Rich, You could certainly write a "converter" in Fortran if that is easier that then just does the usual MatView() to save it in PETSc binary format for later loading by a PETSc solver code like src/ksp/examples/tutorials. If you do we could add it to src/mat/examples/tests Barry Writing a parallel loader just seems like a huge waste of time, over just reading the ascii in sequentially and then saving as a Petsc matrix for later reading in in parallel. On Tue, 5 Apr 2005, Richard Tran Mills wrote:

Barry Smith wrote:

...

My understanding is that MatrixMarket format is different from Harwell-Boeing?

Yes, they are different. MatrixMarket format is a coordinate format that is considerably easier to parse than Harwell-Boeing, which was really designed with Fortran in mind. The matrices available at Matrix Market are usually available in both formats.

...

I could not find a Harwell-Boeing loader in src/mat/examples/tests (though there is a writer). Does anyone have C (PETSc?) code to read in Harwell-Boeing? If so we could merge it into the MatrixMarket (coordiate storage) reader code and then (with a command line switch) read in either format and dump in PETSc format?

There are some freely distributable and modifiable C routines for working with HB files available at

http://math.nist.gov/~KRemington/harwell_io/harwell_io.html

Incorporating some of these routines into a PETSc viewer seems like the best bet. I don't know of anything else in C. I have usually used Saad's SPARSKIT Fortran routines for working with Harwell-Boeing matrices.

Cheers, Richard