def scale_const(multiplier):
"""
Returns a scaling transformation with fixed parameter (1 output, 1 input):
``output1 = input1 * <multiplier>``.
"""
dtype = dtypes.detect_type(multiplier)
return Transformation(
inputs=1, outputs=1,
code="${o1.store}(${func.mul(i1.dtype, numpy." + str(dtype) + ", out=o1.dtype)}(" +
"${i1.load}, " + dtypes.c_constant(multiplier, dtype=dtype) + "));")
def scale_const(multiplier):
"""
Returns a scaling transformation with fixed parameter (1 output, 1 input):
``output1 = input1 * <multiplier>``.
"""
dtype = dtypes.detect_type(multiplier)
return Transformation(inputs=1,
outputs=1,
code="${o1.store}(${func.mul(i1.dtype, numpy." +
str(dtype) + ", out=o1.dtype)}(" + "${i1.load}, " +
dtypes.c_constant(multiplier, dtype=dtype) + "));")
def prepare_for(self, max_local_size):
local_size, workgroups_num, kwds = self._generate(max_local_size)
kwds.update(dict(
min_mem_coalesce_width=self._device_params.min_mem_coalesce_width[self._basis.dtype.itemsize],
local_mem_banks=self._device_params.local_mem_banks,
get_padding=get_padding,
normalize=self._normalize,
wrap_const=lambda x: dtypes.c_constant(x, dtypes.real_for(self._basis.dtype)),
min_blocks=min_blocks,
takes_kweights=(self.kweights is not None),
pad_in=(self._fft_size != self._fft_size_real and self._pass_num == 0
and not self._reverse_direction),
unpad_out=(self._fft_size != self._fft_size_real and self._last_pass
and self._reverse_direction),
reverse_direction=self._reverse_direction))
local_size = local_size
global_size = local_size * workgroups_num
return global_size, local_size, kwds
def prepare_for(self, max_local_size):
local_size, workgroups_num, kwds = self._generate(max_local_size)
kwds.update(
dict(min_mem_coalesce_width=self._device_params.
min_mem_coalesce_width[self._basis.dtype.itemsize],
local_mem_banks=self._device_params.local_mem_banks,
get_padding=get_padding,
normalize=self._normalize,
wrap_const=lambda x: dtypes.c_constant(
x, dtypes.real_for(self._basis.dtype)),
min_blocks=min_blocks,
takes_kweights=(self.kweights is not None),
pad_in=(self._fft_size != self._fft_size_real
and self._pass_num == 0
and not self._reverse_direction),
unpad_out=(self._fft_size != self._fft_size_real
and self._last_pass and self._reverse_direction),
reverse_direction=self._reverse_direction))
local_size = local_size
global_size = local_size * workgroups_num
return global_size, local_size, kwds
