Present: Frank, Eloisa, Roland, Peter, Jonah, Erik, Vassili, Bhavesh, Ian
Roland asked about overlap of communication and computation during SYNC statements. * communication buffers are all data between pairs of processors * multi-threaded MPI may be required to keep a thread inside of MPI all the time for systems where MPI does no work unless a caller is inside its code. Could use self-compiled OpenMPI on blue waters or multi-threaded Cray MPI. * suggestions to use mpi trace * stand-alone test may be best * need to determine that there is actually a benefit of this overlap
Piraha parser: * currently stop parsing a CCL file at the first CCL syntax error but will parse multiple files for errors. * since syntax errors in piraha include things like mis-spelling CCTK_REAL Roland thinks that aborting at syntax errors is bad also since typically one will modify only a single CCL files when eg working on a thorn * this is however larger than the original intent of the ticket, which was about some specific files not parsing * Roland will provide a set of ccl files that should parse but do not parse
testsuite problems: * see issues on Nebula Haswel * see issues on Roland's laptop Kaby Lake * see issues on Ian's laptop/workstation (Kaby lake)
Workshop this year: * no date set yet
Yours, Roland