[PATCH] D142656: [SVE][codegen] Add pattern for SVE multiply-add accumulate

[Sushant Gokhale via Phabricator via llvm-commits]

sushgokh added a comment.

@SjoerdMeijer  Yes, I will add the patch with full context.

Considering the motive of the patch.  In cases where one of the addend is splat, semantics of  'mla' instruction will mandate extra register moves. This drawback is overcome by giving higher priority to 'mad' instruction than 'mla'. However, this doesnt guarantee complete freedom from extra moves as can be noted in one of the test cases(i.e. with mad, it generated one extra mov). But, overall, mad generated less count of instructions.

I added TOFIX in test cases just to keep track that things can be improved there, either over current patch or current LLVM trunk results. I will remove them as they are misleading


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D142656/new/

https://reviews.llvm.org/D142656