corey-derochie-amd
corey-derochie-amd
@eidenyoshida this PR only adds unit tests, no effect on performance. Regression testing not required.
@nileshnegi @nusislam @thananon I'm looking to get this one off the books. I think it's an important change to our mscclpp build to use build targets instead of procedurally scripting...
@wenkaidu I want to see whether this will cause conflicts or be redundant with the 2.27 sync in progress.
I don't *think* this is problem because we are now using a keyed map instead of an index-based array, but: Is there any chance that conditionally including these kernels could...
@thananon @nusislam this fixes a build defect with mscclpp clipping feature, it wasn't being enabled properly. This actually shows as a warning in the build log.
Recreating #1963
@nusislam This branch should now build. Do you have any comments on it?
[Issue]: RCCL collective call Alltoall is performing way worse than normal MPI Alltoall on Frontier.
Hello, @manver-iitk . Has this issue been resolved for you?