Profile with kineto and warmup for more accurate benchmarking (#3580) #3585
+203
−27
Conversation
This file contains 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
|
This pull request was exported from Phabricator. Differential Revision: D68292871 |
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D68292871 |
q10
pushed a commit
to q10/FBGEMM
that referenced
this pull request
Jan 17, 2025
…h#3585) Summary: Pull Request resolved: pytorch#3585 X-link: facebookresearch/FBGEMM#667 **Summary** This PR introduces: A new warm-up method to ensure sufficient GPU preparation before benchmarking. Benchmark time calculation using the Kineto profiler for measuring the time and bandwidth of inference forward kernels. **Motivation** In small benchmark cases, kernel launch and synchronization overheads can be significant compared to the actual kernel runtime. By leveraging the Kineto profiler: These overheads are eliminated. Users get a more accurate estimation of kernel execution time and bandwidth of the forward kernel. For small kernels the iteration based warm-up might not be sufficient. By leveraging the time based warmup: Users will be confident the GPU has done enough warm-up. **Test instruction** The below script shows how to use this features: python bench/split_table_batched_embeddings_benchmark.py nbit-device-with-spec --export-trace --warmup_ms 50 Pull Request resolved: pytorch#3580 Reviewed By: leitian Differential Revision: D68292871 Pulled By: q10
…h#3585) Summary: X-link: facebookresearch/FBGEMM#667 **Summary** This PR introduces: A new warm-up method to ensure sufficient GPU preparation before benchmarking. Benchmark time calculation using the Kineto profiler for measuring the time and bandwidth of inference forward kernels. **Motivation** In small benchmark cases, kernel launch and synchronization overheads can be significant compared to the actual kernel runtime. By leveraging the Kineto profiler: These overheads are eliminated. Users get a more accurate estimation of kernel execution time and bandwidth of the forward kernel. For small kernels the iteration based warm-up might not be sufficient. By leveraging the time based warmup: Users will be confident the GPU has done enough warm-up. **Test instruction** The below script shows how to use this features: python bench/split_table_batched_embeddings_benchmark.py nbit-device-with-spec --export-trace --warmup_ms 50 Reviewed By: leitian Differential Revision: D68292871 Pulled By: q10
|
This pull request was exported from Phabricator. Differential Revision: D68292871 |
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.
Summary:
X-link: https://github.com/facebookresearch/FBGEMM/pull/667
Summary
This PR introduces:
A new warm-up method to ensure sufficient GPU preparation before benchmarking.
Benchmark time calculation using the Kineto profiler for measuring the time and bandwidth of inference forward kernels.
Motivation
In small benchmark cases, kernel launch and synchronization overheads can be significant compared to the actual kernel runtime. By leveraging the Kineto profiler:
These overheads are eliminated.
Users get a more accurate estimation of kernel execution time and bandwidth of the forward kernel.
For small kernels the iteration based warm-up might not be sufficient.
By leveraging the time based warmup:
Users will be confident the GPU has done enough warm-up.
Test instruction
The below script shows how to use this features:
python bench/split_table_batched_embeddings_benchmark.py nbit-device-with-spec --export-trace --warmup_ms 50
Reviewed By: leitian
Differential Revision: D68292871
Pulled By: q10
This is a re-export of #3580