Skip to content

Improve mapped kernels, window accumulation and special functions - #61

Merged
jph00 merged 1 commit into
mainfrom
pr/improve-mapped-kernels-window-accumulation-and
Nov 15, 2026
Merged

jph00 merged 1 commit into
mainfrom
pr/improve-mapped-kernels-window-accumulation-and

Conversation

@jph00

@jph00 jph00 commented Nov 15, 2026

Copy link
Copy Markdown
Contributor

Convert repeated mixed-dtype broadcast scalars once per row and reuse prepared mappings in matrix-product planning. Use chunked overlap-add for indexed float sums and add reusable transpose-of-window-sum plans for writing into existing output storage.

Use bulk slice kernels for exp, expm1, ln and log1p, and simplify the normal-range tanh calculation. Scale high-order polygamma to avoid premature overflow and underflow, with mpmath reference coverage through order 170. The Accuracy page records the numerical bounds.

Preserve checked-domain behaviour in digamma and bound incomplete beta. Handwritten Formula implementations must declare their NaN-error policy. Bound beta delegates validation and safe-domain facts to its underlying formula, retaining fused execution for proven-safe inputs.

Require Rust 1.100 across the workspace and raise the core/kernel dependency floors to the versions supplying the new shared APIs.

@jph00 jph00 added the enhancement New feature or request label Nov 15, 2026
@jph00
jph00 merged commit f6d8276 into main Nov 15, 2026
@jph00
jph00 deleted the pr/improve-mapped-kernels-window-accumulation-and branch November 15, 2026 08:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant