[PATCH] D106721: [AArch64] Implemnt MSVC __mulh and __umulh builtins and corresponding IR level intrinsics

[Reid Kleckner via Phabricator via llvm-commits]

rnk added a comment.

In D106721#2902586 <https://reviews.llvm.org/D106721#2902586>, @efriedma wrote:

> We won't call compiler-rt for i128 multiply on aarch64.  It's not worthwhile under any circumstances.
>
> Pattern-matching 64/64->128 multiply in SelectionDAG legalization has been reliable in practice, as far as I know.  And even it misses due to weird behavior in instcombine or something like that, we only end up with one or two extra mla instructions, so not a big deal.

Sounds reasonable to use i128 to me then.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D106721/new/

https://reviews.llvm.org/D106721