[AArch64] use isTRNMask to calculate shuffle costs (#171524)

This builds on #169858 to fix the divergence in codegen
(https://godbolt.org/z/a9az3h6oq) between two very similar
functions initially observed in #137447 (represented in the diff by test
cases `@transpose_splat_constants` and `@transpose_constants_splat`:
```
int8x16_t f(int8_t x)
{
  return (int8x16_t) { x, 0, x, 1, x, 2, x, 3,
                       x, 4, x, 5, x, 6, x, 7 };
}

int8x16_t g(int8_t x)
{
  return (int8x16_t) { 0, x, 1, x, 2, x, 3, x,
                       4, x, 5, x, 6, x, 7, x };
}
```

The PR uses an additional `isTRNMask` call in
`AArch64TTIImpl::getShuffleCost` to ensure that we treat shuffle masks
as transpose masks even if `isTransposeMask` fails to recognise them
(meaning that `Kind == TTI::SK_Transpose` cannot be relied upon).

Follow-up work could consider modifying `isTransposeMask`, but that
would also impact other backends than AArch64.

Author: ginsbach

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

@@ -6134,8 +6134,13 @@ AArch64TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, VectorType *DstTy,
   unsigned Unused;
   if (LT.second.isFixedLengthVector() &&
       LT.second.getVectorNumElements() == Mask.size() &&
-      (Kind == TTI::SK_PermuteTwoSrc || Kind == TTI::SK_PermuteSingleSrc) &&
+      (Kind == TTI::SK_PermuteTwoSrc || Kind == TTI::SK_PermuteSingleSrc ||
+       // Discrepancies between isTRNMask and ShuffleVectorInst::isTransposeMask
+       // mean that we can end up with shuffles that satisfy isTRNMask, but end
+       // up labelled as TTI::SK_InsertSubvector. (e.g. {2, 0}).
+       Kind == TTI::SK_InsertSubvector) &&
       (isZIPMask(Mask, LT.second.getVectorNumElements(), Unused, Unused) ||
+       isTRNMask(Mask, LT.second.getVectorNumElements(), Unused, Unused) ||
        isUZPMask(Mask, LT.second.getVectorNumElements(), Unused) ||
        isREVMask(Mask, LT.second.getScalarSizeInBits(),
                  LT.second.getVectorNumElements(), 16) ||