[PATCH] D69686: [LVI][CVP] Use block value when simplifying icmps

[Nikita Popov via Phabricator via llvm-commits]

nikic added a comment.

Oops, I looked at the wrong revision in my previous comment, that was for D69914 <https://reviews.llvm.org/D69914>.

Now looking at the right one: https://llvm-compile-time-tracker.com/compare.php?from=cb392c870d12eb520f84c8b7eb4f57e37483baed&to=d06337c311c36ce4fd0e939bbcef65f6d53c9d47&stat=instructions Compile-time impact is pretty minimal (< 0.1%).

Relevant stats on test-suite:

  correlated-value-propagation.NumCmps | 816 | 1004
  sccp.NumDeadBlocks | 7420 | 7413
  sccp.NumInstRemoved | 18200 | 18168

I think that the addition of range support in SCCP has already covered many of the cases I wanted this for. Still think it makes sense to do this in CVP as well, as we already have the needed information, and LVI and SCCP have somewhat different strengths.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69686/new/

https://reviews.llvm.org/D69686