def setUp(self):
if self.stream == 'null':
self.stream = cupy.cuda.Stream.null
elif self.stream == 'new':
self.stream = cupy.cuda.Stream()
self.my_sum = core.ReductionKernel('T x', 'T out', 'x', 'a + b',
'out = a', '0', 'my_sum')
def test_invalid_kernel_name(self):
with six.assertRaisesRegex(self, ValueError, 'Invalid kernel name'):
core.ReductionKernel('T x',
'T y',
'x',
'a + b',
'y = a',
'0',
name='1')
def getnnz(self, axis=None):
"""Returns the number of stored values, including explicit zeros.
Args:
axis: Not supported yet.
Returns:
int: The number of stored values.
"""
if axis is not None:
raise NotImplementedError(
'getnnz over an axis is not implemented for DIA format')
m, n = self.shape
nnz = core.ReductionKernel(
'int32 offsets, int32 m, int32 n', 'int32 nnz',
'offsets > 0 ? min(m, n - offsets) : min(m + offsets, n)', 'a + b',
'nnz = a', '0', 'dia_nnz')(self.offsets, m, n)
return int(nnz)
import cupy
import cupyx.scipy.fft
from cupy import core
from cupy.core import _routines_math as _math
from cupy.core import fusion
from cupy.lib import stride_tricks
_dot_kernel = core.ReductionKernel('T x1, T x2', 'T y', 'x1 * x2', 'a + b',
'y = a', '0', 'dot_product')
def _choose_conv_method(in1, in2, mode):
if in1.ndim != 1 or in2.ndim != 1:
raise NotImplementedError('Only 1d inputs are supported currently')
if in1.dtype.kind in 'bui' or in2.dtype.kind in 'bui':
return 'direct'
if _fftconv_faster(in1, in2, mode):
return 'fft'
return 'direct'
def _fftconv_faster(x, h, mode):
"""
.. seealso:: :func: `scipy.signal.signaltools._fftconv_faster`
"""
# TODO(Dahlia-Chehata): replace with GPU-based constants.
def setUp(self):
self.my_sum = core.ReductionKernel('T x', 'T out', 'x', 'a + b',
'out = a', '0', 'my_sum')
# TODO(okuta): Implement delete
# TODO(okuta): Implement insert
# TODO(okuta): Implement append
# TODO(okuta): Implement resize
_first_nonzero_krnl = core.ReductionKernel(
'T data, int64 len',
'int64 y',
'data == T(0) ? len : _j',
'min(a, b)',
'y = a',
'len',
'first_nonzero'
)
def trim_zeros(filt, trim='fb'):
"""Trim the leading and/or trailing zeros from a 1-D array or sequence.
Returns the trimmed array
Args:
filt(cupy.ndarray): Input array
trim(str, optional):
'fb' default option trims the array from both sides.
