Closed SanderHulst closed 1 year ago
Hi @SanderHulst, could you please share your test case?
Hi Sander, I just saw your bug report. In this case (see index of -2147483648 TAUHFT) is in 99% some NaN or INF in wave action. Basically, I added some abort there but i need to have this into the develop. Basically, your run has either some non-defined errors due to a bug in the code or it has blown up. Cheers
Describe the bug We are investigating whether a term for atmospheric stability improves our day-to-day forecast. We therefor compiled our WAVEWATCH-6.07 with the Intel OneAPI 2022.1.0 and Intel MPI 2021.6.0 in hybrid mode with the STAB3 switch. The model runs fine as long as ww3_prnc writes the wind field without air-sea temperature difference (WND), even on a cold start. However, as soon as ww3_prnc writes the wind field and the air-sea temperature difference (WNS), then ww3_multi aborts.
When I compile with debugging enabled I get (mind you, 6.07)
In the computation of subscript 3, the action density is used. Some print statements showed that it was NaN and that propagated into the index.
To Reproduce I have a testcase available that hold the wind field with and without air-sea temperature difference. I can share it.
Expected behavior WAVEWATCH should not crash on a cold start + active STAB3.
Screenshots
Additional context I prepared a fix that I'm testing in my fork of v6.07 (release v6.07.5). I've also merged it to the development branch of my fork.