For an academic numerical project we use as direct solvers mostly pardiso (from mkl) and cholmod. Now I switch to Apple Silicon (amazing fast, a quarter of the build time!) and cannot use Intel's mkl for native compiling any more.
- For BLAS openblas runs fine
- For positive definite systems cholmod works fine
- For non positive definite systems (where pardiso is great), I'm searching for a replacement
- By 01.2022 there is no academic Pardiso available any more.
One could use Apple's Accelerate Framework instead of openBLAS. SuiteSparse finds it by default.
For non-positive definite systems, one can use umfpack from SuiteSparse. However it is (in my experience) slow compared to cholmod.