[PATCH] D69498: IR: Invert convergent attribute handling

[John McCall via Phabricator via cfe-commits]

rjmccall added a comment.

As I've said before, I absolutely view this polarity flip as good for GPU compilers, and I don't mind Clang needing to take some patches to update how we operate when targeting the GPU and to update some GPU-specific tests.  I do not think we should view GPU targets as "marginal".  But I do think that GPUs have different base semantics from CPUs (under normal, non-vector-transformed rules) and that it's not appropriate to unconditionally make those weaker GPU semantics the base semantics of LLVM IR.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69498/new/

https://reviews.llvm.org/D69498