Page 2 of 2

Re: Compiling parallel scientific codes on parallella

PostPosted: Thu Feb 18, 2016 8:09 pm
by sebraa
I have never used MPI. I do believe that mpirun/mpiexec do not exist on the Epiphany implementation and that another way is used. But I am not sure.

Re: Compiling parallel scientific codes on parallella

PostPosted: Fri Feb 19, 2016 3:48 am
by jar
Sebraa is correct. Currently, mpirun/mpiexec do not exist for Epiphany. You can still use OpenMPI on the ARM host with mpirun/mpiexec if you want. Threaded MPI is the model most recently published. You can read about it in this paper: http://arxiv.org/abs/1506.05442, get some of the software here: https://www.parallella.org/2015/04/09/t ... hitecture/ and see other sample codes available here: https://github.com/USArmyResearchLab/mpi-epiphany

Threaded MPI requires offload semantics (host code) because Epiphany exists as a co-processor. The actual parallel portion that runs on the Epiphany looks more like your standard MPI code, however. But this could change in the future.

Re: Compiling parallel scientific codes on parallella

PostPosted: Fri Feb 19, 2016 8:47 am
by gmogni
Yuk, this looks much more complicated than simply compiling and running "mpirun -np <nproc> <name_program> <input_file>" which is what I've been doing so far :( Maybe I'll wait a few more years before buying parallella until its MPI support will become more mature. Thanks for your help anyway.....