Skip to content

Commit

Permalink
Set use_cuda_graphs in fp8_gemm_rowwise
Browse files Browse the repository at this point in the history
Summary: The default value for use_cuda_graphs was changed to False in D64471087 and this caused slowdowns in triton/ck kernels for fp8_gemm_rowwise.

Reviewed By: danzimm

Differential Revision: D65140285

fbshipit-source-id: 4ab77537afeb9108dab7cdef6cac34aaa39d7d73
  • Loading branch information
karthik-man authored and facebook-github-bot committed Oct 29, 2024
1 parent 7d471dc commit 5099840
Showing 1 changed file with 1 addition and 0 deletions.
1 change: 1 addition & 0 deletions tritonbench/operators/fp8_gemm_rowwise/operator.py
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,7 @@ def __init__(
self, tb_args: argparse.Namespace, extra_args: Optional[List[str]] = None
):
super().__init__(tb_args, extra_args)
self.use_cuda_graphs = True
addmm_args = parse_args(self.extra_args)
if addmm_args.m and addmm_args.n and addmm_args.k:
self.shapes = [(addmm_args.m, addmm_args.n, addmm_args.k)]
Expand Down

0 comments on commit 5099840

Please sign in to comment.