[PATCH] D78239: [SveEmitter] Add builtins for FP conversions

Thu Apr 16 13:56:36 PDT 2020

sdesmalen added a comment.

In D78239#1985102 <https://reviews.llvm.org/D78239#1985102>, @efriedma wrote:

> > as the LLVM IR intrinsics use the <vscale x 16 x i1> as the predicate.
>
> Why are the fp conversion intrinsics special here?  Should we fix the LLVM intrinsic definitions?

I questioned that as well, but didn't really want to complicate this patch and also wasn't sure whether it is worth it because the convert intrinsics are really only ever used by the ACLE.
On the other hand, it would let us get rid of the IsFPConvert flag and the definition would better match the behaviour of the instruction.

I'm happy to change their definitions and remove the IsFPConvert flag if that has your preference.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D78239/new/

https://reviews.llvm.org/D78239