def from_numpy_dtype(np_dtype):
"""
:param np_dtype: the NumPy dtype (e.g. np.dtype(np.double))
:return: a dtype type representation
"""
from numba.typesystem import numpy_support
return numpy_dtype(numpy_support.map_dtype(np_dtype))
def get_default_match_table(u):
"""
Get a matcher table: { (type -> bool) : (value -> type) }
"""
table = {
is_NULL:
lambda value: numba.typesystem.null,
is_dtype_constructor:
lambda value: numba.typesystem.from_numpy_dtype(np.dtype(value)),
is_numpy_scalar:
lambda value: numpy_support.map_dtype(value.dtype),
is_ctypes:
lambda value: from_ctypes(value, u),
cffi_support.is_cffi_func:
lambda value: from_cffi(value, u),
is_numba_exttype:
lambda value: getattr(type(value), '__numba_ext_type'),
numbawrapper.is_numba_wrapper:
lambda value: u.jit_function(value),
is_autojit_func:
lambda value: u.autojit_function(value),
is_registered:
lambda value: from_typefunc(value, u),
}
return table
def from_numpy_dtype(np_dtype):
"""
:param np_dtype: the NumPy dtype (e.g. np.dtype(np.double))
:return: a dtype type representation
"""
from numba.typesystem import numpy_support
return numpy_dtype(numpy_support.map_dtype(np_dtype))
def get_default_match_table(u):
"""
Get a matcher table: { (type -> bool) : (value -> type) }
"""
table = {
is_NULL:
lambda value: numba.typesystem.null,
is_dtype_constructor:
lambda value: numba.typesystem.from_numpy_dtype(np.dtype(value)),
is_numpy_scalar:
lambda value: numpy_support.map_dtype(value.dtype),
is_ctypes:
lambda value: from_ctypes(value, u),
cffi_support.is_cffi_func:
lambda value: from_cffi(value, u),
is_numba_exttype:
lambda value: getattr(type(value), '__numba_ext_type'),
numbawrapper.is_numba_wrapper:
lambda value: u.jit_function(value),
is_autojit_func:
lambda value: u.autojit_function(value),
is_registered:
lambda value: from_typefunc(value, u),
}
return table
def promote_int(u, promote, type1, type2):
"""
Promote two integer types. We have to remain portable at this point, e.g.
promote_int(Py_ssize_t, int) should be Py_ssize_t, and not int64 or int32.
"""
t1, t2 = type1.get_dtype(), type2.get_dtype()
return numpy_support.map_dtype(np.result_type(t1, t2))
def numba_type_from_sig(ufunc_signature):
"""
Convert ufunc type signature string (e.g. 'dd->d') to a function
"""
args, ret = ufunc_signature.split('->')
to_numba = lambda c: numpy_support.map_dtype(np.dtype(c))
signature = to_numba(ret)(*map(to_numba, args))
return signature
def type_ndarray(value):
if isinstance(value, np.ndarray):
dtype = numpy_support.map_dtype(value.dtype)
return u.array(dtype, value.ndim)
def get_default_typing_rules(u, typeof, promote):
"""
Get a table mapping Python classes to handlers (value -> type)
:param u: The type universe
"""
from numba.typesystem import numpy_support
table = {}
def register(*classes):
def dec(func):
for cls in classes:
table[cls] = lambda u, value: func(value)
return func
return dec
@register(int, long)
def type_int(value):
if abs(value) < 1:
bits = 0
else:
bits = math.ceil(math.log(abs(value), 2))
if bits < 32:
return u.int_
elif bits < 64:
return u.int64
else:
raise ValueError("Cannot represent %s as int32 or int64", value)
@register(np.ndarray)
def type_ndarray(value):
if isinstance(value, np.ndarray):
dtype = numpy_support.map_dtype(value.dtype)
return u.array(dtype, value.ndim)
#is_c_contig=value.flags['C_CONTIGUOUS'],
#is_f_contig=value.flags['F_CONTIGUOUS'])
@register(tuple, list, dict)
def type_container(value):
assert isinstance(value, (tuple, list, dict))
if isinstance(value, dict):
key_type = type_container(value.keys(), promote, typeof)
value_type = type_container(value.values(), promote, typeof)
return u.dict_(key_type, value_type, size=len(value))
if isinstance(value, tuple):
container_type = u.tuple_
else:
container_type = u.list_
if 0 < len(value) < 30:
# Figure out base type if the container is not too large
# base_type = reduce(promote, (typeof(child) for child in value))
ty = typeof(value[0])
if all(typeof(child) == ty for child in value):
base_type = ty
else:
base_type = u.object_
else:
base_type = u.object_
return container_type(base_type, size=len(value))
register(
np.dtype)(lambda value: u.numpy_dtype(numpy_support.map_dtype(value)))
register(types.ModuleType)(lambda value: u.module(value))
register(itypesystem.Type)(lambda value: u.meta(value))
return table
def type_ndarray(value):
if isinstance(value, np.ndarray):
dtype = numpy_support.map_dtype(value.dtype)
return u.array(dtype, value.ndim)
def get_default_typing_rules(u, typeof, promote):
"""
Get a table mapping Python classes to handlers (value -> type)
:param u: The type universe
"""
table = {}
def register(*classes):
def dec(func):
for cls in classes:
table[cls] = lambda u, value: func(value)
return func
return dec
@register(int, long)
def type_int(value):
if abs(value) < 1:
bits = 0
else:
bits = math.ceil(math.log(abs(value), 2))
if bits < 32:
return u.int_
elif bits < 64:
return u.int64
else:
raise ValueError("Cannot represent %s as int32 or int64", value)
@register(np.ndarray)
def type_ndarray(value):
if isinstance(value, np.ndarray):
dtype = numpy_support.map_dtype(value.dtype)
return u.array(dtype, value.ndim)
#is_c_contig=value.flags['C_CONTIGUOUS'],
#is_f_contig=value.flags['F_CONTIGUOUS'])
@register(tuple, list, dict)
def type_container(value):
assert isinstance(value, (tuple, list, dict))
if isinstance(value, dict):
key_type = type_container(value.keys(), promote, typeof)
value_type = type_container(value.values(), promote, typeof)
return u.dict_(key_type, value_type, size=len(value))
if isinstance(value, tuple):
container_type = u.tuple_
else:
container_type = u.list_
if 0 < len(value) < 30:
# Figure out base type if the container is not too large
# base_type = reduce(promote, (typeof(child) for child in value))
ty = typeof(value[0])
if all(typeof(child) == ty for child in value):
base_type = ty
else:
base_type = u.object_
else:
base_type = u.object_
return container_type(base_type, size=len(value))
register(np.dtype)(lambda value: u.numpy_dtype(numpy_support.map_dtype(value)))
register(types.ModuleType)(lambda value: u.module(value))
register(itypesystem.Type)(lambda value: u.meta(value))
return table
