[PATCH] D87188: [InstCombine] Canonicalize SPF to abs intrinc

Tue Nov 10 06:32:17 PST 2020

spatel added a comment.

In D87188#2348343 <https://reviews.llvm.org/D87188#2348343>, @spatel wrote:

> In D87188#2348326 <https://reviews.llvm.org/D87188#2348326>, @nikic wrote:
>
>> Reopening this so we don't forget...
>>
>> I believe @spatel is working on the cost modelling. I did not have much luck tracking down the miscompile, at least did not spot anything incriminating in the llvm-diff.
>
> Yes, I'm working through the mess of the TTI cost model with things like:
> c963bde0152a <https://reviews.llvm.org/rGc963bde0152a741d58b9e1e9ac485d8f2e6c6d58>
> It's a slog...

D90554 <https://reviews.llvm.org/D90554> / f7eac51b9b <https://reviews.llvm.org/rGf7eac51b9b3f780c96ca41913293851c5acb465b>
I think that change makes this patch ready to try again. If we do see regressions, then it should now be easy to adjust target-specific cost modeling of abs() intrinsics to fix those.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D87188/new/

https://reviews.llvm.org/D87188