Describe what you are looking for

On-the-fly transposition is prohibitively expensive some ISAs like SME. This leads to our logically cheaper symmetric operations being much slower than packed once, despite requiring less memory traffic. We should consider pre-packing both inputs before passing to our multiplication APIs, but it it introduces issues for some ISAs like AMX, where the swizzling order of A and B arguments is fundamentally different.

Can you contribute to the implementation?

• I can contribute

Is your feature request specific to a certain interface?

It applies to everything

Contact Details

No response

Is there an existing issue for this?

• I have searched the existing issues

Code of Conduct

• I agree to follow this project's Code of Conduct