Skip to content

Allow specifiying the use of persistent kernel #5129

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 1 commit into from
Closed

Allow specifiying the use of persistent kernel #5129

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 7 additions & 1 deletion fbgemm_gpu/experimental/gemm/triton_gemm/fp8_gemm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1212,6 +1212,8 @@ def matmul_fp8_row(
imprecise_acc: bool = False,
tma_persistent: bool = True,
no_use_persistent: Optional[bool] = None,
# add an option to explicitly require the use of persistent process
use_persistent: Optional[bool] = None,
use_warp_specialization: bool = False,
) -> torch.Tensor:
"""
Expand All @@ -1232,12 +1234,16 @@ def matmul_fp8_row(
Returns:
torch.Tensor: [M, N] Output tensor a @ b / (a_scale[:, None] * b_scale[None, :])
"""
if no_use_persistent is None:
if use_persistent:
no_use_persistent = False
elif no_use_persistent is None:
# Default True for AMD and False for Nvidia.
if torch.version.hip is not None:
no_use_persistent = True
else:
no_use_persistent = False
# if use_persistent is explicitly requested, set o_use_persistent to False

# Get datatypes and constants to use.
pt_fp8_dtype, _, _, _ = get_fp8_constants()
# Handle 3D+ a shape
Expand Down
Loading