[Perf][Distributed] Add low-SM multimem reduce-scatter for SM100/SM103 (#55072)
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com> Signed-off-by: Summer Yang <girasoleyang@gmail.com> Co-authored-by: Summer Yang <girasoleyang@gmail.com>
This commit is contained in:
co-authored by
Summer Yang
parent
8b365ff949
commit
442bd00d9f
@@ -3159,6 +3159,26 @@ def mnnvl_lamport_reduce_scatter(
|
||||
)
|
||||
|
||||
|
||||
def mnnvl_multimem_reduce_scatter(
|
||||
fa: int,
|
||||
inp: torch.Tensor,
|
||||
out: torch.Tensor,
|
||||
local_buffer: int,
|
||||
multicast_buffer: int,
|
||||
stage_sz_bytes: int,
|
||||
block_limit: int,
|
||||
) -> None:
|
||||
torch.ops._C_custom_ar.mnnvl_multimem_reduce_scatter(
|
||||
fa,
|
||||
inp,
|
||||
out,
|
||||
local_buffer,
|
||||
multicast_buffer,
|
||||
stage_sz_bytes,
|
||||
block_limit,
|
||||
)
|
||||
|
||||
|
||||
def dispose(fa: int) -> None:
|
||||
torch.ops._C_custom_ar.dispose(fa)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user