[PATCH] D136858: [AArch64-SVE]: Force generating code compatible to streaming mode for sve-fixed-length tests.

[Sander de Smalen via Phabricator via llvm-commits]

sdesmalen added inline comments.


================
Comment at: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp:12931
       useSVEForFixedLengthVectorVT(
           SrcVT, OverrideNEON && Subtarget->useSVEForFixedLengthVectors())) {
 
----------------
Instead of updating the OverrideNEON variable, I suspect that you actually want to do something like this:

  if (SrcVT.isScalableVector() ||
      useSVEForFixedLengthVectorVT(
          SrcVT, OverrideNEON && Subtarget->useSVEForFixedLengthVectors()) ||
      useSVEForFixedLengthVectorVT(
          SrcVT, Subtarget->forceStreamingCompatibleSVE()))

so to not alter the behaviour for non-streaming fixed-length vectors. This avoids the regressions in sve-fixed-length-int-reduce.ll where the SVE variants require an additional predicate (whereas the NEON reduction operations are unpredicated and thus only 1 instruction).


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D136858/new/

https://reviews.llvm.org/D136858