Present: Vassili, Peter, Erik, Ian, Philipp, Yosef * goals for next release ** suggestion: improve scaling and performance of (hydro) code ** start discussion on mailing list
* multi-line error messages ** suggestion that write(,) with proper unit numbers might be workable ** Fortran 2003 defines standard module ISO_FORTRAN_ENV which defines the error and output unit numbers. Seems to exist on ifort and gfortran (starting from 4.6) * performance improvement goals for workshop ** single processor speed improvements ** speed up GRHydro ** consider using OpenCL for at least some tests ** scaling and load balancing of Hydro code compared to vacuum code ** improve restriction and prolongation scheduling and communication * hangs on stampede ** Yosef tested with and without the parity symmetry thorn ** there are two failure modes and both are seen by Yosef even when not using the thorn * M_PI not defined by default in c99 * M_PI not defined in c99. Defined by C++ and OpenCL, usually can be made accessible by special compiler options. ** possible solutions: *** check in autoconf that M_PI is defined and abort if not *** define M_PI if not defined *** add CFLAGS options in Cactus' configure stage, requires most likely a new option BASE_CFLAGS to let users override the Cactus default. Provide only support for common (intel, gcc) compilers only. ** the last solution (CLAGS set by Cactus' configure stage) is the preferred method right now * misc ** feature request to have CarpetRegrid2 check that its grid scalars are consistent on all processes Yours, Roland
signature.asc
Description: OpenPGP digital signature
_______________________________________________ Users mailing list [email protected] http://lists.einsteintoolkit.org/mailman/listinfo/users
