[SVE] Remove getNumElements() warnings in InstCombiner::visitBitCast

[david-arm]

Summary

Whilst trying to compile this test to assembly:

CodeGen/aarch64-sve-intrinsics/acle_sve_reinterpret.c

I discovered some warnings were firing in InstCombiner::visitBitCast
due to calls to getNumElements() for scalable vector types. These
calls only really made sense for fixed width vectors so I have fixed
up the code appropriately.

[efriedma]

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

2529

Probably should just be if (auto *DestVTy = dyn_cast<FixedVectorType>(DestTy)) {. A bitcast with a scalable vector dest type must have a scalable vector src type, anyway.

2559

We should have a test that we don't do these folds for scalable vectors.

[david-arm]

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

2529

OK, perhaps you're right. I thought that some of the code below (optimizeIntegerToVectorInsertions) may still kick in for scalable vectors. I'll have a deeper look.

2559

Sure. I hit this while trying to compile a clang ACLE test, so I can reproduce some of that code in a smaller test.

[efriedma]

LGTM

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

if (SrcElTy == DstElTy) {
         // zeros.
         if (SrcPTy->getAddressSpace() == 0 || !CanBeNull)
           GEP->setIsInBounds();
       }
       return GEP;
     }
   }
 
-  if (VectorType *DestVTy = dyn_cast<VectorType>(DestTy)) {
+  if (FixedVectorType *DestVTy = dyn_cast<FixedVectorType>(DestTy)) {
     // Beware: messing with this target-specific oddity may cause trouble.
     if (DestVTy->getNumElements() == 1 && SrcTy->isX86_MMXTy()) {
       Value *Elem = Builder.CreateBitCast(Src, DestVTy->getElementType());
       return InsertElementInst::Create(UndefValue::get(DestTy), Elem,
                      Constant::getNullValue(Type::getInt32Ty(CI.getContext())));
     }
 
     if (isa<IntegerType>(SrcTy)) {
if (isa<IntegerType>(SrcTy)) {
       // If the input is an 'or' instruction, we may be doing shifts and ors to
       // assemble the elements of the vector manually.  Try to rip the code out
       // and replace it with insertelements.
       if (Value *V = optimizeIntegerToVectorInsertions(CI, *this))
         return replaceInstUsesWith(CI, V);
     }
   }
 
-  if (VectorType *SrcVTy = dyn_cast<VectorType>(SrcTy)) {
+  if (FixedVectorType *SrcVTy = dyn_cast<FixedVectorType>(SrcTy)) {
     if (SrcVTy->getNumElements() == 1) {
       // If our destination is not a vector, then make this a straight
       // scalar-scalar cast.
       if (!DestTy->isVectorTy()) {
         Value *Elem =
           Builder.CreateExtractElement(Src,
                      Constant::getNullValue(Type::getInt32Ty(CI.getContext())));
         return CastInst::Create(Instruction::BitCast, Elem, DestTy);
       }

llvm/test/Transforms/InstCombine/AArch64/sve-bitcast.ll

+; RUN: opt -instcombine -mtriple=aarch64-linux-gnu -mattr=+sve -S < %s | FileCheck %s
+
+; We shouldn't fold bitcast(insert <vscale x 1 x iX> .., iX %val, i32 0)
+; into bitcast(iX %val) for scalable vectors.
+define <vscale x 2 x i8> @bitcast_of_insert_i8_i16(i16 %val) #0 {
+; CHECK-LABEL: @bitcast_of_insert_i8_i16(
+; CHECK-NOT:   bitcast i16 %val to <vscale x 2 x i8>
+; CHECK:       bitcast <vscale x 1 x i16> %op2 to <vscale x 2 x i8>
+entry:
+  %op2 = insertelement <vscale x 1 x i16> undef, i16 %val, i32 0
+  %0 = bitcast <vscale x 1 x i16> %op2 to <vscale x 2 x i8>
+  ret <vscale x 2 x i8> %0
+}