pydata/sparse - CodSpeed

sparse

Performance History

Latest Results

[pre-commit.ci] pre-commit autoupdate updates: - [github.com/astral-sh/ruff-pre-commit: v0.15.11 → v0.15.12](https://github.com/astral-sh/ruff-pre-commit/compare/v0.15.11...v0.15.12)

pre-commit-ci-update-config

2 days ago

perf: short-circuit COO.reshape when -1 resolves to self.shape (#935) * perf: short-circuit COO.reshape when -1 resolves to self.shape COO.reshape returns self when self.shape equals the requested shape, but only checks before resolving any -1 in the target. sparse.tensordot passes shapes like (-1, N) even for 2D x 2D matmul that doesn't actually change shape, so the short-circuit never fires and a full reshape runs (linear_loc + coord rebuild + new COO allocation). Moving the -1 resolution before the equality check avoids that work for callers that pass a -1 factorization of the current shape. Behavior is a strict subset ("fewer copies"): any reshape that already returned self before still does, and reshape((-1, ...)) that resolves to the current shape now also returns self, matching the documented contract. Measured ~16% speedup on a warm conservative-regrid loop (xarray-regrid ConservativeRegridder.regrid) whose tensordot call sits on the hot path; bit-identical output. All 6050 existing numba-backend tests pass. * perf: use `-1 in shape` for presence check in COO.reshape Replace `any(d == -1 for d in shape)` with `-1 in shape`. The latter is a C-level tuple containment, the former a Python-level generator. On this machine (micro): 221 ns -> 45 ns per check. End-to-end on the PR's repro (median of 3): reshape(a.shape): 0.4 us -> 0.2 us (matches main; erases the regression introduced by running the -1 check unconditionally) reshape((-1, K)): 2.7 us -> 2.3 us (small incremental win) Pure readability / perf nit; semantics are identical since shape is already a tuple of ints by the line above.

main

8 days ago

perf: use `-1 in shape` for presence check in COO.reshape Replace `any(d == -1 for d in shape)` with `-1 in shape`. The latter is a C-level tuple containment, the former a Python-level generator. On this machine (micro): 221 ns -> 45 ns per check. End-to-end on the PR's repro (median of 3): reshape(a.shape): 0.4 us -> 0.2 us (matches main; erases the regression introduced by running the -1 check unconditionally) reshape((-1, K)): 2.7 us -> 2.3 us (small incremental win) Pure readability / perf nit; semantics are identical since shape is already a tuple of ints by the line above.

thodson-usgs:perf/reshape-resolve-neg1-before-shortcircuit

8 days ago

perf: short-circuit COO.reshape when -1 resolves to self.shape COO.reshape returns self when self.shape equals the requested shape, but only checks before resolving any -1 in the target. sparse.tensordot passes shapes like (-1, N) even for 2D x 2D matmul that doesn't actually change shape, so the short-circuit never fires and a full reshape runs (linear_loc + coord rebuild + new COO allocation). Moving the -1 resolution before the equality check avoids that work for callers that pass a -1 factorization of the current shape. Behavior is a strict subset ("fewer copies"): any reshape that already returned self before still does, and reshape((-1, ...)) that resolves to the current shape now also returns self, matching the documented contract. Measured ~16% speedup on a warm conservative-regrid loop (xarray-regrid ConservativeRegridder.regrid) whose tensordot call sits on the hot path; bit-identical output. All 6050 existing numba-backend tests pass.

thodson-usgs:perf/reshape-resolve-neg1-before-shortcircuit

8 days ago

[pre-commit.ci] pre-commit autoupdate (#934)

main

9 days ago

[pre-commit.ci] pre-commit autoupdate updates: - [github.com/astral-sh/ruff-pre-commit: v0.15.8 → v0.15.11](https://github.com/astral-sh/ruff-pre-commit/compare/v0.15.8...v0.15.11)

pre-commit-ci-update-config

9 days ago

[pre-commit.ci] pre-commit autoupdate updates: - [github.com/astral-sh/ruff-pre-commit: v0.15.8 → v0.15.10](https://github.com/astral-sh/ruff-pre-commit/compare/v0.15.8...v0.15.10)

pre-commit-ci-update-config

16 days ago

fix: return 0-D array for full reductions per Array API standard (#932)

main

18 days ago

Latest Branches

+25%

[pre-commit.ci] pre-commit autoupdate#936

2 days ago

3b07981

pre-commit-ci-update-config

-18%

perf: short-circuit COO.reshape when -1 resolves to self.shape#935

8 days ago

487809e

thodson-usgs:perf/reshape-resolve-neg1-before-shortcircuit

-21%

fix: return 0-D array for full reductions per Array API standard#932

18 days ago

335edc9

Abineshabee:fix/array-api-reduction-returns-0d-array

Home Terms Privacy Docs