[gmx-users] setting up mpi code
Cesar Araujo
cesar.araujo at oulu.fi
Thu Nov 10 07:58:42 CET 2005
Hi,
Floating-point comparisons are not exact due to the limited precision
representation and in most cases they suffer from round-off errors. This is not
the best way to compare a floating-point variable value against zero value or
even between them. Please, check the book "Numerical Recipes in C" for further
details about this topic. Here you will find a good practical explanation:
http://www.cygnus-software.com/papers/comparingfloats/comparingfloats.htm
If the things doesn't work as expected this could be a probable cause.
Regards,
César.-
----------------------------------
Hi
I have been trying to get gromacs 3.2.1 set up with mpi on our sgi cluster at
work. I get a number of errors that i have not received when i have compiled
previously on serial intel nodes, e.g:
orires.c(95): warning #1572: floating-point equality and inequality comparisons
are unreliable
if (ir->orires_tau == 0)
^
orires.c(104): warning #1572: floating-point equality and inequality comparisons
are unreliable
if (ir->orires_tau == 0)
^
orires.c(209): warning #1572: floating-point equality and inequality comparisons
are unreliable
(eig[j] != 0) ? eig[i]/eig[j] : 0,v[XX][i],v[YY][i],v[ZZ][i]);
^
orires.c(230): warning #1572: floating-point equality and inequality comparisons
are unreliable
bTAV = (od->edt != 0);
^
orires.c(415): warning #1572: floating-point equality and inequality comparisons
are unreliable
if (od->fc != 0) {
^
orires.c(416): warning #1572: floating-point equality and inequality comparisons
are unreliable
bTAV = (od->edt != 0);
^
plus lots lots more along the same lines!
But the code does not fail and seems to compile completely, i.e. luck works!
Should I be concerned about these errors?
I am using icc, ifort and lmpi
Thanks
Jo
--
Cesar Araujo, Lic. of Chemistry
Department of Molecular Endocrynology
Oulu University Hospital
FIN-90029 OYS, FINLAND
phone: +358 8 3155632
e-mail: cesar.araujo at oulu.fi
More information about the gromacs.org_gmx-users
mailing list