Floating point exception inf pcg.c

[AlbertNichols3]

At pcg.c line 998, i_prod is negative and bi_prod is positive, Both on the order of 1e-12. The code is taking the square root of the ratio.

[rfalgout]

Hi Al. Do you have a way to turn up the print level so we can get more details on this particular solve? Both of these variables are supposed to stay positive, but they might drift away from that for a variety of reasons. Some of the questions I have include: Are you using a preconditioner? If so, what are the preconditioner settings (e.g., smoother settings for AMG, etc.)? Was this a serial or parallel run? GPU or CPU? ...

We have already added a bunch of checking in PCG for these variables, but it's clear we need to add some safeguards for the sqrt operations. We'll work on that in the meantime.

Thanks!