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 __init__(self, name, comm=None):
self.comm = comm # for possible future use
self.dims = {}
try:
if self.comm.rank == 0:
if os.path.isfile(name):
os.rename(name, name[:-5] + '.old'+name[-5:])
self.comm.barrier()
except AttributeError:
if os.path.isfile(name):
os.rename(name, name[:-5] + '.old'+name[-5:])
if self.comm.size > 1:
comm = self.comm.get_c_object()
else:
comm = None
self.file = File(name, 'w', comm)
self.dims_grp = self.file.create_group("Dimensions")
self.params_grp = self.file.create_group("Parameters")
self.file.attrs['title'] = 'gpaw_io version="0.1"'
self.hdf5 = True
class Writer:
def __init__(self, name, comm=None):
self.comm = comm # for possible future use
self.dims = {}
try:
if self.comm.rank == 0:
if os.path.isfile(name):
os.rename(name, name[:-5] + '.old'+name[-5:])
self.comm.barrier()
except AttributeError:
if os.path.isfile(name):
os.rename(name, name[:-5] + '.old'+name[-5:])
if self.comm.size > 1:
comm = self.comm.get_c_object()
else:
comm = None
self.file = File(name, 'w', comm)
self.dims_grp = self.file.create_group("Dimensions")
self.params_grp = self.file.create_group("Parameters")
self.file.attrs['title'] = 'gpaw_io version="0.1"'
self.hdf5 = True
def dimension(self, name, value):
if name in self.dims.keys() and self.dims[name] != value:
raise Warning('Dimension %s changed from %s to %s' % \
(name, self.dims[name], value))
self.dims[name] = value
self.dims_grp.attrs[name] = value
def __setitem__(self, name, value):
# attributes must be written collectively
self.params_grp.attrs[name] = value
def add(self, name, shape, array=None, dtype=None,
parallel=False, write=True):
if array is not None:
array = np.asarray(array)
# self.dtype, type, itemsize = self.get_data_type(array, dtype)
if dtype is None:
self.dtype = array.dtype
else:
self.dtype = dtype
shape = [self.dims[dim] for dim in shape]
if not shape:
shape = [1,]
self.dset = self.file.create_dataset(name, shape, self.dtype)
if array is not None:
self.fill(array, parallel=parallel, write=write)
def fill(self, array, *indices, **kwargs):
parallel = kwargs.pop('parallel', False)
write = kwargs.pop('write', True)
if parallel:
collective = True
else:
collective = False
if not write:
selection = None
elif indices:
selection = HyperslabSelection(indices, self.dset.shape)
else:
selection = 'all'
self.dset.write(array, selection, collective)
def get_data_type(self, array=None, dtype=None):
if dtype is None:
dtype = array.dtype
if dtype in [int, bool]:
dtype = np.int32
dtype = np.dtype(dtype)
type = {np.int32: 'int',
np.float64: 'float',
np.complex128: 'complex'}[dtype.type]
return dtype, type, dtype.itemsize
def append(self, name):
raise NotImplementedError('Append with HDF5 not available.')
def close(self):
mtime = int(time.time())
self.file.attrs['mtime'] = mtime
self.dims_grp.close()
self.params_grp.close()
self.dset.close()
self.file.close()
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()
