Index: lib/Transforms/Vectorize/LoopVectorize.cpp =================================================================== --- lib/Transforms/Vectorize/LoopVectorize.cpp +++ lib/Transforms/Vectorize/LoopVectorize.cpp @@ -574,6 +574,10 @@ /// Returns (and creates if needed) the trip count of the widened loop. Value *getOrCreateVectorTripCount(Loop *NewLoop); + /// Returns a bitcasted value to the requested vector type. + /// Also handles bitcasts of float <--> pointer types. + Value* createBitCast(Value *V, VectorType* DstVTy, const DataLayout& DL); + /// Emit a bypass check to see if the vector trip count is zero, including if /// it overflows. void emitMinimumIterationCountCheck(Loop *L, BasicBlock *Bypass); @@ -2852,6 +2856,7 @@ if (Instr != Group->getInsertPos()) return; + const DataLayout &DL = Instr->getModule()->getDataLayout(); Value *Ptr = getPointerOperand(Instr); // Prepare for the vector type of the interleaved load/store. @@ -2926,7 +2931,7 @@ // If this member has different type, cast the result type. if (Member->getType() != ScalarTy) { VectorType *OtherVTy = VectorType::get(Member->getType(), VF); - StridedVec = Builder.CreateBitOrPointerCast(StridedVec, OtherVTy); + StridedVec = createBitCast(StridedVec, OtherVTy, DL); } if (Group->isReverse()) @@ -2955,9 +2960,10 @@ if (Group->isReverse()) StoredVec = reverseVector(StoredVec); - // If this member has different type, cast it to an unified type. + // If this member has different type, cast it to a unified type. + if (StoredVec->getType() != SubVT) - StoredVec = Builder.CreateBitOrPointerCast(StoredVec, SubVT); + StoredVec = createBitCast(StoredVec, SubVT, DL); StoredVecs.push_back(StoredVec); } @@ -3292,6 +3298,40 @@ return VectorTripCount; } +Value* InnerLoopVectorizer::createBitCast(Value *V, VectorType* DstVTy, + const DataLayout& DL) { + // Do a direct cast if a safe direct cast is possible. + if (CastInst::isBitOrNoopPointerCastable(V->getType(), DstVTy, DL)) { + return Builder.CreateBitOrPointerCast(V, DstVTy); + } + // Verify that V is a vector type with same number of elements as DstVTy. + unsigned VF = DstVTy->getNumElements(); + VectorType *SrcVecTy = dyn_cast(V->getType()); + assert(SrcVecTy); + assert(VF == SrcVecTy->getNumElements()); + Type *SrcElemTy = SrcVecTy->getElementType(); + Type *DstElemTy = DstVTy->getElementType(); + assert(DL.getTypeSizeInBits(SrcElemTy) == DL.getTypeSizeInBits(DstElemTy)); + + // The previous castable check does not cover the bitcasts between + // vector and vector types and may fail. So try another time + // but using element types. + if (CastInst::isBitOrNoopPointerCastable(SrcElemTy, DstElemTy, DL)) { + return Builder.CreateBitOrPointerCast(V, DstVTy); + } + // V cannot be directly casted to desired vector type. + // May happen when V is a floating point vector but DstVTy is a vector of pointers + // or vice-versa. Handle this using a two-step bitcast using an intermediate Integer + // type for the bitcast i.e. Ptr <-> Int <-> Float. + assert(DstElemTy->isPointerTy() != SrcElemTy->isPointerTy()); + assert(DstElemTy->isFloatingPointTy() != SrcElemTy->isFloatingPointTy()); + Type *IntTy = IntegerType::getIntNTy(V->getContext(), + DL.getTypeSizeInBits(SrcElemTy)); + VectorType *VecIntTy = VectorType::get(IntTy, VF); + Value *CastVal = Builder.CreateBitOrPointerCast(V, VecIntTy); + return Builder.CreateBitOrPointerCast(CastVal, DstVTy); +} + void InnerLoopVectorizer::emitMinimumIterationCountCheck(Loop *L, BasicBlock *Bypass) { Value *Count = getOrCreateTripCount(L); Index: test/CodeGen/ARM/loopvectorize_pr33804_1.ll =================================================================== --- /dev/null +++ test/CodeGen/ARM/loopvectorize_pr33804_1.ll @@ -0,0 +1,90 @@ +; RUN: opt -loop-vectorize -S < %s | FileCheck %s + +; This checks we don't crash when vectorizing if vectorizer ends up +; requiring casting float to a pointer type. + +; ModuleID = 'bugpoint-reduced-simplified.bc' +source_filename = "bugpoint-output-26dbd81.bc" +target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64" +target triple = "armv7--linux-gnueabihf" + +%struct.CvNode1D = type { float, %struct.CvNode1D* } + +@.str.13 = external unnamed_addr constant [1 x i8], align 1 + +; CHECK-LABEL: @cvCalcEMD2 +; CHECK: vector.body +; CHECK: store <8 x %struct.CvNode1D*> +define void @cvCalcEMD2() local_unnamed_addr #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) { +entry: + invoke void @cvGetMat() + to label %invoke.cont unwind label %lpad.loopexit.split-lp + +invoke.cont: ; preds = %entry + invoke void @cvGetMat() + to label %invoke.cont3 unwind label %lpad.loopexit.split-lp + +invoke.cont3: ; preds = %invoke.cont + invoke void @_Znaj() #3 + to label %call.i.i.i1408.noexc unwind label %lpad.loopexit.split-lp + +lpad.loopexit.split-lp: ; preds = %invoke.cont3, %entry, %invoke.cont + %lpad.loopexit.split-lp2387 = landingpad { i8*, i32 } + cleanup + resume { i8*, i32 } undef + +call.i.i.i1408.noexc: ; preds = %invoke.cont3 + invoke void @_ZNSsC1EPKcRKSaIcE() + to label %invoke.cont188.i unwind label %lpad187.i + +invoke.cont188.i: ; preds = %call.i.i.i1408.noexc + br label %invoke.cont203.i + +invoke.cont203.i: ; preds = %invoke.cont188.i + invoke void @_ZN2cv5errorERKNS_9ExceptionE() + to label %invoke.cont206.i unwind label %lpad205.i + +invoke.cont206.i: ; preds = %invoke.cont203.i + br label %for.body14.i.i + +lpad187.i: ; preds = %call.i.i.i1408.noexc + %0 = landingpad { i8*, i32 } + cleanup + unreachable + +lpad205.i: ; preds = %invoke.cont203.i + %1 = landingpad { i8*, i32 } + cleanup + unreachable + +for.body14.i.i: ; preds = %for.body14.i.i, %invoke.cont206.i + %i.1424.i.i = phi i32 [ %inc21.i.i, %for.body14.i.i ], [ 0, %invoke.cont206.i ] + %arrayidx15.i.i1427 = getelementptr inbounds %struct.CvNode1D, %struct.CvNode1D* undef, i32 %i.1424.i.i + %val.i.i = getelementptr inbounds %struct.CvNode1D, %struct.CvNode1D* %arrayidx15.i.i1427, i32 0, i32 0 + store float 0xC415AF1D80000000, float* %val.i.i, align 4 + %next19.i.i = getelementptr inbounds %struct.CvNode1D, %struct.CvNode1D* undef, i32 %i.1424.i.i, i32 1 + store %struct.CvNode1D* undef, %struct.CvNode1D** %next19.i.i, align 4 + %inc21.i.i = add nuw nsw i32 %i.1424.i.i, 1 + %exitcond438.i.i = icmp eq i32 %inc21.i.i, 0 + br i1 %exitcond438.i.i, label %for.end22.i.i, label %for.body14.i.i + +for.end22.i.i: ; preds = %for.body14.i.i + unreachable +} + +declare void @cvGetMat() local_unnamed_addr #1 + +declare i32 @__gxx_personality_v0(...) + +declare void @_ZN2cv5errorERKNS_9ExceptionE() local_unnamed_addr #1 + +declare void @_ZNSsC1EPKcRKSaIcE() unnamed_addr #1 + +; Function Attrs: nobuiltin +declare void @_Znaj() local_unnamed_addr #2 + +attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+dsp,+neon,+vfp3,-thumb-mode" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #1 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+dsp,+neon,+vfp3,-thumb-mode" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #2 = { nobuiltin "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+dsp,+neon,+vfp3,-thumb-mode" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #3 = { builtin } + Index: test/CodeGen/ARM/loopvectorize_pr33804_2.ll =================================================================== --- /dev/null +++ test/CodeGen/ARM/loopvectorize_pr33804_2.ll @@ -0,0 +1,90 @@ +; RUN: opt -loop-vectorize -S < %s | FileCheck %s + +; This checks we don't crash when vectorizing if vectorizer ends up +; requiring casting pointer to a float type. + +; ModuleID = 'bugpoint-reduced-simplified.bc' +source_filename = "bugpoint-output-26dbd81.bc" +target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64" +target triple = "armv7--linux-gnueabihf" + +%struct.CvNode1D = type { %struct.CvNode1D*, float } + +@.str.13 = external unnamed_addr constant [1 x i8], align 1 + +; CHECK-LABEL: @cvCalcEMD2 +; CHECK: vector.body +; CHECK: store <8 x float> +define void @cvCalcEMD2() local_unnamed_addr #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) { +entry: + invoke void @cvGetMat() + to label %invoke.cont unwind label %lpad.loopexit.split-lp + +invoke.cont: ; preds = %entry + invoke void @cvGetMat() + to label %invoke.cont3 unwind label %lpad.loopexit.split-lp + +invoke.cont3: ; preds = %invoke.cont + invoke void @_Znaj() #3 + to label %call.i.i.i1408.noexc unwind label %lpad.loopexit.split-lp + +lpad.loopexit.split-lp: ; preds = %invoke.cont3, %entry, %invoke.cont + %lpad.loopexit.split-lp2387 = landingpad { i8*, i32 } + cleanup + resume { i8*, i32 } undef + +call.i.i.i1408.noexc: ; preds = %invoke.cont3 + invoke void @_ZNSsC1EPKcRKSaIcE() + to label %invoke.cont188.i unwind label %lpad187.i + +invoke.cont188.i: ; preds = %call.i.i.i1408.noexc + br label %invoke.cont203.i + +invoke.cont203.i: ; preds = %invoke.cont188.i + invoke void @_ZN2cv5errorERKNS_9ExceptionE() + to label %invoke.cont206.i unwind label %lpad205.i + +invoke.cont206.i: ; preds = %invoke.cont203.i + br label %for.body14.i.i + +lpad187.i: ; preds = %call.i.i.i1408.noexc + %0 = landingpad { i8*, i32 } + cleanup + unreachable + +lpad205.i: ; preds = %invoke.cont203.i + %1 = landingpad { i8*, i32 } + cleanup + unreachable + +for.body14.i.i: ; preds = %for.body14.i.i, %invoke.cont206.i + %i.1424.i.i = phi i32 [ %inc21.i.i, %for.body14.i.i ], [ 0, %invoke.cont206.i ] + %next19.i.i = getelementptr inbounds %struct.CvNode1D, %struct.CvNode1D* undef, i32 %i.1424.i.i, i32 0 + store %struct.CvNode1D* undef, %struct.CvNode1D** %next19.i.i, align 4 + %arrayidx15.i.i1427 = getelementptr inbounds %struct.CvNode1D, %struct.CvNode1D* undef, i32 %i.1424.i.i + %val.i.i = getelementptr inbounds %struct.CvNode1D, %struct.CvNode1D* %arrayidx15.i.i1427, i32 0, i32 1 + store float 0xC415AF1D80000000, float* %val.i.i, align 4 + %inc21.i.i = add nuw nsw i32 %i.1424.i.i, 1 + %exitcond438.i.i = icmp eq i32 %inc21.i.i, 0 + br i1 %exitcond438.i.i, label %for.end22.i.i, label %for.body14.i.i + +for.end22.i.i: ; preds = %for.body14.i.i + unreachable +} + +declare void @cvGetMat() local_unnamed_addr #1 + +declare i32 @__gxx_personality_v0(...) + +declare void @_ZN2cv5errorERKNS_9ExceptionE() local_unnamed_addr #1 + +declare void @_ZNSsC1EPKcRKSaIcE() unnamed_addr #1 + +; Function Attrs: nobuiltin +declare void @_Znaj() local_unnamed_addr #2 + +attributes #0 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+dsp,+neon,+vfp3,-thumb-mode" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #1 = { "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+dsp,+neon,+vfp3,-thumb-mode" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #2 = { nobuiltin "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+dsp,+neon,+vfp3,-thumb-mode" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #3 = { builtin } +