Sorry I lost track of this email thread. I got caught up with other things. I'm wondering whether you have tried compiling with fp-precise? I'm just wondering whether it could be a problem with Intel's optimization.
--Steve
On 9/1/2021 6:21 AM, Hee Il Kim wrote:
Hi,
As James Healy mentioned in the post "[Users] Einstein Toolkit and modern AMD supercomputer", I found static_tov.par passed the test for the Intel issue. So some bns-generic codes would make NaNs. But less probable because the only bns-generic code would be NSTracker but I've already encountered the NaN production issue w/wo NSTracker.
NaNChecker itself could be suspected but both bns and tove tests are using NaNChecker.
Rather, it could be a generic issue of Intel compiler. There are some bugs reported and new features in handling NaNs for the case of ifx. Those look just irrelevant yet.
Hee Il
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users