INFO not checked by pdgesvd?
Posted: Mon Jul 13, 2015 6:07 pmI have a case of PDGESVD returning the "heterogeneity detected" error. I tried doing blacs_set_(ictxt, TOPSCOHRNT, 1) after sl_init_() was called, but it did not fix the problem. [maybe I need to do an explicit grid_set?]
But here's the real reason I'm writing.
PDGESVD makes one to three calls that return the INFO status word: once to DBDSQR() and twice to PDORMBR(). BUT ... PDGESVD never checks INFO.
Therefore its theoretically possible that an earlier failure would be masked by a later success.
Meantime I will divide-and-conquer to find out where my 0-rank process first diverges from the other 2 regarding the singular values. (relative error is only 1e-15, but when the numbers are printed to 17 digits, they agree exactly so I need to dig deeper.)
But here's the real reason I'm writing.
PDGESVD makes one to three calls that return the INFO status word: once to DBDSQR() and twice to PDORMBR(). BUT ... PDGESVD never checks INFO.
Therefore its theoretically possible that an earlier failure would be masked by a later success.
Meantime I will divide-and-conquer to find out where my 0-rank process first diverges from the other 2 regarding the singular values. (relative error is only 1e-15, but when the numbers are printed to 17 digits, they agree exactly so I need to dig deeper.)