Nusrat Islam e1c20e7f24 ext-src: Improved allreduce performance in cpx mode for MI308 (#1393) ...

To get the improved performance for TP=4, the user needs to use
RCCL_MSCCL_FORCE_ENABLE=1 and MSCCLPP_READ_ALLRED=1. For TP=8, the
user should use MSCCLPP_HIERARCHICAL_ALLRED=1.

[ROCm/rccl commit: 0fb3b5eba9]

2024-10-30 08:30:15 -05:00

projects/rccl

ext-src: Improved allreduce performance in cpx mode for MI308 (#1393)

2024-10-30 08:30:15 -05:00

S

Description

No description provided

282 MiB

Languages

C++ 67.5%

C 20.6%

Python 6.6%

CMake 3.4%

Shell 0.6%

Other 1.1%