Skip to content

[None][feat] Support DeepGEMM swap-AB on sm100 #7355

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

Merged
litaotju merged 2 commits into from
Sep 1, 2025
Merged

[None][feat] Support DeepGEMM swap-AB on sm100 #7355

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
edeac06
Support DeepGEMM swap-AB on sm100
Barry-Delaney Aug 29, 2025
f4978a0
Reduce jit time for swap AB
Barry-Delaney Sep 1, 2025
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
3 changes: 2 additions & 1 deletion .gitmodules
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,4 +28,5 @@
url = https://github.com/zeromq/cppzmq.git
[submodule "3rdparty/DeepGEMM"]
path = 3rdparty/DeepGEMM
url = https://github.com/deepseek-ai/DeepGEMM.git
url = https://github.com/ruoqianguo/DeepGEMM.git
branch = dev/ruoqiang/swapab_sm100
11 changes: 6 additions & 5 deletions tensorrt_llm/_torch/custom_ops/torch_custom_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -926,8 +926,9 @@ def get_valid_tactics(
inputs: List[torch.Tensor],
profile: OptimizationProfile,
) -> List[int]:
# Encode swap_ab as False (0) and True (1). Currently only add one tactic here.
return [0]
# Encode swap_ab as False (0) and True (1). Currently enabled when GEMM m <= 128.
input, _, _ = inputs
return [0, 1] if input.shape[0] <= 128 else [0]

def forward(
self,
Expand All @@ -941,9 +942,9 @@ def forward(
device=input.device,
dtype=self.output_dtype,
)
# TODO: add swap_ab=tactic == 0 to detemrmine the swap_ab value
# Treat the default tactic=-1 as swap_ab=False
deep_gemm.fp8_gemm_nt(

forward_func = deep_gemm.fp8_gemm_ntt if tactic == 1 else deep_gemm.fp8_gemm_nt
forward_func(
(a, a_sf),
(weight, weight_scale),
output,
Expand Down