S V N Vishwanathan <[email protected]> writes: > Hi > >> > I am wondering whether I can multiply MATELEMENTAL and MATAIJ >> > matrices. If I cannot multiply those, could you build the >> > multiplication? >> >> What are the dimensions of the matrices? >> >> Jack, does Clique support multiplication of sparse matrices with >> [MC,MR] dense matrices? > > The dense matrix is very flat and long (e.g. 10 x few 1000s) and the > sparse matrix is also flat and very very long (e.g. few 1000s x few > millions). I saw the code which is used to multiply a MPIAIJ matrix with > a dense matrix. We need something very similar, but which uses elemental > matrices.
In practice, you'll probably get the (sparse) * (dense) version instead. These aren't implemented yet for Elemental matrices, but you can MatConvert to MPIDENSE and then do the sparse-dense product. Why is the original (few 1000s)x10 matrix in Elemental format? PETSc does not currently support the [VC,*] distributions. We can do it, and maybe should, but it's not what the first users were asking for and it's basically the same as MATDENSE except in its ability to interact with [MC,MR]. > There seems to be some inconsistency right now in PETSc. I can multiple > a MPIAIJ matrix with MPIDENSE, but I cannot multiply MPIDENSE with > MPIDENSE since there is no PLAPACK support anymore. MPIDENSE was a bad distribution for squarish matrices. > On other hand, I can multiply a MATELEMENTAL with MATELEMENTAL but I > cannot multiply a MPIDENSE with MATELEMENTAL. Is the eventual plan to > remove MPIDENSE completely and replace it with MATELEMENTAL? Yes, at least partially. Elemental is sort of a heavy dependency for the large faction that only need Sparse x [VC,*]. For more general distributions and some other algorithms (in the realm that PLAPACK was used in the past), Elemental is the right way to go.
pgpbbrAg8bjwb.pgp
Description: PGP signature
