[PATCH] D113094: [ARM] Move VPTBlock pass after post-ra scheduling

Dave Green via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Thu Nov 4 02:51:23 PDT 2021


dmgreen added a comment.

In D113094#3108283 <https://reviews.llvm.org/D113094#3108283>, @SjoerdMeijer wrote:

> Sounds like a good idea. I think its current place in the pipeline was semi arbitrary, i.e. it seemed like a good place and not much thought was given about scheduling.
> I can't think of any, but just checking, any (potential) disadvantages to moving this later?

I couldn't think of any. There may be different scheduling of instructions in and around VPT blocks, which may be better or worse depending on the exact case, but isn't something we control very well at the moment (to try and create the minimum number of blocks for example). It should be an improvement for tail predicated loops and for the most part vpt blocks, I believe, even if we may find not all cases are better.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113094/new/

https://reviews.llvm.org/D113094



More information about the llvm-commits mailing list