Skip to content

[vla, data] perf: pin pyav decoder to single thread #135

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
nullnonenilNULL merged 1 commit into from
Jul 30, 2026
Merged

[vla, data] perf: pin pyav decoder to single thread #135

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: 8 additions & 0 deletions loongforge/embodied/data/datasets/video_backends.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,6 +114,14 @@ def _decode_pyav(video_path, timestamps=None):
import av
container = av.open(str(video_path))
stream = container.streams.video[0]
# Single-threaded decode. CPU parallelism here is owned by --num-workers;
# libav slice threading would be a second knob over the same cores, and the
# two multiply (num_workers * world_size * threads-per-core). It is also the
# worse knob for this access pattern (many small independent seek-and-grab
# decodes): processes scale near-linearly, slice threading gives ~1.6x at
# best, and thread_count=0 ("auto") pays a per-core thread-pool setup on
# every av.open that a few decoded frames never amortize.
stream.thread_count = 1
first_ts = min(timestamps)
last_ts = max(timestamps)
seek_pts = int(first_ts / stream.time_base)
Expand Down
Loading