[Executorch] make slice_copy parallel #15830
Open
+41
−6
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
When doing large prefills in LLMs, slice_copy takes about 5-10% time. Mainly coming from slicing in the rope implementation. Differential Revision: [D85532081](https://our.internmc.facebook.com/intern/diff/D85532081/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/15830
Note: Links to docs will display an error until the docs builds have been completed. ❌ 123 New FailuresAs of commit 6f77646 with merge base 7600df8 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
kimishpatel
added a commit
that referenced
this pull request
Nov 14, 2025
When doing large prefills in LLMs, slice_copy takes about 5-10% time. Mainly coming from slicing in the rope implementation. Differential Revision: [D85532081](https://our.internmc.facebook.com/intern/diff/D85532081/) ghstack-source-id: 323355686 Pull Request resolved: #15830
manuelcandales
approved these changes
Nov 14, 2025
manuelcandales
added
the
release notes: ops & kernels
Copilot finished reviewing on behalf of
mergennachin
November 17, 2025 16:18
Contributor
There was a problem hiding this comment.
Pull Request Overview
This pull request adds parallel processing to the slice_copy operation in ExecutorTorch to improve performance during large prefills in LLMs, where slice_copy can take 5-10% of execution time (primarily from rope implementation slicing).
Key Changes:
- Added multithreading support to
compute_slicefunction with workload-based thresholds - Parallel execution distributes work across leading dimensions using
parallel_for - Single-threaded fallback maintained for smaller workloads
Reviewed Changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| kernels/portable/cpu/util/targets.bzl | Adds threadpool dependency required for parallel execution support |
| kernels/portable/cpu/util/slice_util.cpp | Implements parallel slice_copy with multithreading when leading_dims ≥ 8 and total_elements ≥ 32768, maintaining single-threaded fallback for smaller workloads |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| if (use_multithreading) { | ||
| // Use parallel_for to distribute work across leading dimensions | ||
| // Calculate grain size based on number of elements per leading dimension | ||
| const int64_t elements_per_leading_dim = length * trailing_dims; |
There was a problem hiding this comment.
The variable elements_per_leading_dim is calculated but never used. It appears this was intended for grain size calculation but MIN_LEADING_DIMS_FOR_MT is used instead. Consider removing this unused variable or using it to calculate a more dynamic grain size based on workload characteristics.
Suggested change
| const int64_t elements_per_leading_dim = length * trailing_dims; |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
When doing large prefills in LLMs, slice_copy takes about 5-10% time.
Mainly coming from slicing in the rope implementation.
Differential Revision: D85532081