[PATCH] D136946: [AMDGPU] Enable `update/mov.dpp` selection with `+dpp,+gfx8-insts`

[Pierre van Houtryve via Phabricator via llvm-commits]

Pierre-vh added a comment.

In D136946#3891728 <https://reviews.llvm.org/D136946#3891728>, @foad wrote:

> Typo "gfx10" in title.
>
> Why do you need +dpp as well as +gfx8-insts? Doesn't +gfx8-insts kind of imply +dpp already?

I'm not sure - I used both to be safe because the intrinsic is dpp specific. If everyone agrees we just need gfx8-insts then sure I'll remove the need for DPP.
Though, the change in `pseudoToMCOpcode` will still need to be DPP-specific because if I only check for GFX8-insts, it creates a mess (about 90 tests failures + various crashes). It seems like it needs to return -1 in some cases so I did a very targeted fix there (which I don't really like, if there's a better option I'd love to do something else really)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D136946/new/

https://reviews.llvm.org/D136946