ENH: linalg: Pythranized version of _solve_toeplitz.pyx

[serge-sans-paille]

No description provided.

[serge-sans-paille]

@rgommers I don't understand the remaining failure :-/

[ilayn]

re-run the test, now passing

[ilayn]

These need better error messages

[ilayn]

Needs a better error message

[serge-sans-paille]

@ilayn I did most of the requested changes, And tested the performance impact, not finding any significant changes wrt. cython implementation.

@rgommers I also added support for using blas with Pythran in meson.build, but that was my first interaction with it, so feel free to correct me.

[serge-sans-paille]

I've added the test case to pythran codebase to avoid regressing on it, see serge-sans-paille/pythran#2117

[serge-sans-paille]

The failing check seems unrelated

[serge-sans-paille]

@ilayn any though on that new revision?

[rgommers]

@rgommers I also added support for using blas with Pythran in meson.build, but that was my first interaction with it, so feel free to correct me.

The PYTHRAN_BLAS_* define you changed seems to only be used for this snippet in pythran/pythonic/numpy/dot.hpp:

#ifdef PYTHRAN_BLAS_NONE
#error pythran configured without BLAS but BLAS seem needed
#endif

#if defined(PYTHRAN_BLAS_ATLAS) || defined(PYTHRAN_BLAS_SATLAS)
extern "C" {
#endif
#include <cblas.h>
#if defined(PYTHRAN_BLAS_ATLAS) || defined(PYTHRAN_BLAS_SATLAS)
}
#endif

That immediately makes a few alarm bells go off for me - we don't have any direct cblas.h includes in SciPy, and I'm almost certain it will cause problems. See, .e.g,

scipy/scipy/_build_utils/src/npy_cblas.h

Lines 2 to 4 in e1f0d97

	* This header provides numpy a consistent interface to CBLAS code. It is needed
	* because not all providers of cblas provide cblas.h. For instance, MKL provides
	* mkl_cblas.h and also typedefs the CBLAS_XXX enums.

[serge-sans-paille]

@rgommers for that PR, I've removed the call to @ and added a FIXME. It should not be a big deal to have pythran support 1D dot product without linking to blas, but that's on the pythran side!

[rgommers]

Thanks, that sounds good!

[serge-sans-paille]

I don't understand the failure, otherwise looks good to me?

[ilayn]

Nevermind that failure, it is not related but #18654 This looks good to me. But that error change might break some code somewhere. So I'm not sure about it

[serge-sans-paille]

Gentle ping :-)