b, <4 x ; CHECK-NEXT: [[B1:%.*]] = extractelement <4 x float> %b, i32 1 ; CHECK-NEXT: [[B2:%.*]] = extractelement <4 x float> %b, i32 2 ; CHECK-NEXT: [[B3:%.*]] = extractelement <4 x float> %b, i32 3 -; CHECK-NEXT: [[CMP0:%.*]] = icmp ne i32 [[C0]], 0 -; CHECK-NEXT: [[CMP1:%.*]] = icmp ne i32 [[C1]], 0 -; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> undef, i32 [[C2]], i32 0 -; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C3]], i32 1 +; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> undef, i32 [[C0]], i32 0 +; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1 ; CHECK-NEXT: [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer -; CHECK-NEXT: [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]] -; CHECK-NEXT: [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]] -; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x float> undef, float [[A2]], i32 0 -; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x float> [[TMP4]], float [[A3]], i32 1 -; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x float> undef, float [[B2]], i32 0 -; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x float> [[TMP6]], float [[B3]], i32 1 -; CHECK-NEXT: [[TMP8:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP5]], <2 x float> [[TMP7]] -; CHECK-NEXT: [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0 -; CHECK-NEXT: [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1 -; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x float> [[TMP8]], i32 0 -; CHECK-NEXT: [[RC:%.*]] = insertelement <4 x float> undef, float [[TMP9]], i32 2 -; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x float> [[TMP8]], i32 1 -; CHECK-NEXT: [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[TMP10]], i32 3 +; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i32> undef, i32 [[C2]], i32 0 +; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i32> [[TMP4]], i32 [[C3]], i32 1 +; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <2 x i32> [[TMP5]], zeroinitializer +; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x float> undef, float [[A0]], i32 0 +; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[A1]], i32 1 +; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x float> undef, float [[B0]], i32 0 +; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[B1]], i32 1 +; CHECK-NEXT: [[TMP11:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP8]], <2 x float> [[TMP10]] +; CHECK-NEXT: [[TMP12:%.*]] = insertelement <2 x float> undef, float [[A2]], i32 0 +; CHECK-NEXT: [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[A3]], i32 1 +; CHECK-NEXT: [[TMP14:%.*]] = insertelement <2 x float> undef, float [[B2]], i32 0 +; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[B3]], i32 1 +; CHECK-NEXT: [[TMP16:%.*]] = select <2 x i1> [[TMP6]], <2 x float> [[TMP13]], <2 x float> [[TMP15]] +; CHECK-NEXT: [[TMP17:%.*]] = extractelement <2 x float> [[TMP11]], i32 0 +; CHECK-NEXT: [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP17]], i32 0 +; CHECK-NEXT: [[TMP18:%.*]] = extractelement <2 x float> [[TMP11]], i32 1 +; CHECK-NEXT: [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP18]], i32 1 +; CHECK-NEXT: [[TMP19:%.*]] = extractelement <2 x float> [[TMP16]], i32 0 +; CHECK-NEXT: [[RC:%.*]] = insertelement <4 x float> undef, float [[TMP19]], i32 2 +; CHECK-NEXT: [[TMP20:%.*]] = extractelement <2 x float> [[TMP16]], i32 1 +; CHECK-NEXT: [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[TMP20]], i32 3 ; CHECK-NEXT: ret <4 x float> [[RD]] ; ; ZEROTHRESH-LABEL: @simple_select_no_users( -- 2.7.4