Looks interesting: http://librsb.sourceforge.net/

Decent speedups. No Julia interface at this point.

How does it compare with SuiteSparseGraphBLAS.jl?

Has anyone attempted to write a pure Julia package for sparse linear algebra supporting generic numeric types?

Currently, most of the effort in that area is going towards dense implementations. My guess is that the next area of focus will be generic factorizations, and that a really good sparse library is a ways away.

I think, gains wise, it should be the opposite.

There is a lot to do on sparse matrices with a language as flexible as Julia.

The BLAS thing is almost saturated.

The reason for starting on BLAS like problems is that it is really good for improving LoopVectorization, which has uses other than BLAS. Specifically, these improvements help dense algebra with operations other than `*`

and `+`

, and more complicated loop orders (eg batched matrix multiplication).