Kristoffer Carlsson
Kristoffer Carlsson
I haven't tried this at all personally with Pardiso 7. Did they change any of the API?
Are the older versions not available for download?
The `set_nprocs!` calls into [`mkl_domain_set_num_threads`](https://software.intel.com/en-us/mkl-developer-reference-c-mkl-domain-set-num-threads) with the domain 4 which should be the Pardiso domain. Not sure why that wouldn't work.
Something seems off with your MKL, it works for me ``` julia> ps = MKLPardisoSolver(); julia> set_nprocs!(ps, 2) julia> get_nprocs(ps) 2 julia> set_nprocs!(ps, 8) julia> get_nprocs(ps) 8 ``` Perhaps upgrade...
Does it work if you do not have the non-MKL Pardiso loaded? Remove `ENV["JULIA_PARDISO"]`, run `build Pardiso` and try re-run the tests. I had some issues with this when both...
We have to open some libraries for non-MKL Pardiso to work and I am guessing one of them causes some issue with MKL...
> This example works on Julia v1.6.5. That's surprising to me. I don't see how a different Julia version would influence this.
Can you post the code that generates the matrix and the code you use to factorize it. Would be good to have for eventual future investigation.
Hm, it is missing the error from https://github.com/JuliaSparse/Pardiso.jl/blob/e202350860e53f62f7d4a119a8e357296e173907/deps/build.jl#L55 (I guess `stderr` doesn't go to the log?). If you run `pkg> build --verbose` what does it say?
Does it fail to just run `Libdl.dlopen` on it in a normal julia session?