use sparse multiplication for sparse arrays time BitMatrix and BitVector and wrappers

[abraunst]

My attempt to fix #39474
Not really sure about the types renaming, can be rolled back or changed of course.

[abraunst]

I can add some tests if this is deemed the way forward.

[ViralBShah]

@dkarrasch Can you also take a look?

[ViralBShah]

Let's add the tests. I'm not sure - but is nanosoldier functional at the moment?

[abraunst]

Let's add the tests. I'm not sure - but is nanosoldier functional at the moment?

About the tests... how can we target the fact that the multiplication is the sparse one?
Or maybe we just need benchmarking?

[dkarrasch]

About the tests... how can we target the fact that the multiplication is the sparse one?

I think you can do something like

m = @which mul!(C, A, B, true, false)
@test m.module == SparseArrays

Or maybe we just need benchmarking?

It would be interesting to see whether we have tests that cover the BitVector case. But to make sure we don't regress from where we currently are, let's run Nanosoldier.

@nanosoldier runbenchmarks("sparse", vs=":master")

[abraunst]

@nanosoldier runbenchmarks("sparse", vs=":master")

nanosoldier seems MIA :(

[dkarrasch]

Sometimes it's actually running but fails to leave a footprint. Unfortunately, I don't know where to look for whether it's running. I'll wait two or three hours and then try again.

[dkarrasch]

LGTM. I think this is carefully written to include "BitVecOrMat", but otherwise not change current behavior. I'd not expect to learn much from a Nanosoldiers run.

[ViralBShah]

That's what I thought as well - I think it should be good to merge.

I think nanosoldier is out of commission (and physical access to MIT is needed to fix it - which is not likely to be approved anytime soon).

[ViralBShah]

@dkarrasch Please merge if good.