def func(nglobal):
d = np.array(comm.rank)
with filter_comm(comm.rank < nglobal, comm) as newcomm:
if newcomm is not None:
newcomm.Allreduce(MPI.IN_PLACE, as_mpi(d))
d = comm.bcast(d)
n = min(comm.size, nglobal)
assert d == n * (n - 1) // 2
def func(nglobal):
d = np.array(comm.rank)
with filter_comm(comm.rank < nglobal, comm) as newcomm:
if newcomm is not None:
newcomm.Allreduce(MPI.IN_PLACE, as_mpi(d))
d = comm.bcast(d)
n = min(comm.size, nglobal)
assert d == n * (n - 1) // 2
def diffTdiff(input, output, axis=0, scalar=1., comm=None):
"""
Inplace discrete difference transpose times discrete difference
"""
if not isinstance(input, np.ndarray):
raise TypeError('Input array is not an ndarray.')
if input.dtype != var.FLOAT_DTYPE:
raise TypeError('The data type of the input array is not ' + \
str(var.FLOAT_DTYPE.type) + '.')
if axis < 0:
raise ValueError("Invalid negative axis '" + str(axis) + "'.")
if comm is None:
comm = MPI.COMM_WORLD
scalar = np.asarray(scalar, var.FLOAT_DTYPE)
ndim = input.ndim
if ndim == 0:
output.flat = 0
return
if axis >= ndim:
raise ValueError("Invalid axis '" + str(axis) + "'. Expected value is" \
' less than ' + str(ndim) + '.')
inplace = input.__array_interface__['data'][0] == \
output.__array_interface__['data'][0]
if axis != 0 or comm.size == 1:
if input.size == 0:
return
tmf.operators.difftdiff(input.ravel(), output.ravel(), ndim-axis,
np.asarray(input.T.shape), scalar, inplace)
return
if product(input.shape[1:]) == 0:
return
with filter_comm(input.shape[0] > 0, comm) as fcomm:
if fcomm is not None:
status = tmf.operators_mpi.difftdiff(input.ravel(), output.ravel(),
ndim-axis, np.asarray(input.T.shape), scalar, inplace,
fcomm.py2f())
if status != 0: raise RuntimeError()
