[PATCH] D62787: [X86] Add the vector integer min/max instructions to isAssociativeAndCommutative.

Wed Jun 5 06:19:41 PDT 2019

spatel accepted this revision.
spatel added a comment.
This revision is now accepted and ready to land.

In D62787#1529783 <https://reviews.llvm.org/D62787#1529783>, @lebedev.ri wrote:

> In D62787#1529704 <https://reviews.llvm.org/D62787#1529704>, @spatel wrote:
>
> > In D62787#1529701 <https://reviews.llvm.org/D62787#1529701>, @RKSimon wrote:
> >
> > > This might be a sign that we need to review the generic CPU - for instance should we enable PostRA?
> >
> >
> > We need to differentiate here: the generic CPU ("target-cpu"="x86-64") is set by clang and uses the SandyBridge scheduler model, but these regression tests have *no* CPU model rather than the generic model. Maybe we should have llc set the generic model as the default?
>
>
> That will probably result in every single test being regenerated?

I haven't tried the experiment, but yes, I'd guess there would be substantial churn. But it's not good that we use/test a default configuration that does not match what most users see.

This patch has dedicated tests now, so LGTM.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62787/new/

https://reviews.llvm.org/D62787