perf: cache default ArraySpec for regular chunk grids#3908
perf: cache default ArraySpec for regular chunk grids#3908d-v-b wants to merge 10 commits intozarr-developers:mainfrom
Conversation
For regular grids, all chunks have the same codec_shape, so we can build the ArraySpec once and reuse it for every chunk — avoiding the per-chunk ChunkGrid.__getitem__ + ArraySpec construction overhead. Adds _get_default_chunk_spec() and uses it in _get_selection and _set_selection. Saves ~5ms per 1000 chunks. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #3908 +/- ##
=======================================
Coverage 93.23% 93.24%
=======================================
Files 87 87
Lines 11696 11702 +6
=======================================
+ Hits 10905 10911 +6
Misses 791 791
🚀 New features to boost your workflow:
|
Merging this PR will improve performance by 11.1%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ⚡ | WallTime | test_slice_indexing[None-(slice(None, None, None), slice(None, None, None), slice(None, None, None))-memory] |
389.8 ms | 352.5 ms | +10.59% |
| ⚡ | WallTime | test_slice_indexing[None-(slice(0, None, 4), slice(0, None, 4), slice(0, None, 4))-memory] |
386.7 ms | 348.3 ms | +11.01% |
| ⚡ | WallTime | test_slice_indexing[None-(slice(10, -10, 4), slice(10, -10, 4), slice(10, -10, 4))-memory] |
211.6 ms | 190.4 ms | +11.1% |
| ⚡ | WallTime | test_slice_indexing[None-(slice(None, None, None), slice(0, 3, 2), slice(0, 10, None))-memory_get_latency] |
4.3 ms | 3.9 ms | +10.54% |
Comparing d-v-b:perf/cache-default-chunk-spec (6fa93cb) with main (029c376)2
Footnotes
-
6 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩
-
No successful run was found on
main(c4730be) during the generation of this report, so 029c376 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩
For regular grids, all chunks have the same codec_shape, so we can
build the ArraySpec once and reuse it for every chunk — avoiding the
per-chunk ChunkGrid.getitem + ArraySpec construction overhead.
Adds _get_default_chunk_spec() and uses it in _get_selection and
_set_selection. Saves ~5ms per 1000 chunks.
Co-Authored-By: Claude Opus 4.6 (1M context) noreply@anthropic.com