[LoopVectorizer] Use two step casting for float to pointer types.

Summary:
LoopVectorizer is creating casts between vec<ptr> and vec<float> types
on ARM when compiling OpenCV. Since, tIs is illegal to directly cast a
floating point type to a pointer type even if the types have same size
causing a crash. Fix the crash using a two-step casting by bitcasting
to integer and integer to pointer/float.
Fixes PR33804.

Reviewers: mkuper, Ayal, dlj, rengolin, srhines

Reviewed By: rengolin

Subscribers: aemerson, kristof.beyls, mkazantsev, Meinersbur, rengolin, mzolotukhin, llvm-commits

Differential Revision: https://reviews.llvm.org/D35498

llvm-svn: 312331

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -585,6 +585,11 @@ class InnerLoopVectorizer {
   /// 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 vector<float> <-> vector<pointer> types.
+  Value *createBitOrPointerCast(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);
@@ -2866,6 +2871,7 @@ void InnerLoopVectorizer::vectorizeInterleaveGroup(Instruction *Instr) {
   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.
@@ -2940,7 +2946,7 @@ void InnerLoopVectorizer::vectorizeInterleaveGroup(Instruction *Instr) {
         // 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 = createBitOrPointerCast(StridedVec, OtherVTy, DL);
         }
 
         if (Group->isReverse())
@@ -2969,9 +2975,10 @@ void InnerLoopVectorizer::vectorizeInterleaveGroup(Instruction *Instr) {
       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 = createBitOrPointerCast(StoredVec, SubVT, DL);
 
       StoredVecs.push_back(StoredVec);
     }
@@ -3272,6 +3279,36 @@ Value *InnerLoopVectorizer::getOrCreateVectorTripCount(Loop *L) {
   return VectorTripCount;
 }
 
+Value *InnerLoopVectorizer::createBitOrPointerCast(Value *V, VectorType *DstVTy,
+                                                   const DataLayout &DL) {
+  // Verify that V is a vector type with same number of elements as DstVTy.
+  unsigned VF = DstVTy->getNumElements();
+  VectorType *SrcVecTy = cast<VectorType>(V->getType());
+  assert((VF == SrcVecTy->getNumElements()) && "Vector dimensions do not match");
+  Type *SrcElemTy = SrcVecTy->getElementType();
+  Type *DstElemTy = DstVTy->getElementType();
+  assert((DL.getTypeSizeInBits(SrcElemTy) == DL.getTypeSizeInBits(DstElemTy)) &&
+         "Vector elements must have same size");
+
+  // Do a direct cast if element types are castable.
+  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()) &&
+         "Only one type should be a pointer type");
+  assert((DstElemTy->isFloatingPointTy() != SrcElemTy->isFloatingPointTy()) &&
+         "Only one type should be a floating point type");
+  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);

llvm/test/CodeGen/AArch64/loopvectorize_pr33804_double.ll

@@ -0,0 +1,66 @@
+; RUN: opt -loop-vectorize -S < %s | FileCheck %s
+
+; These tests check that we don't crash if vectorizer decides to cast
+; a double value to be stored into a pointer type or vice-versa.
+
+; This test checks when a double value is stored into a pointer type.
+
+; ModuleID = 'bugpoint-reduced-simplified.bc'
+source_filename = "bugpoint-output-26dbd81.bc"
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-unknown-linux-gnu"
+
+%struct.CvNode1D = type { double, %struct.CvNode1D* }
+
+; CHECK-LABEL: @cvCalcEMD2
+; CHECK: vector.body
+; CHECK: store <{{[0-9]+}} x %struct.CvNode1D*>
+define void @cvCalcEMD2() local_unnamed_addr #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
+entry:
+  br label %for.body14.i.i
+
+for.body14.i.i:                                   ; preds = %for.body14.i.i, %entry
+  %i.1424.i.i = phi i32 [ %inc21.i.i, %for.body14.i.i ], [ 0, %entry ]
+  %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 double 0xC415AF1D80000000, double* %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
+}
+
+; This test checks when a pointer value is stored into a double type.
+
+%struct.CvNode1D2 = type { %struct.CvNode1D2*, double }
+
+; CHECK-LABEL: @cvCalcEMD2_2
+; CHECK: vector.body
+; CHECK: store <{{[0-9]+}} x double>
+define void @cvCalcEMD2_2() local_unnamed_addr #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
+entry:
+  br label %for.body14.i.i
+
+for.body14.i.i:                                   ; preds = %for.body14.i.i, %entry
+  %i.1424.i.i = phi i32 [ %inc21.i.i, %for.body14.i.i ], [ 0, %entry ]
+  %next19.i.i = getelementptr inbounds %struct.CvNode1D2, %struct.CvNode1D2* undef, i32 %i.1424.i.i, i32 0
+  store %struct.CvNode1D2* undef, %struct.CvNode1D2** %next19.i.i, align 4
+  %arrayidx15.i.i1427 = getelementptr inbounds %struct.CvNode1D2, %struct.CvNode1D2* undef, i32 %i.1424.i.i
+  %val.i.i = getelementptr inbounds %struct.CvNode1D2, %struct.CvNode1D2* %arrayidx15.i.i1427, i32 0, i32 1
+  store double 0xC415AF1D80000000, double* %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 i32 @__gxx_personality_v0(...)
+
+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" "unsafe-fp-math"="false" "use-soft-float"="false" }
+

llvm/test/CodeGen/ARM/loopvectorize_pr33804.ll

@@ -0,0 +1,66 @@
+; RUN: opt -loop-vectorize -S < %s | FileCheck %s
+
+; These tests check that we don't crash if vectorizer decides to cast
+; a float value to be stored into a pointer type or vice-versa.
+
+; This test checks when a float value is stored into 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-unknown-linux-gnueabihf"
+
+%struct.CvNode1D = type { float, %struct.CvNode1D* }
+
+; CHECK-LABEL: @cvCalcEMD2
+; CHECK: vector.body
+; CHECK: store <{{[0-9]+}} x %struct.CvNode1D*>
+define void @cvCalcEMD2() local_unnamed_addr #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
+entry:
+  br label %for.body14.i.i
+
+for.body14.i.i:                                   ; preds = %for.body14.i.i, %entry
+  %i.1424.i.i = phi i32 [ %inc21.i.i, %for.body14.i.i ], [ 0, %entry ]
+  %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
+}
+
+; This test checks when a pointer value is stored into a float type.
+
+%struct.CvNode1D2 = type { %struct.CvNode1D2*, float }
+
+; CHECK-LABEL: @cvCalcEMD2_2
+; CHECK: vector.body
+; CHECK: store <{{[0-9]+}} x float>
+define void @cvCalcEMD2_2() local_unnamed_addr #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
+entry:
+  br label %for.body14.i.i
+
+for.body14.i.i:                                   ; preds = %for.body14.i.i, %entry
+  %i.1424.i.i = phi i32 [ %inc21.i.i, %for.body14.i.i ], [ 0, %entry ]
+  %next19.i.i = getelementptr inbounds %struct.CvNode1D2, %struct.CvNode1D2* undef, i32 %i.1424.i.i, i32 0
+  store %struct.CvNode1D2* undef, %struct.CvNode1D2** %next19.i.i, align 4
+  %arrayidx15.i.i1427 = getelementptr inbounds %struct.CvNode1D2, %struct.CvNode1D2* undef, i32 %i.1424.i.i
+  %val.i.i = getelementptr inbounds %struct.CvNode1D2, %struct.CvNode1D2* %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 i32 @__gxx_personality_v0(...)
+
+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" }
+