[PATCH] D130675: [RFC][DAG] Match select of constant equivalents in foldBinOpIntoSelect.

[Amaury SECHET via Phabricator via llvm-commits]

deadalnix added inline comments.


================
Comment at: llvm/test/CodeGen/X86/memcmp-more-load-pairs-x32.ll:165
+; X86-NEXT:    movl $-1, %eax
+; X86-NEXT:    cmovael %ecx, %eax
 ; X86-NEXT:    popl %esi
----------------
deadalnix wrote:
> deadalnix wrote:
> > gchatelet wrote:
> > > gchatelet wrote:
> > > > RKSimon wrote:
> > > > > deadalnix wrote:
> > > > > > @RKSimon Is this an improvement or a regression? Or it doesn't matter?
> > > > > Avoiding cmov for similar codesize is usually a win - they can't be predicted and have tight eflags dependencies
> > > > The codegen of llvm-libc `memcmp` and `bcmp` should directly benefit from it :-) Thx!
> > > Ha sorry I read it in the opposite direction. So this would be a regression...
> > I'm working on a fix for this :) unfortunately, it creates some regressions in the powerpc target, so it's a bit like playing wack a mole.
> So D130765 fixes this, but in turn seems to introduce some regressions for PowerPC.
The regression for PowerPC are gone, I'm going to try to rebase this.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D130675/new/

https://reviews.llvm.org/D130675