[PATCH] D77065: [ARM][MVE] Add VHADD and VHSUB patterns

[Dave Green via Phabricator via llvm-commits]

dmgreen added a comment.

We had these patterns before and took them out because they were not correct. My understanding is that these instructions do trunc(shift(add(sext(a), sext(b)), 1)). They internally operate in a higher bitwidth than we natively have.

See rGe9e1daf2b9e800c00a3c08db53650c1569288a1b <https://reviews.llvm.org/rGe9e1daf2b9e800c00a3c08db53650c1569288a1b> for the revert, which includes some examples in the commit message. I think I left the tests in for them though.

Maybe there's something we can do here with no wrap flags? Or patterns with extensions in them?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D77065/new/

https://reviews.llvm.org/D77065