microbenchmarks: add buffer rotation option#677
Open
matthiasdiener wants to merge 2 commits into
Open
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
Adds an opt-in --rotating-buffers mode to the microbenchmarks to cycle input tensors through a buffer ring, reducing artificial cache-residency effects and aligning measurements with common GEMM benchmarking methodology.
Changes:
- Introduces buffer-rotation utilities (
rotating,make_input) and a new CLI flag--rotating-buffers [N]in the shared microbenchmark runner. - Updates several benchmarks (GEMM, FP8 GEMM, normalization, grouped GEMM, casting) to use rotation-aware input construction.
- Documents the new option and usage examples in the microbenchmarks README.
Reviewed changes
Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| benchmarks/microbenchmarks/utils.py | Adds rotation machinery + CLI flag plumbing and helper APIs used by benchmarks. |
| benchmarks/microbenchmarks/README.md | Documents --rotating-buffers behavior and provides example invocations. |
| benchmarks/microbenchmarks/benchmark_normalization.py | Switches normalization benchmarks to rotation-aware input generation. |
| benchmarks/microbenchmarks/benchmark_grouped_gemm.py | Updates grouped GEMM benchmark to optionally follow rotated activation buffers. |
| benchmarks/microbenchmarks/benchmark_gemm.py | Switches GEMM benchmark to rotation-aware input generation. |
| benchmarks/microbenchmarks/benchmark_gemm_fp8.py | Switches FP8 GEMM benchmark to rotation-aware input generation. |
| benchmarks/microbenchmarks/benchmark_casting.py | Uses rotation-aware inputs for quantize path and rotates FP8 tensors for dequantize path. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
matthiasdiener
marked this pull request as ready for review
July 23, 2026 21:48
matthiasdiener
requested review from
ipanfilo,
wangye805 and
wenchenvincent
as code owners
July 23, 2026 21:48
matthiasdiener
force-pushed
the
mdiener/benchmarks-rotating
branch
from
July 23, 2026 22:05
ed4f3e7 to
3209e8e
Compare
alextmagro
requested changes
Jul 24, 2026
| python benchmark_casting.py --rotating-buffers 16 # or fix the ring size (16 buffers) | ||
| ``` | ||
|
|
||
| Passing `--rotating-buffers N` fixes the number of buffers; omitting `N` |
Contributor
There was a problem hiding this comment.
Generally we would specify the # of bytes that the rotating buffer uses rather than the # of buffers -- I think that can be generally set to a safely large value, such as 4/8
| ), | ||
| ) | ||
| parser.add_argument( | ||
| "--rotating-buffers", nargs="?", type=int, const=0, default=None, metavar="N", |
Contributor
There was a problem hiding this comment.
I would argue for this to be on by default, as a standard benchmarking practice
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
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.
Description
Add a new option, --rotating-buffers, to rotate input buffers (off by default).
Follows the discussions in:
Type of change
Changes
Checklist: