Evgeni Burovski

so that there's a chance to catch things like gh-273

Add a test run with torch with `torch.set_default_dtype(torch.float64)`

1

To avoid things like gh-273.

torch matmul does not handle different dtypes

3

The spec [requires that `matmul` follows the type promotion rules for the arguments](https://data-apis.org/array-api/draft/API_specification/generated/array_api.array.__matmul__.html#api-specification-generated-array-api-array-matmul--page-root ), but pytorch requires that the dtypes match: ``` In [3]: import array_api_strict as xp In [5]:...

Make `cupyx.scipy` submodule imports lazy

1

closes https://github.com/cupy/cupy/issues/8336 Two changes: First, formerly `import cupyx` eagerly imported `cupyx.scipy.fft`, `linalg` and `special`. This led to strange effects downstream: `import cupyx` made `cupyx.scipy.special` available but not `cupyx.scipy.ndimage`. Fix this...

BUG: For integer arguments, `round` returns a view while `ceil` returns a copy

5

### Describe the issue: I expected that all of `ceil`, `trunc` and `round` always return either a view or a copy for all dtypes. However, for integer arguments, `round` returns...

BUG: make round consistently return a copy

2

Otherwise, `round` returns a view for integer arguments and a copy otherwise. All other "rounding" functions (ceil, floor, trunc, rint), always return copies. Thus, make `round` consistent with the rest...

ENH: ILP64 BLAS/LAPACK support

9

#### Reference issue towards https://github.com/scipy/scipy/issues/21889 #### What does this implement/fix? Add changes needed to build SciPy with 64-bit (ILP64) BLAS and LAPACK. The main bulk of work for LAPACK detection...

A matching PR for https://github.com/data-apis/array-api-strict/pull/115, which removes an undocumented `__array__` from array-api-strict and replaces it the python 3.12+ implementation of the buffer protocol. Do not merge as is: this currently...

#### Reference issue Continues and includes gh-23052 #### What does this implement/fix? Convert - [x] PPoly - [x] BPoly - [x] CubicHermiteSpline - [x] PchipInterpolator - [x] Akima1DInterpolator - [x]...

ENH: `linalg.{inv, solve}`: follow-up improvements to low-level ND support

1

Potential follow-ups to https://github.com/scipy/scipy/pull/22924, which moved looping over batches in `linalg.inv` to C: - [ ] further memory and performance improvements (there are quite a few, potentially). For one, we...