[PATCH] D108411: [X86][TLI] SimplifyDemandedVectorEltsForTargetNode(): don't break apart broadcasts from which not just the 0'th elt is demanded

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

Note that this does not appear to have test coverage before it's parent patch.

In D108411#2959009 <https://reviews.llvm.org/D108411#2959009>, @RKSimon wrote:

> Something that I've been wondering about is whether SelectionDAG::isSplatValue should be moved to TLI to make it possible to peek through target opcodes as well to help identify more splats/broadcasts

That does sound plausible.

> - have you noticed anything that might benefit?

I don't really know, but the change doesn't seem too intrusive so i suppose i could just try doing that and seeing the fallout.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D108411/new/

https://reviews.llvm.org/D108411