Kernels
Kernels copied to clipboard
Fortran MPI p2p is buggy
Errors with Open-MPI + NVHPC and MPICH + GCC. Adding lots of barriers doesn't fix it.
jhammond@orin:~/PRK/FORTRAN$ make mpi -j && /usr/bin/mpirun.mpich -n 12 ./transpose-p2p-mpi 100 $((12*1000))
make: Nothing to be done for 'mpi'.
Parallel Research Kernels
Fortran MPI Matrix transpose: B = A^T
Number of MPI procs 12
Number of iterations 100
Matrix order 12000
ERROR: Aggregate squared error ******************************exceeds threshold 0.000000010000000