[PATCH] D41856: [PowerPC] Zero-extend the compare operand for ATOMIC_CMP_SWAP

[Eli Friedman via Phabricator via llvm-commits]

efriedma added inline comments.


================
Comment at: lib/Target/PowerPC/PPCISelLowering.cpp:8876
+  // ATOMIC_CMP_SWAP is Legal and ATOMIC_CMP_SWAP_WITH_SUCCESS needs to be
+  // Expanded.
+  return ToExpand ? SDValue() : Op;
----------------
nemanjai wrote:
> efriedma wrote:
> > nemanjai wrote:
> > > efriedma wrote:
> > > > Do you really need to custom-lower ATOMIC_CMP_SWAP_WITH_SUCCESS, given that we're just going to call into the custom lowering code again with the generated ATOMIC_CMP_SWAP?
> > > Well, unless your patch or Uli's (D41798 or D38215) to zero-extend the operand to the generated `SETCC` lands, I need to do it for both, don't I? I certainly continue to get failures if I don't do this for both nodes without one of those patches.
> > Oh, yes, you're right.  Hopefully we can get that landed soon, though.
> OK. So I'll update this to only handle `ATOMIC_CMP_SWAP` assuming that the other patch will be committed soon. When they land, I'll ask for both to be merged in 6.0. Does that sound like a reasonable plan?
Yes, sounds good.


Repository:
  rL LLVM

https://reviews.llvm.org/D41856