Join GitHub today
GitHub is home to over 50 million developers working together to host and review code, manage projects, and build software together.Sign up
GitHub is where the world builds software
Millions of developers and companies build, ship, and maintain their software on GitHub — the largest and most advanced development platform in the world.
Support for the STRUMPACK sparse direct solver and preconditioner [strumpack-dev] #222
Pull request for comments about the integration of
I just had a chance to look through this and I think its an excellent start. It seems to closely follow the SuperLU implementation as we had planned. The issue you had mentioned regarding copying the matrix is also an issue with the SuperLU implementation. Perhaps in the future we can find a common solution but I see no immediate need to address this.
I would like to try to build and test the new code using the same version of STRUMPACK that you have developed against. Is "STRUMPACK-Sparse 1.1.0" from November 8, 2016 (on http://portal.nersc.gov/project/sparse/strumpack/) a suitable choice or is there something more recent?
to set STRUMPACK solver options. Pass the command line arguments to STRUMPACK. Remove some debugging statements. Make the indentation more consistent.
Last week I moved the STRUMPACK code to a public github repository:
And here are the steps I took in order to compile mfem with STRUMPACK. In the same folder as where I have the mfem/ source I run the following commands:
make -j 4 VERBOSE=1
I will run some benchmarks next week with STRUMPACK as a preconditioner. This requires tuning some options. I think for now, it would be good if you can just try it as a direct solver, which is the default setting for now.
STRUMPACK also depends on ParMetis and Metis.
In INSTALL, it says that "STRUMPACK requires the PT-Scotch and Scalapack libraries." But STRUMPACK also requires ParMetis and Metis, just like SuperLUDist.