[RFC]: Add BLAS bindings and implementations for linear algebra (tracking issue)

Question

[RFC]: Add BLAS bindings and implementations for linear algebra (tracking issue)

aman-095 opened this issue 2 years ago · 13 comments

Description

This RFC proposes adding BLAS routines. The purpose of this issue is to serve as a tracking issue for adding BLAS routines.

Implementation Status for BLAS Routines

Level 1

Single Precision

Double Precision

Complex Single

Complex Double

Level 2

Single Precision

Double Precision

Complex Single

Complex Double

Level 3

Single Precision

Double Precision

Complex Single

Complex Double

Related Issues

None.

Questions

No.

Other

No.

Checklist

I have read and understood the Code of Conduct.
Searched for existing issues and pull requests.
The issue name begins with RFC:.

Answer 1 · 2024-04-10T10:25:22.000Z

Working on idamax and isamax level 1 routines.

Answer 2 · 2024-04-12T05:31:28.000Z

@performant23 That one is already being worked on in #2104.

Answer 3 · 2024-04-12T05:32:13.000Z

@performant23 Maybe try dznrm2?

Answer 4 · 2024-04-12T05:37:19.000Z

Oh, hey! I think csscal is different right? In cscal we have a complex scalar and in csscal we have a real scalar if I remember correctly.

Answer 5 · 2024-04-12T05:40:20.000Z

Ah, you're right. That one should be straightforward.

Answer 6 · 2024-04-15T20:45:06.000Z

Hi @kgryte, we would need a generic interface to perform Givens rotation with vectors of other data types like integers, right? If so, I'd like to work on such an interface i.e. grot to get an idea of how generic g* interfaces are sampled from their equivalent d* implementations and also since we have implementations for those available for this package (JS implementation for drot is reviewed), this should be good to go!

Answer 7 · 2024-04-16T04:39:56.000Z

Working on Level 1 routine drotm

Answer 8 · 2024-04-16T05:41:29.000Z

@performant23 Yes, go for it. The one thing you need to do is ensure accessor array support. See gcopy and gswap for an idea.

Answer 9 · 2024-12-22T14:06:41.000Z

For SROTMG do we need to perform the scaling check as done in the Lapack Fortren implementation .
It seems that the SciPy implementation doesn’t include this step.

Answer 10 · 2024-12-22T14:19:49.000Z

SROTMG in SciPy only returning the transformation matrix and flag .

Answer 11 · 2025-01-25T17:26:25.000Z

I have given a try to create a SROTMG package please do check it!