search

SGpp / DisCoTec

MPI-based code for distributed HPC simulations with the sparse grid combination technique
https://sparsegrids.org/
GNU Lesser General Public License v3.0
8 stars 6 forks source link

Broadcast combischeme instead of reading it from `.json` in every rank #75

Closed freifrauvonbleifrei closed 1 year ago

freifrauvonbleifrei commented 1 year ago

It appears there are set-ups where it is not possible that every rank reads ~4MB of data in .json files.

To this end, it would be good if this section https://github.com/SGpp/DisCoTec/blob/d1a85860cb42eda856154bc9df4cb62df6160ead/src/combischeme/CombiMinMaxScheme.hpp#L144 could be reworked such that only one rank (maybe per process group) reads the data and broadcasts the resulting vectors to the other ranks.