avoid generic matmul fallback in kernel

[lkdvos]

This is a very minor change that avoids falling back to generic_matmatmul for the signature mul!(::StridedView, ::StridedView, ::Matrix, ::Number, ::Number), simply by promoting the matrix to a StridedView as well.

Seems to lead to a slight speed-up in my cases, but is typically subleading. Nevertheless, probably better in the long run.

[kshyatt]

This is already being worked on in Strided

[lkdvos]

Is it possible to capture all cases like this?
I was under the impression that it would be really hard to capture all combinations of StridedView vs AbstractArray, without dealing with the ambiguity issues that then follow every time LinearAlgebra changes a signature somewhere, but I haven't thought that through in great detail.
(Note that this is not about GPU vs CPU mixtures, it really is just that the buffers are StridedView wrapped and the basis transform isn't, if that makes a difference)

[kshyatt]

This is exactly the case I already hit, in which case we know the basis transform is a raw Matrix (or CuMatrix), not wrapped by anything, so I think we can provide a generic option in Strided for that case.

[kshyatt]

FWIW I really don't think we need to worry about all cases, it's probably enough to cover Matrix, Adjoint{Matrix}, Transpose{Matrix} and the equivalents for the GPU matrices.

[kshyatt]

Doing this in Strided also allows people who run into similar situations not based on this specific basis transformation use case to benefit from the speedup 🚀

[lkdvos]

Hmmm, I see your point, but honestly I would still just add the wrapper here, rather than implicitly assuming this is handled upstream.

If we ever decide to change the matrix type, and this now changes into a type that you did not specifically handle in Strided, this would break again, and the addition here really doesn't hurt 😉.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/tensors/indexmanipulations.jl	77.00% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.