[PATCH] D138545: [VectorCombine] Enable scalarizeBinopOrCmp for scalable vectors

[Sanjay Patel via Phabricator via llvm-commits]

spatel accepted this revision.
spatel added a comment.
This revision is now accepted and ready to land.

LGTM - see inline for a couple of additions.



================
Comment at: llvm/lib/Transforms/Vectorize/VectorCombine.cpp:1724
+
+    if (isa<VectorType>(I.getType()))
+      MadeChange |= scalarizeBinopOrCmp(I);
----------------
Add a comment here to make the logic difference explicit:
  // This transform works with scalable and fixed vectors.

Should there be a TODO comment about allowing more scalable transforms?


================
Comment at: llvm/test/Transforms/VectorCombine/AArch64/scalarize-scalable.ll:4
+
+define <vscale x 4 x float> @scalarize_scalable(float %0, float %1, float %2, float %3) {
+; CHECK-LABEL: @scalarize_scalable(
----------------
This shows a series of folds, but it would be good to have a couple of minimum tests too. Something like this should give us coverage for integer types and compares?


```
define <vscale x 4 x i32> @scalarize_scalable_udiv(i32 %x, i32 %y) {
  %splatx = insertelement <vscale x 4 x i32> poison, i32 %x, i64 0
  %splaty = insertelement <vscale x 4 x i32> poison, i32 %y, i64 0
  %r = udiv <vscale x 4 x i32> %splatx, %splaty
  ret <vscale x 4 x i32> %r
}

define <vscale x 4 x i1> @scalarize_scalable_icmp(i32 %x, i32 %y) {
  %splatx = insertelement <vscale x 4 x i32> poison, i32 %x, i64 0
  %splaty = insertelement <vscale x 4 x i32> poison, i32 %y, i64 0
  %r = icmp sgt <vscale x 4 x i32> %splatx, %splaty
  ret <vscale x 4 x i1> %r
}

```


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D138545/new/

https://reviews.llvm.org/D138545