Problems in the distributed EV IO
Wrong eigenvectors are read in a film noco case in some MPI configurations. Two out files are attached, the eigenvalues are the same in both, but l-like charge is different. Tracking the error back, I found that zMat%data_c read in cdnval differs.
MPI configuration lead to the wrong results: 4 nodes, 2 MPI/node (run time about 20 min). MPI configuration lead to the right results: 4 nodes, 8 MPI/node (run time about 10 min).
When run with the "-eig hdf" flag, both hang in the IO write (killed after time limit 4:30 hours)