[PATCH] D52886: [X86] Move ReadAfterLd functionality into X86FoldableSchedWrite (PR36957)

Fri Oct 5 06:19:19 PDT 2018

RKSimon added a comment.

In https://reviews.llvm.org/D52886#1256461, @courbet wrote:

> In https://reviews.llvm.org/D52886#1256458, @RKSimon wrote:
>
> > I've raised https://bugs.llvm.org/show_bug.cgi?id=39188 to discuss what to do about the broadwell load latencies - changing them would involve a notable update to the model and I'd be a lot more comfortable if we had (llvm-exegesis?) data to confirm them
>
>
> What's wrong with the data I gave above ? :)

:) I started creating a patch and realised how much of a diff it was going to cause - and that was before I'd altered any of the InstRW overrides.....  I've CC'd you on PR39188, we can continue work on it there.

PR36957 is actually a problem for us (both llvm-mca accuracy and codegen scheduling), and I'd like to concentrate fixing this first.

Repository:
  rL LLVM

https://reviews.llvm.org/D52886