[PATCH] D107272: AMDGPU/GlobalISel: Stop using NarrowScalar/FewerElements for unaligned splitting

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp:1248
               unsigned DstSize = DstTy.getSizeInBits();
               if (!isPowerOf2_32(DstSize)) {
                 // We're probably decomposing an odd sized store. Try to split
----------------
foad wrote:
> Is this case still required for some reason, even though you've removed the corresponding code from the scalar case above?
This one you can technically remove, but it results in worse code since it hits the full scalarization path below. 


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D107272/new/

https://reviews.llvm.org/D107272