Dear Joe, Yes I am interested in contributing. I have reasons to believe this will bring a significant performance boost to my simulations, but please correct me if I’m wrong.
I’ve had a look at the mumps files in kwant source, and although I can’t make full sense of it, I’m sure I’ll be able to do the job with support from the kwant community. Kindly giude me in the right direction as to where I should start. Best regards, Abdul > On 31 Mar 2016, at 10:05, Joseph Weston <joseph.westo...@gmail.com> wrote: > > If I recall correctly there was no fundamental barrier to doing this, it was > just that other things had priority. Michael will know more about this > because he wrote most of the MUMPS wrapping code. > > I had a quick look and it seems that extending to support parallel would > consist of setting > a few extra flags and possibly manually distributing the Hamiltonian matrices > over the cores that are to doing the computation (one would need to. After > that it should just be a case of running the Kwant script with `mpirun`. > > If you are motivated to work on this we would be more than happy to accept > contributions. > If that's the case you can coordinate with us (perhaps on the kwant-devel > mailing list) and we'll give you a hand to get started. > > Thanks! > > Joe > > On 30 March 2016 at 22:11, Abdulkareem Afandi <abdulkareem.afa...@gmail.com > <mailto:abdulkareem.afa...@gmail.com>> wrote: > Dear Kwant community, > > Is there a particular reason why kwant is configured to work with sequential > MUMPS instead of parallel? I am working with very large systems (~10^8 sites) > on a well equipped computer cluster, and was wondering if I could easily > extend MUMPS support to parallel instead of just sequential? > > Best regards, > Abdul >
