[PATCH] D140840: Tests + Improve cases for optimizing out some icmp(binop) patterns (mostly mul)

[Noah Goldstein via Phabricator via llvm-commits]

goldstein.w.n added a comment.

@nikic This is somewhat unrelated so if this is not the place to ask, let me know.

I see your commit:

  [InstCombine] Make TTI comment more forceful (NFC)
  
  To save people not familiar with InstCombine from creating patches
  that are immediately rejected by policy.

I was hoping to make a patch to avoid the **de-optimization**
that occurs in `ShrinkDemandedConstant` when the reduced
bits does something like break `imm8` sign extension and forces
an `imm64` (causes `movabs` for X86, ot `movk + lsl`/`ldq` on
aarch64).

I figured the best way to do this was with `getIntImmCostInst`
but your comment is pretty forceful.

Is this an acceptable use case that could warrant an exception?
If not, do you have an advise for how to avoid this de-optimization?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D140840/new/

https://reviews.llvm.org/D140840