[PATCH] D95959: [ValueTracking] improve analysis for "C << X" and "C >> X"

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

In D95959#2539817 <https://reviews.llvm.org/D95959#2539817>, @spatel wrote:

> In D95959#2539683 <https://reviews.llvm.org/D95959#2539683>, @RKSimon wrote:
>
>> This is very similar to D90479 <https://reviews.llvm.org/D90479> which is trying to do this more generally by using the KnownBits shift helpers.
>
> Ah, I forgot about that patch. Was anything holding it up?
> Yes, they're showing almost the same test diffs apart from the zext+icmp test. I don't have a preference for the fix, so we should push one of these.

D90479 <https://reviews.llvm.org/D90479> noticed that we were assuming a poison shift could be 'known' to be all zero - so I needed to go back and fix that first but I never did (got distracted by reductions, and other squirrels.....).

If you want to get this patch that's fine - D90479 <https://reviews.llvm.org/D90479> will just remove this again when I get around to looking at it.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D95959/new/

https://reviews.llvm.org/D95959