[PATCH] D62614: Fix for the OCL/LC to failure on some OCLPerf tests

[Alexander via Phabricator via llvm-commits]

alex-t added a comment.

I don't agree that the enhancement the definition of the "divergence" to the scope is correct way at all.
literally, the value is uniform if all threads observe same value. Nothing about exec mask, lanes or GPU :)
All threads in our case are all executing loop body. That's it.

My goal is to select instructions to SALU or VALU form in dependence of the both DA results and context.
Assigning correct register classes to the cross block values is part of this task. The problem is that current implementation only consults to DA making the decision.
All I need is to augment the DA results with the context.

Also please note: this review is about temporary solution that allow to not revert https://reviews.llvm.org/D59990#1521409

Incremental changes a easier to integrate.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62614/new/

https://reviews.llvm.org/D62614