These are chat archives for elemental/chat

Feb 2017
Aidan Dang
Feb 17 2017 13:56
Hi Jack, just a heads up, I think there might be another issue with SVD. I'm getting the "Secular solver did not converge in 400 iterations" (I've tried upping the iterations to 10000 too), and I think that if I try it on the Debug build, I get the "Produced non-finite sigmaEst" error. I'll get back to you with an example.
Jack Poulson
Feb 17 2017 15:28
thanks, it would be good for us to check whether or not this also effects LAPACK
there shouldn't be too many differences in the numerics
Ryan H. Lewis
Feb 17 2017 17:54
@jeffhammond hey, I just tried to build elemental with the v 17 compiler update 1, and I was able to replicate that Issue elemental/Elemental#215
Ryan H. Lewis
Feb 17 2017 18:15
@poulson ugh, forgot my jenkins password, again.
Ryan H. Lewis
Feb 17 2017 18:52
@YingzhouLi i took a stab at updating the Homebrew formula. It appears homebrew doesn’t have support for requiring C++14 which is now a requirement for Elemental.
@YingzhouLi also, the mpi issue you mentioned appears to be a bug in Homebrew, not the elemental package. It doesn’t recognize MPICH as satsifying the :mpi dependency. Not sure why.
@YingzhouLi you may as well just build from source, thats all homebrew will do anyways, there are no bottles built.
Ryan H. Lewis
Feb 17 2017 22:46
@poulson I updated our TSVD branch. I now have it compiling and running a small test case, where it segfaults.
Instead of computing the InfinityNorm(A) for operators, I borrowed a norm estimate which is what TSVD.jl by @andreasnoack is doing.