Change totalVolumeChange from ==0 to <= 1e-12 #1325
Conversation
This fixes a floating invalid error that occurs on cori-knl intel. The totalVolumeChange may be a very small value, i.e. 1e-100, because it is computed as a difference: totalVolumeChange = currentVolume - sums(variableIndex) In the previous version, it does not equal zero exactly, but still produces a floating invalid when dividing.
|
This fixes the second error described in |
|
@jonbob Please take a quick look. I would like to include this in the batch of analysis debug fixes today. It is a reported bug by nkeen on cori-knl. I am testing on MPAS stand-alone and in ACME with |
|
@mark-petersen - I won't have time to review it today, since I have a couple of critical runs to get in before leaving for the day. As long as you've tested this in ACME, I'm OK with it. But if you're not confident for any reason, please don't include it and we'll bring it in separately |
|
This passed MPAS-O nightly regression suite with gnu (bfb with previous), and ACME tests: |
This fixes a floating invalid error that occurs on cori-knl intel. The
totalVolumeChange may be a very small value, i.e. 1e-100, because it is
computed as a difference:
totalVolumeChange = currentVolume - sums(variableIndex)In the current code, it does not equal zero exactly, but still
produces a floating invalid when dividing.