[PATCH] D124419: [InstCombine] try to fold icmp with mismatched extended operands

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

In D124419#3473046 <https://reviews.llvm.org/D124419#3473046>, @nikic wrote:

> Relevant: https://discourse.llvm.org/t/aggressive-conversion-of-sext-to-zext-blocks-indvarsimplify/61561/9?u=nikic

Yes, I saw that. Putting flags on casts would probably change the way this is implemented, but it does not change the need for this transform? It seems similar to the recent bf09a925f21a <https://reviews.llvm.org/rGbf09a925f21ad04bcd541dd2dbf51bc880ea813d> - we'd end up using the ValueTracking call to annotate the zext somewhere else, then pattern match the 'zext nsw' here.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D124419/new/

https://reviews.llvm.org/D124419