hysop.core.mpi package

Hysop interface to the mpi implementation.

It contains :

  • mpi basic variables (main communicator, rank, size …)

  • hysop.topology.topology.CartesianTopology : mpi process distribution + local mesh

This package is used to hide the underlying mpi interface in order to make any change of this interface, if required, easiest.

At this time we use mpi4py : http://mpi4py.scipy.org

hysop.core.mpi.Wtime() float

Function to return elapsed time since some time in the past. Usage: tref = Wtime() # proceed with some computations … elapsed = Wtime() - tref # -> elapsed == time for ‘some computations’ on the current mpi process

hysop.core.mpi.default_mpi_params()[source]
hysop.core.mpi.host_comm = <mpi4py.MPI.Intracomm object>

Intrahost communicator

hysop.core.mpi.host_rank = 0

Intrahost rank

hysop.core.mpi.host_size = 1

Intrahost size

hysop.core.mpi.interhost_comm = <mpi4py.MPI.Intracomm object>

Interhost communicator (between each host local master rank)

hysop.core.mpi.interhost_rank = 0

Communicator rank between hosts

hysop.core.mpi.interhost_size = 1

Communicator size between hosts

hysop.core.mpi.intershm_comm = <mpi4py.MPI.Intracomm object>

Communicator between shared memory local master ranks

hysop.core.mpi.intershm_rank = 0

Communicator rank between shm masters

hysop.core.mpi.intershm_size = 1

Communicator size between shm masters

hysop.core.mpi.is_multihost = False

True if the programm runs on different hosts

hysop.core.mpi.is_multishm = False

True if the programm runs on different shared memory communicators

hysop.core.mpi.main_comm = <mpi4py.MPI.Intracomm object>

Main communicator

hysop.core.mpi.main_rank = 0

Rank of the current process in main communicator

hysop.core.mpi.main_size = 1

Number of mpi process in main communicator

hysop.core.mpi.processor_hash = 1138720636

MPI hashed processor name as integer (fits into a 32bit signed integer)

hysop.core.mpi.processor_name = 'runner-cqxjjwvr-project-13672-concurrent-0'

MPI processor name

hysop.core.mpi.shm_comm = <mpi4py.MPI.Intracomm object>

Shared memory communicator

hysop.core.mpi.shm_rank = 0

Shared memory process id in shm_comm (ie. NUMA node id)

hysop.core.mpi.shm_size = 1

Shared memory process count in shm_comm (ie. NUMA nodes count)

Submodules