Under what circumstances does Elemental's distributed SVD not work? [closed]

I am playing around with Elemental's distributed singular value decomposition and am running into two particular issues.

Building the test at tests/lapack_like/SVD.cpp, and running with

mpirun -np 4 ./SVD --height 3 --width 3


I get

SVD residual was unacceptably large


for the distributed Complex<float> test (but sequential is OK).

Additionally, when I run with --height 50 --width 52, I run into some problem

H and A must have the same height


The Elemental documentation for SVD is outdated, and I don't know what requirements there are on the dimensions of a matrix in order to take its SVD.

I have Elemental built with pulled OpenBLAS, Scalapack, Metis and Parmetis. I could not manage to get libFLAME linked, even using the reference BLAS/LAPACK implementations.

• I think that question is too specific to Elemental. Maybe you should ask it in the mailing list: users@libelemental.org. Sep 20 '16 at 12:55
• Yeah, I should probably do that too. I was thinking that it might be some algorithmic detail that puts specific requirements on the dimensions/values on an allowable matrix, which I would have look outside of the documentation for. Sep 20 '16 at 14:26