#962: parallelize Multipole using OpenMP and MPI ------------------------------------+--------------------------------------- Reporter: rhaas | Owner: Type: enhancement | Status: review Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: Resolution: | Keywords: Multipole ------------------------------------+---------------------------------------
Comment (by rhaas):
I lack data. The single run I just looked at spends about half of ANALYSIS (NSNS run, so there are only Hydro_Analysis, Multipole and QLM but not eg AHFinderDirect) in Multipole. ANLYSIS was about 10% of the total run time, so I doubt Multipole is significant for runs, in particular if AEILocalInterp were OpenMP parallelized (which it currently is not). The patch *does* make MP much faster in the test case. I had the patches and logic lying around from Ylm_Decomp. I am sure I can find some Swedish person's law to claim that all part of a code must be parallelized for scalability :-)
Which do you find more objectionable: the while loop to reconstruct el and em from lm, or the MPI_Gatherv?