#################################################################################
# WaterTAP Copyright (c) 2020-2024, The Regents of the University of California,
# through Lawrence Berkeley National Laboratory, Oak Ridge National Laboratory,
# National Renewable Energy Laboratory, and National Energy Technology
# Laboratory (subject to receipt of any required approvals from the U.S. Dept.
# of Energy). All rights reserved.
#
# Please see the files COPYRIGHT.md and LICENSE.md for full copyright and license
# information, respectively. These files are also available online at the URL
# "https://github.com/watertap-org/watertap/"
#################################################################################
import numpy
from watertap.tools.parallel.results import LocalResults
from watertap.tools.parallel.parallel_manager import build_and_execute, ParallelManager
[docs]class MPIParallelManager(ParallelManager):
[docs] def __init__(self, MPI, **kwargs):
self.MPI = MPI
self.comm = self.MPI.COMM_WORLD
self.rank = self.comm.Get_rank()
self.num_procs = self.comm.Get_size()
self.results = None
[docs] def is_root_process(self):
return self.rank == self.ROOT_PROCESS_RANK
[docs] def get_rank(self):
return self.comm.Get_rank()
[docs] def number_of_worker_processes(self):
return self.num_procs
[docs] def sync_with_peers(self):
self.comm.Barrier()
[docs] def sync_array_with_peers(self, data):
"""
Broadcast the array to all processes. this call acts as a synchronization point
when run by multiple peer mpi processes.
"""
if self.num_procs > 1:
self.comm.Bcast(data, root=self.ROOT_PROCESS_RANK)
[docs] def sync_pyobject_with_peers(self, obj):
"""
Broadcast the object to all processes from the root. this call acts as a synchronization point
when run by multiple peer mpi processes.
"""
return self.comm.bcast(obj, root=self.ROOT_PROCESS_RANK)
[docs] def combine_data_with_peers(self, data):
return self.comm.allgather(data)
[docs] def gather_arrays_to_root(self, sendbuf, recvbuf_spec):
self.comm.Gatherv(sendbuf, recvbuf_spec, root=self.ROOT_PROCESS_RANK)
[docs] def sum_values_and_sync(self, sendbuf, recvbuf):
self.comm.Allreduce(sendbuf, recvbuf)
[docs] def scatter(
self,
do_build,
do_build_kwargs,
do_execute,
all_parameters,
):
# Split the total list of values into NUM_PROCS chunks,
# one per each of the MPI ranks
divided_parameters = numpy.array_split(all_parameters, self.num_procs)
# The current process's portion of the total workload
local_parameters = divided_parameters[self.rank]
results = build_and_execute(
do_build, do_build_kwargs, do_execute, local_parameters
)
self.results = LocalResults(
self.get_rank(),
local_parameters,
results,
)
[docs] def gather(self):
return self.comm.allgather(self.results)
[docs] def results_from_local_tree(self, results):
return [
result for result in results if result.process_number == self.get_rank()
]