def run(self, q0=None, dt=None, tend=None, iterations=None, **kw):
"""Run in parallel (PFASST)."""
if q0 is None:
raise ValueError, "missing initial condition"
rank = self.mpi.rank
state = self.state
ntime = self.mpi.ntime
levels = self.levels
self.state.dt = dt
self.state.tend = tend
for F, G in self.fine_to_coarse:
F.tmat, F.tmask = time_interpolation_matrix(F.sdc.nodes, G.sdc.nodes)
F.rmat, F.rmask = time_interpolation_matrix(G.sdc.nodes, F.sdc.nodes)
# time block loop
nblocks = int(math.ceil(tend / (dt * ntime)))
for block in range(nblocks):
step = block * ntime + rank
t0 = dt * step
state.set(t0=t0, block=block, step=step)
self.spread_q0(q0, t0, dt, **kw)
if self.mpi.ntime > 1:
self.predictor(t0, dt, **kw)
for F in self.levels:
F.q0[...] = F.qSDC[0]
for k in range(iterations):
self.iteration(k, t0, dt, **kw)
# loop
if nblocks > 1 and block < nblocks:
if self.mpi.nspace > 1:
raise ValueError, "looping not implemented for nspace > 1"
T = levels[0]
self.mpi.comm.Bcast(T.qend, root=self.mpi.ntime - 1)
q0 = T.qend
for F in levels:
F.call_hooks("end-step", **kw)
# done
for F in levels:
F.call_hooks("end", **kw)
def run(self, u0, dt, tend, iterations, cycle='V', **kwargs):
"""Run in parallel (PFASST)."""
#### short cuts, state, options
levels = self.levels
nlevels = len(levels)
rank = self.mpi.rank
ntime = self.mpi.ntime
T = levels[0] # finest/top level
B = levels[nlevels-1] # coarsest/bottom level
self.state.dt = dt
self.state.tend = tend
iterations = optdb.iterations or iterations
cycles = self.cycles(optdb.cycle or cycle)
#### build time interpolation matrices
for l in range(nlevels-1):
nodesF = levels[l].sdc.nodes
nodesG = levels[l+1].sdc.nodes
levels[l].time_interp_mat = time_interpolation_matrix(nodesF, nodesG)
#### block loop
nblocks = int(math.ceil(tend/(dt*ntime)))
for block in range(nblocks):
step = block*ntime + rank
t0 = dt*step
self.state.t0 = t0
self.state.block = block
self.state.step = step
self.set_initial_conditions(u0, t0, dt, **kwargs)
self.predictor(t0, dt, **kwargs)
# iterations
for k in range(iterations):
self.state.iteration = k
self.iteration(k, t0, dt, cycles, **kwargs)
# loop
if nblocks > 1 and block < nblocks:
if self.mpi.nspace > 1:
raise ValueError, 'looping not implemented for nspace > 1'
self.mpi.comm.Bcast(T.qend, root=self.mpi.ntime-1)
u0 = T.qend
T.call_hooks('end-step', **kwargs)
#### done
T.call_hooks('end', **kwargs)
