[Users] Difference in NaN checks results of GRHydro

Hee Il Kim heeilkim at gmail.com
Sun Jul 18 20:16:47 CDT 2010


Recently, I found the nancheks of GRHydro show different results sometimes
depending on the compilers and codes. As far as I've tested, nans appear
only if I use the private thorn GRHydro_RNSID (just a clone of Whisky_RNSID)
with XL compilers on AIX machines. I don't see any nans with Intel compilers
for the same parfile. Also old Whisky with BSSN_MoL passes the nanchecks

The difference happens actually for somewhat specific situations. It only
happens for differentially rotating stars. It disappears if I take a smaller
# of refinement levels.

I've tested it applying the IEEE strict floating point options to both
compilers (Intel and XL), but the results were same.

Since the problem doesn't happen for Whisky+BSSN_MoL, I think it is not the
problem of the private thorn only. My GRHydro_RNSID is basically the same
with Whisky_RNSID. Assuming most of you are using the similar thorn, I
attached the thornlist and the parfile. I hope you can examine and fix the


Hee Il
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.einsteintoolkit.org/pipermail/users/attachments/20100719/ef6395d1/attachment-0001.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test_l3.par
Type: application/octet-stream
Size: 12419 bytes
Desc: not available
Url : http://lists.einsteintoolkit.org/pipermail/users/attachments/20100719/ef6395d1/attachment-0002.obj 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: rns_et.th
Type: application/octet-stream
Size: 1856 bytes
Desc: not available
Url : http://lists.einsteintoolkit.org/pipermail/users/attachments/20100719/ef6395d1/attachment-0003.obj 

More information about the Users mailing list