Skip to content

[None][feat] Add SM107 CuTe DSL BF16 dense GEMM/BMM custom ops and dispatch #18761

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

Open
farazkh80 wants to merge 3 commits into
base: main
Choose a base branch
from
Open

[None][feat] Add SM107 CuTe DSL BF16 dense GEMM/BMM custom ops and dispatch #18761

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ad1c7eb
[None][feat] Add SM107 CuTe DSL BF16 dense GEMM/BMM custom ops and di…
farazkh80 Sep 5, 2026
8ef4fb8
[None][feat] Register the SM107 BF16 CuTe DSL ops as in-place for tor…
farazkh80 Sep 5, 2026
a8121fb
[None][fix] Apply the SM107 CTA_N=256/cluster_n=2 exclusion to the pr…
farazkh80 Sep 9, 2026
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
9 changes: 7 additions & 2 deletions tensorrt_llm/_torch/attention/mla.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1402,9 +1402,14 @@ def forward_context(
)

def _bmm_bf16_out(self, a, b_no_transpose, b_transposed, output):
"""BMM with optional CuTe DSL bf16 acceleration on Blackwell."""
"""BMM with optional CuTe DSL bf16 acceleration on Blackwell/SM107."""
if self.use_cute_dsl_bf16_bmm and is_sm_100f():
torch.ops.trtllm.cute_dsl_bf16_bmm_blackwell(a, b_no_transpose, output)
bf16_bmm_op = (
torch.ops.trtllm.cute_dsl_bf16_bmm_rubin
if get_sm_version() == 107
else torch.ops.trtllm.cute_dsl_bf16_bmm_blackwell
)
bf16_bmm_op(a, b_no_transpose, output)
else:
torch.ops.trtllm.bmm_out(a, b_transposed, output)

Expand Down
6 changes: 6 additions & 0 deletions tensorrt_llm/_torch/compilation/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -217,6 +217,12 @@ def inplace_info():
"fp8_block_scaling_bmm_out": {
1: "out"
},
"cute_dsl_bf16_bmm_rubin": {
1: "output"
},
"cute_dsl_bf16_gemm_rubin": {
1: "output"
},
"gate_forward": {
1: "out_weights",
2: "out_indices"
Expand Down
Loading
Loading