Present: Roland, Peter D, Steve, David Bradley, Erik S, Ken, Miguel G, Zach, Maria
SPEC benchmark contribution * Steve provided update on status * SPEC org is proceeding ahead * will have to check license given for code ** Cactus is LGPL (v2+) release planning * Roland provided an update on POWER status * needs documenation * will not include monitor for strain due to problems getting this to work with FFI due to ringing * Zach and Maria asked about computing radiated energy. Roland reported that this is not currently present to keep POWER doing one thing and do that well. Will add how to compute energy to documenation. * feature requests: ** energy and angular momentum ** reading in ASCII files in addition to HDF5 files * need to start testing on important systems since we are at the 2 months mark making LORENE2 the default ET LORENE * Zach reported on state from 2018 and that Josh had identified and issue with convergence in LORENE but that will not affect the parts the ET uses * Roland to contact Bruno G. about updates * would at that point also update included tarball simfactory Python3 transition * are using Python3 by default right now * no breakage observed tickets: * https://bitbucket.org/einsteintoolkit/tickets/issues/2516 Zach suggested to try perturbing initial data to roundoff and check what difference is observed and if that is larger than the differences seen here ** general agreement that there should be not difference baring con2prim issues and roundoff ** Roland would want to run a short test himself to see if any differences is observerd * https://bitbucket.org/einsteintoolkit/tickets/issues/963 Peter will take a look but has not had time yet * https://bitbucket.org/einsteintoolkit/tickets/issues/2407 Erik says he will pursue this after the next release. Roland reports that currently all simfactory clusters should support this but this will require a minimum version of gcc 6.X either to compile or as a libstc++ backend for the Intel compiler. Discussed how much speed benefit (if any) the Intel compiler provides compared to GNU compilers. Zach reports on benchmarks with Baikal benchmarks and found that GNU provided faster single core performance, but some other codes (eg LORENE) was much faster with Intel. Erik and Zach suggest a single node benchmark on a real physics setup. Peter mentioned extra options required to get best performance out of compilers in particular wrt to vectorization. ET as a library * Zach reported on efforts to use ET as analysis tool for black holes at home http://blackholesathome.net/ * meant as means to increase waveform generation throughput * Zach could reproduce the gallery example GW150914 using 2GB of memory with waveforms that looks as good or better * this will require a binary library of the ET for OSX, Linux and Windows * Roland has some initial work to compile ET using the MSYS Windows build system (package system for mingw a small footprint variant of Cygwin) * Erik reported on efforts at PI to provide material for high school students Minute taker next week: Steve Chair next weel: Peter D -- My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://pgp.mit.edu .
pgpJE3afLOC6j.pgp
Description: OpenPGP digital signature
_______________________________________________ Users mailing list [email protected] http://lists.einsteintoolkit.org/mailman/listinfo/users
