class Reader:
def __init__(self, name, comm):
self.comm = comm # used for broadcasting replicated data
if self.comm.size > 1:
comm = self.comm.get_c_object()
else:
comm = None
self.file = File(name, 'r', comm)
self.dims_grp = self.file['Dimensions']
self.params_grp = self.file['Parameters']
self.hdf5 = True
def dimension(self, name):
value = self.dims_grp.attrs[name]
return value
def __getitem__(self, name):
value = self.params_grp.attrs[name]
try:
value = eval(value, {})
except (SyntaxError, NameError, TypeError):
pass
return value
def has_array(self, name):
return name in self.file.keys()
def get(self, name, *indices, **kwargs):
parallel = kwargs.pop('parallel', False)
read = kwargs.pop('read', True)
out = kwargs.pop('out', None)
broadcast = kwargs.pop('broadcast', False)
assert not kwargs
if parallel:
collective = True
else:
collective = False
dset = self.file[name]
if indices:
selection = HyperslabSelection(indices, dset.shape)
mshape = selection.mshape
else:
selection = 'all'
mshape = dset.shape
if not read:
selection = None
real2complex = False
if out is None:
array = np.ndarray(mshape, dset.dtype, order='C')
elif out.dtype == complex and dset.dtype == float:
# For real to complex reading i.e TDDFT temporary buffer
# is also needed
array = np.ndarray(mshape, dset.dtype, order='C')
real2complex = True
else:
assert type(out) is np.ndarray
# XXX Check the shapes are compatible
assert out.shape == mshape
assert (out.dtype == dset.dtype or
(out.dtype == complex and dset.dtype == float))
array = out
dset.read(array, selection, collective)
if real2complex:
out[:] = array[:]
array = out
if broadcast:
self.comm.broadcast(array, 0)
if array.shape == ():
return array.item()
else:
return array
def get_reference(self, name, indices, length=None):
return HDF5FileReference(self.file, name, indices, length)
def get_parameters(self):
return self.params_grp.attrs
def close(self):
self.file.close()
