[PATCH] D50992: [InstCombine] try to fold insertelt + vector op into scalar op + insertelt

[Sanjay Patel via Phabricator via llvm-commits]

spatel abandoned this revision.
spatel added a comment.

I'm finally going to abandon this. D80885 <https://reviews.llvm.org/D80885> would introduce the same transform, but it includes load-from-memory and cost model constraints, so should be safer. And I expect that we would still run -instcombine after -vector-combine in a typical opt pipeline, so there shouldn't be much chance of losing overall optimization power.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D50992/new/

https://reviews.llvm.org/D50992