Next: Implementation of reductions, Previous: The c_mpi backend, Up: The c_mpi backend [Contents][Index]
c_mpi
When
ncptl is passed
--backend=c_mpi as a
command-line option,
c_mpi
processes the following backend-specific
command-line options:
In the generated code, MPI_Isend()
and
MPI_Irecv()
are used for asynchronous communication and
MPI_Send()
and MPI_Recv()
are normally
used for synchronous communication. However, the
c_mpi
-specific compiler option
--ssend
instructs
c_mpi
to replace all calls to
MPI_Send()
in the generated code with calls to MPI_Ssend()
, MPI’s
synchronizing send function. A program’s log files indicate whether
the program was built to use MPI_Send()
or
MPI_Ssend()
.
The default reduction operation is MPI_SUM
but a different
operation can be specified using the
c_mpi
-specific --reduce compiler option. A
program’s log files indicate the reduction operation that was
used.