Too many tiles for larger runs of time_dpotrf_tile
I am trying to run Chameleon on 144 nodes (1 process per node with N=20k/node) which leads to the following error:
aprun -N 1 -cc none -n 144 -d 24 ./timing/time_dpotrf_tile -n 245760 --nb=320 -P 12
Profiling throught FxT has not been enabled in StarPU runtime (configure StarPU with --with-fxt)
CHAMELEON FATAL ERROR: RUNTIME_desc_create(): Too many tiles in the descriptor for MPI tags
Is this an inherent problem (caused by the limits of the MPI tag space) or is there a way to work around it?