[PATCH] D122918: [RISCV][CodeGen] Support float-arith in Zfinx

Hugh Perkins via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Apr 13 15:31:22 PDT 2022


hughperkins added a comment.

In D122918#3448555 <https://reviews.llvm.org/D122918#3448555>, @asb wrote:

> I see your point, though the main reasoning about keeping F and Zfinx together is that they are largely the same instructions, and the hope is to define instructions and patterns together where feasible.

Yeah, I totally understand that. And when I first looked at this PR, I thought that made total sense. But then I saw how challenging it is to try to combine the two.

As far as the general principle of 'lets keep things together with similar patterns', the same argument can be levied against the pair of F and D, and yet they are in separate files, and I currently feel putting F and D in separate files is likely a reasonable idea. Despite their apparent similarities (32 vs 64 bits...), it is simpler and easier to keep D and F separate. Compared to D vs F, F vs Zfinx are I feel more different, and there is more reason to keep them apart. But even if they were the same difference as D and F, I feel keeping Zfinx and F coudl be a good idea:

- Zfinx is fairly niche, and seems unfair to 'pollute' F with it?
- I suspect Zfinx might be easier to implement if it is kept separaet fomr F.

Anyway, I seem likely to at least dabble in creating a decoupled version of Zfinx. I sort of have a hunch it will be eiaser. If it's not, well, I will come back, and say "well, yes, you were right, it is good to keep Zfinx and F together because ... [reasons X, Y, Z]".


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D122918/new/

https://reviews.llvm.org/D122918



More information about the llvm-commits mailing list