def __init__(self, size=128, ctype=ctypes.c_double, from_cptr=None):
# __new__ creates or sets _inst_ptr, so always supply that to super's
# `from_cptr`'
VitalObject.__init__(self, self._inst_ptr)
def __new__(cls,
rows=2,
cols=1,
dynamic_rows=False,
dynamic_cols=False,
dtype=numpy.double,
from_cptr=None,
owns_data=True):
"""
Create a new Vital Eigen matrix and interface
:param rows: Number of rows in the matrix
:param cols: Number of columns in the matrix
:param dynamic_rows: If we should not use compile-time generated types
in regards to the row specification.
:param dynamic_cols: If we should not use compile-time generated types
in regards to the column specification.
:param dtype: numpy dtype to use
:param from_cptr: Optional existing C Eigen matrix instance pointer to
use instead of constructing a new one.
:param owns_data: When given a c-pointer, if we should take ownership of
the underlying data.
:return: Interface to a new or existing Eigen matrix instance.
:raises ValueError: If a `c_ptr` is provided but is of a different
compile-time shape type. E.g. cannot create a 2x1 matrix with
dynamic rows from a 2x1 matrix
:raises VitalInvalidStaticEigenShape: An invalid (row, column) was
specified without stating dynamic rows or columns. This is because
Vital and Eigen defines only so many shapes at compile time.
"""
dtype = numpy.dtype(dtype)
func_spec, func_map, c_type = \
cls._init_func_map(rows, cols, dynamic_rows, dynamic_cols, dtype)
op_c_type, op_c_type_ptr = cls.TYPE_CACHE.get_types(func_spec)
# Create new Eigen matrix
# Check if the shape given is either fully dynamic or is a valid
# compile-time shape
try:
cls.VITAL_LIB[func_map['new_sized']]
except AttributeError:
raise VitalInvalidStaticEigenShape(
"Array shape %s is not a valid compile-time specified "
"shape given that dynamic rows/cols were specified as %s." %
((rows, cols), (bool(dynamic_rows), bool(dynamic_cols))))
if from_cptr is None:
c_new = cls.VITAL_LIB[func_map['new_sized']]
c_new.argtypes = [c_ptrdiff_t, c_ptrdiff_t]
c_new.restype = op_c_type_ptr
inst_ptr = c_new(c_ptrdiff_t(rows), c_ptrdiff_t(cols))
if not bool(inst_ptr):
raise RuntimeError("Failed to construct new Eigen matrix")
owns_data = True
else:
if not isinstance(from_cptr, op_c_type_ptr):
raise ValueError("Given C-Pointer is not correct for the "
"shape-type '%s' (given: %s)" %
(func_spec, type(from_cptr)))
inst_ptr = from_cptr
# Get information, data pointer and base transformed array
rows, cols, row_stride, col_stride, data = \
cls._get_data_components(inst_ptr, op_c_type_ptr, c_type, func_map)
# Might have to swap out the use of ``dtype_bytes`` for
# inner/outer size values from Eigen if matrices are ever NOT
# densely packed.
dtype_bytes = numpy.dtype(c_type).alignment
strides = (row_stride * dtype_bytes, col_stride * dtype_bytes)
if data:
# data has no __array_interface__ yet, thus...
b = numpy.ctypeslib.as_array(data, (rows * cols, ))
else:
b = buffer('')
# args: (subclass, shape, dtype, buffer, offset, strides, order)
obj = numpy.ndarray.__new__(cls, (rows, cols), dtype, b, 0, strides)
# local properties
# instance-specific opaque type
obj.C_TYPE = op_c_type
obj.C_TYPE_PTR = op_c_type_ptr
obj._dynamic_rows = dynamic_rows
obj._dynamic_cols = dynamic_cols
obj._func_map = func_map
obj._c_type = c_type
obj._owns_data = owns_data
# Always going to have an instance pointer at this point due to above
# logic
VitalObject.__init__(obj, from_cptr=inst_ptr)
return obj
def __new__(cls, rows=2, cols=1, dynamic_rows=False, dynamic_cols=False,
dtype=numpy.double, from_cptr=None, owns_data=True):
"""
Create a new Vital Eigen matrix and interface
:param rows: Number of rows in the matrix
:param cols: Number of columns in the matrix
:param dynamic_rows: If we should not use compile-time generated types
in regards to the row specification.
:param dynamic_cols: If we should not use compile-time generated types
in regards to the column specification.
:param dtype: numpy dtype to use
:param from_cptr: Optional existing C Eigen matrix instance pointer to
use instead of constructing a new one.
:param owns_data: When given a c-pointer, if we should take ownership of
the underlying data.
:return: Interface to a new or existing Eigen matrix instance.
:raises ValueError: If a `c_ptr` is provided but is of a different
compile-time shape type. E.g. cannot create a 2x1 matrix with
dynamic rows from a 2x1 matrix
:raises VitalInvalidStaticEigenShape: An invalid (row, column) was
specified without stating dynamic rows or columns. This is because
Vital and Eigen defines only so many shapes at compile time.
"""
dtype = numpy.dtype(dtype)
func_spec, func_map, c_type = \
cls._init_func_map(rows, cols, dynamic_rows, dynamic_cols, dtype)
op_c_type, op_c_type_ptr = cls.TYPE_CACHE.get_types(func_spec)
# Create new Eigen matrix
# Check if the shape given is either fully dynamic or is a valid
# compile-time shape
try:
cls.VITAL_LIB[func_map['new_sized']]
except AttributeError:
raise VitalInvalidStaticEigenShape(
"Array shape %s is not a valid compile-time specified "
"shape given that dynamic rows/cols were specified as %s."
% ((rows, cols), (bool(dynamic_rows), bool(dynamic_cols)))
)
if from_cptr is None:
c_new = cls.VITAL_LIB[func_map['new_sized']]
c_new.argtypes = [c_ptrdiff_t, c_ptrdiff_t]
c_new.restype = op_c_type_ptr
inst_ptr = c_new(c_ptrdiff_t(rows), c_ptrdiff_t(cols))
if not bool(inst_ptr):
raise RuntimeError("Failed to construct new Eigen matrix")
owns_data = True
else:
if not isinstance(from_cptr, op_c_type_ptr):
raise ValueError("Given C-Pointer is not correct for the "
"shape-type '%s' (given: %s)"
% (func_spec, type(from_cptr)))
inst_ptr = from_cptr
# Get information, data pointer and base transformed array
rows, cols, row_stride, col_stride, data = \
cls._get_data_components(inst_ptr, op_c_type_ptr, c_type, func_map)
# Might have to swap out the use of ``dtype_bytes`` for
# inner/outer size values from Eigen if matrices are ever NOT
# densely packed.
dtype_bytes = numpy.dtype(c_type).alignment
strides = (row_stride * dtype_bytes,
col_stride * dtype_bytes)
if data:
# data has no __array_interface__ yet, thus...
b = numpy.ctypeslib.as_array(data, (rows * cols,))
else:
b = buffer('')
# args: (subclass, shape, dtype, buffer, offset, strides, order)
obj = numpy.ndarray.__new__(cls, (rows, cols), dtype, b, 0, strides)
# local properties
# instance-specific opaque type
obj.C_TYPE = op_c_type
obj.C_TYPE_PTR = op_c_type_ptr
obj._dynamic_rows = dynamic_rows
obj._dynamic_cols = dynamic_cols
obj._func_map = func_map
obj._c_type = c_type
obj._owns_data = owns_data
# Always going to have an instance pointer at this point due to above
# logic
VitalObject.__init__(obj, from_cptr=inst_ptr)
return obj
def __init__(self, size=128, ctype=ctypes.c_double, from_cptr=None):
# __new__ creates or sets _inst_ptr, so always supply that to super's
# `from_cptr`'
VitalObject.__init__(self, self._inst_ptr)
