Andreas Noack
Andreas Noack
I should mention that some might not like the current behavior where `Q` can act as both square and tall so in a future version of `LinearAlgebra`, we might want...
The representation we use in Julia was completely changed, i.e. we use to hold just a pointer to the `SPQR` object but based on popular demand, I changed it to...
We should just implement our own. It's no longer an option to use the SPQR methods since we store the data in a Julia struct. It's probably not hard to...
I think we should have better support for this is base. At least `AbstractUnitRange` slices of columns of sparse matrices. I've a few ad hoc implementations of these in various...
I think your first reaction is right. I think the solver code in the `CHOLMOD` needs adjustment/renaming such that three argument `ldiv!` becomes supported. Actually, I though it already was....
It was the latter I had in mind.
@Sacha0 Would it be possible to get back to 0.5 performance if structural zeros didn't follow IEEE in `broadcast`?
> using dot ops directly (rather than via * and /) in the benchmarks above for * and / closes most of the gap This is because of the literal....
See also https://github.com/JuliaLang/julia/issues/19102
I think there might be a bug in MKL here although it seems unlikely. This program ```C #include "mkl_spblas.h" #include "stdio.h" #define M 2 #define N 2 #define NNZ 3...