[MAP] Parallel ICOOL on NERSC: update to 332, important bug fix
Berg, J Scott
jsberg at bnl.gov
Wed Oct 9 09:58:53 EDT 2013
Just to clarify, the bug for more than 300,000 particles only occurs when you are reading the particles from a for003.dat file, so runs should only be considered suspect if you were reading from a for003.dat file.
-Scott
> -----Original Message-----
> From: Berg, J Scott
> Sent: Wed, Oct 09, 2013 09:03
> To: 'map-l at lists.bnl.gov'
> Subject: Parallel ICOOL on NERSC: update to 332, important bug fix
>
> For those of you using the parallel ICOOL on NERSC: there is an
> updated version of Rob Ryne's parallelization of ICOOL available at
>
> /project/projectdirs/map/Codes/ICOOL/mpi/bin/icool-332-mpi-00
>
> This version has all the features/fixes in the serial version 332.
> More importantly, a buffer overrun bug has been fixed which could
> cause problems for any runs with more than 300,000 particles. Any
> runs with more than 300,000 particles on any other parallel version
> should be considered suspect, so please switch to this version. If
> anyone has a need for the earlier version with that bug fixed, let
> me know and I can easily make one.
>
> The source code is at
>
> /project/projectdirs/map/Codes/ICOOL/mpi/src/332-mpi-00
>
> As a reminder, any executables in the Codes/icool331.1-mpi*
> directories will produce results with minor statistical anomalies
> due to their use of the PGI random number generator (not an ICOOL
> problem per se), so please don't use them.
>
> To re-iterate: at this point, the only ICOOL MPI version that will
> produce correct (hopefully!) results with more than 300,000
> particles is icool-332-mpi-00. If you need a parallel ICOOL with
> the physics models from an earlier ICOOL version, let me know.
>
> Send me any questions, requests, or bug reports. Thanks,
>
> -Scott
More information about the MAP-l
mailing list