[Users] different results of runs on different machines

[Roland Haas]

Hello Vasilios,

> i don't know whats causing this, could the problem be caused by the
> different compilers on the two machines?
We run the Cactus testsuites on all supported machine before releases.
With very few exceptions for exotic machines/compilers all tests pass on
all machines with all compilers (see
http://einsteintoolkit.org/release-info/parse_testsuite_results.php)
using identical test data. So what I am trying to say is that while
there will be differences between different compilers (and compiler
options), they tend to be small (1e-12 relative deviation or so).
Exceptions are codes that discretely change behaviour based on data (ie.
shock capturing) and everything that involves 1/r terms and evaluations
close to zero. In particular Intel and gcc compiler will differ (unless
both are passed the -O0 option) since Intel defaults to --fast-math
which produces non ieee compliant floating point arithmetic, while
gfortran if standards comforming unless to otherwise.

My guess based on you saying "it does not work" (please provide more
information: which machines, how many MPI processes and OpenMP threads
each, ideally the full output of Cactus/Carpet is useful) would be to
check that you run with the same number of MPI processes and OpenMP
threads on both machines (Carpet outputs this information on startup).
If the number of threads is 1 in one case and larger than one in the
other then you have an OpenMP problem (somewhere).

Yours,
Roland

-- 
My email is as private as my paper mail. I therefore support encrypting
and signing email messages. Get my PGP key from http://keys.gnupg.net.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
Url : http://lists.einsteintoolkit.org/pipermail/users/attachments/20121007/3691f136/attachment.bin