ROCm
Disable GSU code when GSO=0
GSU=0 should disable all GSU code. This change updates some sections of code that were still generating GSU-related code when GSU was disabled.
The purpose of GSU=0 still keep GSU codes is we want to reduce the number of kernels. @KKyang Did this impact anything if without this change?
The purpose of GSU=0 still keep GSU codes is we want to reduce the number of kernels. @KKyang Did this impact anything if without this change?
Yes, we need the ability to disable GSU code since it is not compatible with StreamK code. Without this change, stream-k kernels with epilogues fail to generate due to some extra GSU code being added. If GSU=0, there should not be any GSU code present.
Or If GSU=0 is only used for StreamK case, please reject solution for GSU=0 + StreamK=False.
Added a reject condition to require either GSU or StreamK to be enabled