Skip to content

Demux refactoring without backend conversion #2339

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
wants to merge 6 commits into
base: master
Choose a base branch
from
+230 −138
Open

Demux refactoring without backend conversion #2339

Show file tree
Hide file tree
Changes from 5 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
9f72886
Reduce lock contention in demux
Menkib64 Sep 29, 2025
9bf29e6
Add preferred batch step to xla backend
Menkib64 Oct 1, 2025
441aadb
Fix demux race between workers and computation destructor
Menkib64 Oct 1, 2025
bec0303
Add preferred batch step heuristic to cuda
Menkib64 Oct 10, 2025
c655c82
Fix slow performance when the last backend gets only a partial split
Menkib64 Oct 14, 2025
6189d90
Fix wrong word in a comment
Menkib64 Nov 6, 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
6 changes: 6 additions & 0 deletions src/neural/backends/cuda/network_cuda.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -893,6 +893,12 @@ class CudaNetwork : public Network {
return 2 * sm_count_;
}

int GetPreferredBatchStep() const override {
int preferred_split = 7;
while (sm_count_ % preferred_split != 0) preferred_split++;
return preferred_split;
}

int GetThreads() const override { return 1 + multi_stream_; }

std::unique_ptr<NetworkComputation> NewComputation() override {
Expand Down
Loading