[PATCH] D30081: [PPC] Eliminate more compare instructions using record-form operation

[Hal Finkel via Phabricator via llvm-commits]

hfinkel added a comment.

In https://reviews.llvm.org/D30081#685127, @nemanjai wrote:

> In https://reviews.llvm.org/D30081#684391, @inouehrs wrote:
>
> > Thank you for the comment.
> >  As you said, I did not confirm the cases for added opcodes other than `andi.`. So I will remove them.
> >  I thought it is safe since they are aliases for special cases of `rlwinm`, which is already included as a target of optimization.
>
>
> They may very well be safe. In fact, I assume they are since I'm sure you've run fairly thorough testing with this patch. But it would be good to understand the impact and try to write test cases that exercise these.


FWIW, it might be easier to write an MIR test for these than IR-level tests.


https://reviews.llvm.org/D30081