search

PRUNERS / FLiT

A project to quickly detect discrepancies in floating point computation across hardware, compilers, libraries and software.
Other
36 stars 6 forks source link

MPI support: does not work with MPICH #222

Closed mikebentley15 closed 5 years ago

mikebentley15 commented 6 years ago

Bug Report

Description of the problem The MPI support was developed against OpenMPI, but the mpic++ command from OpenMPI has different flags than the one from MPICH. When trying to run FLiT under MPICH, it fails immediately.

Suggested fix In Makefile.in, either do a check to see what kind of mpi we are using and change the flag accordingly, or just try both sets of flags, allowing the backup set to be used if the primary fails to function correctly.

This allows swapping out the implementations of MPI by just changing your path to point to mpirun and mpic++ from the MPI runtime you desire.

Alternative approaches

There may be other approaches not thought of or discussed in this bug report

mikebentley15 commented 6 years ago

Looks like Intel's version of MPI is similar to MPICH. That means it is likely that when we get support for MPICH fully functional, we may also get it working with Intel MPI for free.

mikebentley15 commented 5 years ago

We think MVAPICH has the same interface as MPICH, so they can both be fixed by this.