Skip to content

[STF] Deprecate scheduling experiments#8195

Open
caugonnet wants to merge 5 commits intoNVIDIA:mainfrom
caugonnet:stf_deprecate_scheduler
Open

[STF] Deprecate scheduling experiments#8195
caugonnet wants to merge 5 commits intoNVIDIA:mainfrom
caugonnet:stf_deprecate_scheduler

Conversation

@caugonnet
Copy link
Contributor

Description

closes

Checklist

  • New or existing tests cover these changes.
  • The documentation is up to date with these changes.

Remove the device_auto automatic task scheduling feature, which was
never tested, documented, or maintained since the STF import. This
includes:

- Delete scheduler.cuh (random, round_robin, HEFT, post_mortem schedulers)
- Delete task_statistics.cuh (calibration infrastructure)
- Delete reorderer.cuh (task reordering framework)
- Remove device_auto place types from places, data_place_impl, data_place_interface
- Remove auto_scheduler and auto_reorderer from backend_ctx
- Simplify timing code in all task scopes to only use DOT timing
- Remove mapping_id, data_footprint, predecessor/successor tracking
- Simplify deferred task infrastructure (no longer needs reordering support)

DOT timing and exec_affinity are preserved.

Made-with: Cursor
With the scheduler and reorderer removed, the deferred task system in
stream_ctx had no purpose — it simply queued tasks to run in submission
order. Remove deferred_stream_task classes, deferred_task(),
deferred_host_launch(), deferred_parallel_for(), and the submit/finalize
coupling. stream_ctx::submit() is kept as a no-op for context variant
dispatch to graph_ctx.

Made-with: Cursor
@copy-pr-bot
Copy link
Contributor

copy-pr-bot bot commented Mar 26, 2026

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@cccl-authenticator-app cccl-authenticator-app bot moved this from Todo to In Progress in CCCL Mar 26, 2026
caugonnet and others added 2 commits March 27, 2026 00:12
The method now only checks if DOT timing is enabled. The old name was a
leftover from the removed scheduling infrastructure.

Made-with: Cursor
@caugonnet
Copy link
Contributor Author

/ok to test 52591c2

@github-actions
Copy link
Contributor

😬 CI Workflow Results

🟥 Finished in 12m 41s: Pass: 10%/48 | Total: 2h 20m | Max: 12m 22s | Hits: 100%/1110

See results here.

@andralex andralex marked this pull request as ready for review March 27, 2026 01:34
@andralex andralex requested a review from a team as a code owner March 27, 2026 01:34
@andralex andralex self-requested a review March 27, 2026 01:34
@cccl-authenticator-app cccl-authenticator-app bot moved this from In Progress to In Review in CCCL Mar 27, 2026
Copy link
Contributor

@andralex andralex left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sad day

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: In Review

Development

Successfully merging this pull request may close these issues.

2 participants