[VectorCombine][SVE] Do not fold bitcast shuffle for scalable type.

[huihuiz]

Summary

First, shuffle cost for scalable type is not known for scalable type;
Second, we cannot reason if the narrowed shuffle mask for scalable type
is a splat or not.

E.g., Bitcast splat vector from type <vscale x 4 x i32> to <vscale x 8 x i16>
will involve narrowing shuffle mask <vscale x 4 x i32> zeroinitializer to
<vscale x 8 x i32> with element sequence of <0, 1, 0, 1, ...>, which cannot be
reasoned if it's a valid splat or not.

[spatel]

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

450

Change to dyn_cast<FixedVectorType> above this instead of adding isa<> checks?

452–453

Move this comment to up to the point that we are excluding ScalableVectorType.

[spatel]

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

452–453

(or remove if it is not possible)

[huihuiz]

Thanks Sanjay for the review!
Addressed review comments.

[spatel]

LGTM

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

 /// moving bitcasts or shuffles together.
 bool VectorCombine::foldBitcastShuf(Instruction &I) {
   Value *V;
   ArrayRef<int> Mask;
   if (!match(&I, m_BitCast(
                      m_OneUse(m_Shuffle(m_Value(V), m_Undef(), m_Mask(Mask))))))
     return false;
 
-  // Disallow non-vector casts and length-changing shuffles.
+  // 1) Do not fold bitcast shuffle for scalable type. First, shuffle cost for
+  // scalable type is unknown; Second, we cannot reason if the narrowed shuffle
+  // mask for scalable type is a splat or not.
+  // 2) Disallow non-vector casts and length-changing shuffles.
   // TODO: We could allow any shuffle.
-  auto *DestTy = dyn_cast<VectorType>(I.getType());
-  auto *SrcTy = cast<VectorType>(V->getType());
-  if (!DestTy || I.getOperand(0)->getType() != SrcTy)
+  auto *DestTy = dyn_cast<FixedVectorType>(I.getType());
+  auto *SrcTy = dyn_cast<FixedVectorType>(V->getType());
+  if (!SrcTy || !DestTy || I.getOperand(0)->getType() != SrcTy)
     return false;
 
   // The new shuffle must not cost more than the old shuffle. The bitcast is
   // moved ahead of the shuffle, so assume that it has the same cost as before.
   if (TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, DestTy) >
       TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, SrcTy))
     return false;
 
-  // FIXME: it should be possible to implement the computation of the widened
-  // shuffle mask in terms of ElementCount to work with scalable shuffles.
-  unsigned DestNumElts = cast<FixedVectorType>(DestTy)->getNumElements();
-  unsigned SrcNumElts = cast<FixedVectorType>(SrcTy)->getNumElements();
+  unsigned DestNumElts = DestTy->getNumElements();
+  unsigned SrcNumElts = SrcTy->getNumElements();
   SmallVector<int, 16> NewMask;
   if (SrcNumElts <= DestNumElts) {
     // The bitcast is from wide to narrow/equal elements. The shuffle mask can
     // always be expanded to the equivalent form choosing narrower elements.
     assert(DestNumElts % SrcNumElts == 0 && "Unexpected shuffle mask");
     unsigned ScaleFactor = DestNumElts / SrcNumElts;
     narrowShuffleMaskElts(ScaleFactor, Mask, NewMask);
   } else {

llvm/test/Transforms/VectorCombine/AArch64/lit.local.cfg

+if not 'AArch64' in config.root.targets:
+    config.unsupported = True

llvm/test/Transforms/VectorCombine/AArch64/vscale-bitcast-shuffle.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -vector-combine -S -mtriple=aarch64-- | FileCheck %s --check-prefixes=CHECK
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-unknown-linux-gnu"
+
+; This test checks we are not crashing with TTI when trying to get shuffle cost.
+; This test also check that shuffle mask <vscale x 4 x i32> zeroinitializer is
+; not narrowed into <0, 1, 0, 1, ...>, which we cannot reason if it's a valid
+; splat or not.
+
+define <vscale x 8 x i16> @bitcast_shuffle(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: @bitcast_shuffle(
+; CHECK-NEXT:    [[I:%.*]] = shufflevector <vscale x 4 x i32> [[A:%.*]], <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT:    [[R:%.*]] = bitcast <vscale x 4 x i32> [[I]] to <vscale x 8 x i16>
+; CHECK-NEXT:    ret <vscale x 8 x i16> [[R]]
+;
+  %i = shufflevector <vscale x 4 x i32> %a, <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer
+  %r = bitcast <vscale x 4 x i32> %i to <vscale x 8 x i16>
+  ret <vscale x 8 x i16> %r
+}