[PATCH] D136478: [LegalizeVectorOps][X86][WIP] Expand vector S/USHLSAT instead of unrolling.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

I was wondering if we could use saturated add/sub (or unpack/ext + left-shift + pack?) ops somehow to better detect overflow (as x86 has really poor right shifts), but I can't think of anything :(

Please can you rebase after rGb483349c5f4f61a67cfd9add21dbee2d69833b77 <https://reviews.llvm.org/rGb483349c5f4f61a67cfd9add21dbee2d69833b77> so we have better coverage than just v4i32?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D136478/new/

https://reviews.llvm.org/D136478