[PATCH] D64508: AMDGPU: Force s_waitcnt after GWS instructions

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D64508#1579479 <https://reviews.llvm.org/D64508#1579479>, @rampitec wrote:

> Is there any documentation describing it? Also this may be a better job for wait count insertion pass or hazard recognizer, depending on the nature of the issue.


The documentation for these is really not great. It needs to be the literal next instruction, since it will jump back 8 bytes and retry in some situations. I'm not sure why it needs to be a full wait. I didn't want to rely on some other pass later to insert it, since it must be there. Having the waitcnt already exist also avoids teaching the waitcnt pass about this special case, and instead the general pre-existing waitcnt handling will take care of it.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D64508/new/

https://reviews.llvm.org/D64508