[PATCH] D123801: [DAG] Combine shuffle(bitcast(X), Mask) to bitcast(shuffle(X, Mask'))

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

This looks like a partial inverse of VectorCombine::foldBitcastShuf(). Would it help to limit that transform instead of or in addition to this?

The x86 diffs seem neutral, but it's a little surprising that we get any changes given that we have a bunch of code to combine shuffles that allows peeking through bitcasts. @RKSimon may have more to say on that.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D123801/new/

https://reviews.llvm.org/D123801