[swanmodel-users] Inconcictencies when compiling with ifort
Brought to you by:
mzijlema
|
From: Rafael G. <r.g...@me...> - 2016-06-13 06:07:39
|
Hi guys, has anyone had issues before with inconsistent results in swan code compiled with ifort please? I had that recently and have still not figured out why. My inconsistencies arise when running an ifort-compiled binary on machines with intel and AMD architectures. there are differences that randomly appear over the domain and oscillate in magnitude over time. For example, the plot attached shows differences in Hs over a 7.5-day nonstationary simulation ran on intel and AMD. The differences are not negligible (over 8cm in hs in some timesteps). and appear to fluctuate On the other hand binaries compiled with gfortran produced consistent (exactly same) results. I also played a bit with ifort optimisation flags (O0, O2, and some other combinations) but no luck - the random differences change but the overall magnitude is somewhat similar to the figure attached. Has anybody experienced something similar please? Is there any specific optimisation option flag or anything you would suggest disabling to ensure ifort-compiled binary produces same (or at least more similar) results regardless where it is running? I'm using version 40.91 by the way. Thanks, Rafael |