implementations.controller_classes.controller_MPI module

class controller_MPI(controller_params, description, comm)

Bases: controller

PFASST controller, running parallel version of PFASST in blocks (MG-style)

check_iteration_estimate(comm)

Routine to compute and check error/iteration estimation

Parameters:

comm – time-communicator

default(num_procs)

Default routine to catch wrong status

it_check(comm, num_procs)

Key routine to check for convergence/termination

it_coarse(comm, num_procs)

Coarse sweep

it_down(comm, num_procs)

Go down the hierarchy from finest to coarsest level

it_fine(comm, num_procs)

Fine sweeps

it_up(comm, num_procs)

Prolong corrections up to finest level (parallel)

pfasst(comm, num_procs)

Main function including the stages of SDC, MLSDC and PFASST (the “controller”)

For the workflow of this controller, check out one of our PFASST talks or the pySDC paper

Parameters:

• comm – communicator

• num_procs (int) – number of parallel processes

predict(comm, num_procs)

Predictor phase

recv(target, source, tag=None, comm=None)

Receive function

Parameters:

• target – level which will receive the values

• source – level which initiated the send

• tag – identifier to check if this message is really for me

• comm – communicator

recv_full(comm, level=None, add_to_stats=False)

Function to perform the recv, including bookkeeping and logging

Parameters:

• comm – the communicator

• level – the level number

• add_to_stats – a flag to end recording data in the hooks (defaults to False)

restart_block(size, time, u0, comm)

Helper routine to reset/restart block of (active) steps

Parameters:

• size – number of active time steps

• time – current time

• u0 – initial value to distribute across the steps

• comm – the communicator

Returns:

block of (all) steps

run(u0, t0, Tend)

Main driver for running the parallel version of SDC, MSSDC, MLSDC and PFASST

Parameters:

• u0 – initial values

• t0 – starting time

• Tend – ending time

Returns:

end values on the finest level stats object containing statistics for each step, each level and each iteration

send_full(comm=None, blocking=False, level=None, add_to_stats=False)

Function to perform the send, including bookkeeping and logging

Parameters:

• comm – the communicator

• blocking – flag to indicate that we need blocking communication

• level – the level number

• add_to_stats – a flag to end recording data in the hooks (defaults to False)

spread(comm, num_procs)

Spreading phase

wait_with_interrupt(request)

Wrapper for waiting for the completion of a non-blocking communication, can be interrupted

Parameters:

request – request to wait for