cannam@167: Next: FFTW MPI Performance Tips, Previous: FFTW MPI Wisdom, Up: Distributed-memory FFTW with MPI [Contents][Index]
cannam@167:An MPI program can deadlock if one process is waiting for a
cannam@167: message from another process that never gets sent. To avoid deadlocks
cannam@167: when using FFTW’s MPI routines, it is important to know which
cannam@167: functions are collective: that is, which functions must
cannam@167: always be called in the same order from every
cannam@167: process in a given communicator. (For example, MPI_Barrier is
cannam@167: the canonical example of a collective function in the MPI standard.)
cannam@167:
cannam@167:
cannam@167:
The functions in FFTW that are always collective are: every
cannam@167: function beginning with ‘fftw_mpi_plan’, as well as
cannam@167: fftw_mpi_broadcast_wisdom and fftw_mpi_gather_wisdom.
cannam@167: Also, the following functions from the ordinary FFTW interface are
cannam@167: collective when they are applied to a plan created by an
cannam@167: ‘fftw_mpi_plan’ function: fftw_execute,
cannam@167: fftw_destroy_plan, and fftw_flops.
cannam@167:
cannam@167:
cannam@167:
cannam@167: