Skip to content

fix(tme): add multi-pool subscription for ThreadedME #925

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
shashankp-mcw wants to merge 1 commit into
base: master
Choose a base branch
from
Open

fix(tme): add multi-pool subscription for ThreadedME #925

Show file tree
Hide file tree
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
23 changes: 11 additions & 12 deletions doc/reST/cli.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -306,10 +306,10 @@ Performance Options
64-bit machines, or 16 for 32-bit machines. If the total number of threads
in the system doesn't obey this constraint, we may spawn fewer threads
than cores which has been empirically shown to be better for performance.
However, when :option:`--threaded-me` is enabled, this behavior is
overridden and a single thread pool larger than 64 threads may be
created. ThreadedME is a singleton job provider, and multiple frame
encoders may push work to it concurrently.
When :option:`--threaded-me` is enabled, small final pools are retained
so its calculated thread allocation is not truncated. ThreadedME remains
a singleton scheduler, but uses proxy job providers when its workers span
multiple physical pools. No physical pool exceeds the bitmap thread limit.

If the five pool features: :option:`--wpp`, :option:`--pmode(deprecated)`,
:option:`--pme(deprecated)`, :option:`--lookahead-slices` and :option:`--threaded-me`
Expand All @@ -320,14 +320,13 @@ Performance Options

When :option:`--threaded-me` is enabled, x265 estimates the number of
threads to assign to ThreadedME based on motion estimation workload and
spawns a dedicated threadpool (threadpool 0) for it. This pool may span
multiple NUMA nodes when the ThreadedME allocation target requires it. The remaining
threads are then used to create the other pools, which are assigned to
frame encoders and lookahead.

Frame encoders are distributed between the available thread pools,
and the encoder will never generate more thread pools than
:option:`--frame-threads`. The pools are used for WPP and for
assigns one or more leading physical thread pools to it. These pools may
span multiple NUMA nodes when the ThreadedME allocation target requires
it. The remaining pools are assigned to frame encoders and lookahead.

Frame encoders are distributed between the available frame pools,
and the encoder will never generate more frame pools than
:option:`--frame-threads`. The pools are used for WPP and for
distributed analysis and motion search.

On Windows, the native APIs offer sufficient functionality to
Expand Down
5 changes: 4 additions & 1 deletion doc/reST/threading.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -281,7 +281,10 @@ as external row dependencies are resolved. The steps are:
ThreadedME has higher threading demand because it must maintain a lead over WPP
to minimize stalls. It computes MVs for CTUs across rows within a frame and
across frames across frame encoders. It is therefore recommended primarily for
many-core systems where threading resources can be balanced.
many-core systems where threading resources can be balanced. Its scheduler is
a singleton, but its configured worker allocation may be split across multiple
physical pools through proxy job providers so that each pool remains within its
worker-bitmap limit.

SAO
===
Expand Down
Loading
Loading