You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
After some tests, I think I'll rather keep rank and add instance_id in the entry points' arguments, and remove those additional arguments when leaving PMPI_Init (and maybe do something else for FORTRAN, as the way args are set for FORTRAN is already specialized in current code base).
This would allow the smpi-msg-masterslave example to work the same way that it is doing now (actor initialization in PMPI_Init), while wiping out these internal arguments from argc/argv once user apps have called MPI_Init.
The way SMPI programs are called seems strange to me in current SMPI.
I can implement the following, which might be saner for users.
What do you think?
Example
Here is a tiny MPI program.
Execution with usual MPI implementation
With OpenMPI, running
outputs
Execution with SMPI
With SimGrid (81e177c), running
outputs
The text was updated successfully, but these errors were encountered: