def _call_kernel(kernel, input, weights, output, weight_dtype=cupy.float64):
"""
Calls a constructed ElementwiseKernel. The kernel must take an input image,
an array of weights, and an output array.
The weights are the only optional part and can be passed as None and then
one less argument is passed to the kernel. If the output is given as None
then it will be allocated in this function.
This function deals with making sure that the weights are contiguous and
float64 or bool*, that the output is allocated and appriopate shaped. This
also deals with the situation that the input and output arrays overlap in
memory.
* weights is always casted to float64 or bool in order to get an output
compatible with SciPy, though float32 might be sufficient when input dtype
is low precision.
"""
if weights is not None:
weights = cupy.ascontiguousarray(weights, weight_dtype)
needs_temp = cupy.shares_memory(output, input, "MAY_SHARE_BOUNDS")
if needs_temp:
output, temp = (
_util._get_output(output.dtype, input, None, weight_dtype),
output,
)
if weights is None:
kernel(input, output)
else:
kernel(input, weights, output)
if needs_temp:
temp[...] = output[...]
output = temp
return output
def test_array_gpu(self):
import cupy as cp
mv = MedicalVolume(np.ones((10, 20, 30)), self._AFFINE)
mv_gpu = mv.to(Device(0))
data = cp.asarray(mv_gpu)
assert cp.shares_memory(data, mv_gpu.volume)
def _call_kernel(
kernel,
input,
weights,
output,
structure=None,
weights_dtype=numpy.float64,
structure_dtype=numpy.float64,
):
"""
Calls a constructed ElementwiseKernel. The kernel must take an input image,
an optional array of weights, an optional array for the structure, and an
output array.
weights and structure can be given as None (structure defaults to None) in
which case they are not passed to the kernel at all. If the output is given
as None then it will be allocated in this function.
This function deals with making sure that the weights and structure are
contiguous and float64 (or bool for weights that are footprints)*, that the
output is allocated and appriopately shaped. This also deals with the
situation that the input and output arrays overlap in memory.
* weights is always cast to float64 or bool in order to get an output
compatible with SciPy, though float32 might be sufficient when input dtype
is low precision. If weights_dtype is passed as weights.dtype then no
dtype conversion will occur. The input and output are never converted.
"""
args = [input]
if weights is not None:
weights = cupy.ascontiguousarray(weights, weights_dtype)
args.append(weights)
if structure is not None:
structure = cupy.ascontiguousarray(structure, structure_dtype)
args.append(structure)
output = _util._get_output(output, input, None, cupy.dtype(weights_dtype))
needs_temp = cupy.shares_memory(output, input, "MAY_SHARE_BOUNDS")
if needs_temp:
output, temp = _util._get_output(output.dtype, input), output
args.append(output)
kernel(*args)
if needs_temp:
temp[...] = output[...]
output = temp
return output
def _get_spline_output(input, output, allow_float32=False):
"""Create workspace array, temp, and the final dtype for the output.
If allow_float32 is False, temp will always have float64 dtype.
If allow_float32 is True, temp will have float32 dtype when ``input`` or
``output`` is single precision.
"""
complex_data = input.dtype.kind == "c"
if complex_data:
min_float_dtype = cupy.complex64 if allow_float32 else cupy.complex128
else:
min_float_dtype = cupy.float32 if allow_float32 else cupy.float64
if isinstance(output, cupy.ndarray):
if complex_data and output.dtype.kind != "c":
raise ValueError(
"output must have complex dtype for complex inputs")
float_dtype = cupy.promote_types(output.dtype, min_float_dtype)
output_dtype = output.dtype
else:
if output is None:
output = output_dtype = input.dtype
else:
output_dtype = cupy.dtype(output)
float_dtype = cupy.promote_types(output, min_float_dtype)
if (isinstance(output, cupy.ndarray)
and output.dtype == float_dtype == output_dtype
and output.flags.c_contiguous):
if output is not input:
output[...] = input[...]
temp = output
else:
temp = input.astype(float_dtype, copy=False)
temp = cupy.ascontiguousarray(temp)
if cupy.shares_memory(temp, input, 'MAY_SHARE_BOUNDS'):
temp = temp.copy()
return temp, float_dtype, output_dtype
def _get_spline_output(input, output):
"""Create workspace array, temp, and the final dtype for the output.
Differs from SciPy by not always forcing the internal floating point dtype
to be double precision.
"""
complex_data = input.dtype.kind == 'c'
if complex_data:
min_float_dtype = cupy.complex64
else:
min_float_dtype = cupy.float32
if isinstance(output, cupy.ndarray):
if complex_data and output.dtype.kind != 'c':
raise ValueError(
'output must have complex dtype for complex inputs'
)
float_dtype = cupy.promote_types(output.dtype, min_float_dtype)
output_dtype = output.dtype
else:
if output is None:
output = output_dtype = input.dtype
else:
output_dtype = cupy.dtype(output)
float_dtype = cupy.promote_types(output, min_float_dtype)
if (isinstance(output, cupy.ndarray)
and output.dtype == float_dtype == output_dtype
and output.flags.c_contiguous):
if output is not input:
output[...] = input[...]
temp = output
else:
temp = input.astype(float_dtype, copy=False)
temp = cupy.ascontiguousarray(temp)
if cupy.shares_memory(temp, input, 'MAY_SHARE_BOUNDS'):
temp = temp.copy()
return temp, float_dtype, output_dtype
def _binary_erosion(input,
structure,
iterations,
mask,
output,
border_value,
origin,
invert,
brute_force=True):
try:
iterations = operator.index(iterations)
except TypeError:
raise TypeError('iterations parameter should be an integer')
if input.dtype.kind == 'c':
raise TypeError('Complex type not supported')
if structure is None:
structure = generate_binary_structure(input.ndim, 1)
all_weights_nonzero = input.ndim == 1
center_is_true = True
default_structure = True
else:
structure = structure.astype(dtype=bool, copy=False)
# transfer to CPU for use in determining if it is fully dense
# structure_cpu = cupy.asnumpy(structure)
default_structure = False
if structure.ndim != input.ndim:
raise RuntimeError('structure and input must have same dimensionality')
if not structure.flags.c_contiguous:
structure = cupy.ascontiguousarray(structure)
if structure.size < 1:
raise RuntimeError('structure must not be empty')
if mask is not None:
if mask.shape != input.shape:
raise RuntimeError('mask and input must have equal sizes')
if not mask.flags.c_contiguous:
mask = cupy.ascontiguousarray(mask)
masked = True
else:
masked = False
origin = _util._fix_sequence_arg(origin, input.ndim, 'origin', int)
if isinstance(output, cupy.ndarray):
if output.dtype.kind == 'c':
raise TypeError('Complex output type not supported')
else:
output = bool
output = _util._get_output(output, input)
temp_needed = cupy.shares_memory(output, input, 'MAY_SHARE_BOUNDS')
if temp_needed:
# input and output arrays cannot share memory
temp = output
output = _util._get_output(output.dtype, input)
if structure.ndim == 0:
# kernel doesn't handle ndim=0, so special case it here
if float(structure):
output[...] = cupy.asarray(input, dtype=bool)
else:
output[...] = ~cupy.asarray(input, dtype=bool)
return output
origin = tuple(origin)
int_type = _util._get_inttype(input)
offsets = _filters_core._origins_to_offsets(origin, structure.shape)
if not default_structure:
# synchronize required to determine if all weights are non-zero
nnz = int(cupy.count_nonzero(structure))
all_weights_nonzero = nnz == structure.size
if all_weights_nonzero:
center_is_true = True
else:
center_is_true = _center_is_true(structure, origin)
erode_kernel = _get_binary_erosion_kernel(
structure.shape,
int_type,
offsets,
center_is_true,
border_value,
invert,
masked,
all_weights_nonzero,
)
if iterations == 1:
if masked:
output = erode_kernel(input, structure, mask, output)
else:
output = erode_kernel(input, structure, output)
elif center_is_true and not brute_force:
raise NotImplementedError(
'only brute_force iteration has been implemented')
else:
if cupy.shares_memory(output, input, 'MAY_SHARE_BOUNDS'):
raise ValueError('output and input may not overlap in memory')
tmp_in = cupy.empty_like(input, dtype=output.dtype)
tmp_out = output
if iterations >= 1 and not iterations & 1:
tmp_in, tmp_out = tmp_out, tmp_in
if masked:
tmp_out = erode_kernel(input, structure, mask, tmp_out)
else:
tmp_out = erode_kernel(input, structure, tmp_out)
# TODO: kernel doesn't return the changed status, so determine it here
changed = not (input == tmp_out).all() # synchronize!
ii = 1
while ii < iterations or ((iterations < 1) and changed):
tmp_in, tmp_out = tmp_out, tmp_in
if masked:
tmp_out = erode_kernel(tmp_in, structure, mask, tmp_out)
else:
tmp_out = erode_kernel(tmp_in, structure, tmp_out)
changed = not (tmp_in == tmp_out).all()
ii += 1
if not changed and (not ii & 1): # synchronize!
# can exit early if nothing changed
# (only do this after even number of tmp_in/out swaps)
break
output = tmp_out
if temp_needed:
temp[...] = output
output = temp
return output
