def _init_gaussian(self, sigma):
"""Create a buffer of the right size according to the width of the gaussian ...
:param sigma: width of the gaussian, the length of the function will be 8*sigma + 1
Same calculation done on CPU
x = numpy.arange(size) - (size - 1.0) / 2.0
gaussian = numpy.exp(-(x / sigma) ** 2 / 2.0).astype(numpy.float32)
gaussian /= gaussian.sum(dtype=numpy.float32)
"""
pyopencl.enqueue_barrier(self.queue).wait()
name = "gaussian_%s" % sigma
size = kernel_size(sigma, True)
wg_size = nextpower(size)
logger.info("Allocating %s float for blur sigma: %s. wg=%s max_wg=%s",
size, sigma, wg_size, self.block_size)
wg1 = self.kernels_wg["gaussian"]
if wg1 >= wg_size:
gaussian_gpu = pyopencl.array.empty(self.queue,
size,
dtype=numpy.float32)
pyopencl.enqueue_barrier(self.queue).wait()
kernel = self.kernels.get_kernel("gaussian")
shm1 = pyopencl.LocalMemory(4 * wg_size)
shm2 = pyopencl.LocalMemory(4 * wg_size)
evt = kernel(
self.queue,
(wg_size, ),
(wg_size, ),
gaussian_gpu.data,
numpy.float32(sigma), # const float sigma,
numpy.int32(size), # const int SIZE
shm1,
shm2) # some shared memory
pyopencl.enqueue_barrier(self.queue).wait()
if self.profile:
self.events.append(("gaussian %s" % sigma, evt))
else:
logger.info(
"Workgroup size error: gaussian wg: %s < max_work_group_size: %s",
wg1, self.block_size)
# common bug on OSX when running on CPU
x = numpy.arange(size) - (size - 1.0) / 2.0
gaus = numpy.exp(-(x / sigma)**2 / 2.0).astype(numpy.float32)
gaus /= gaus.sum(dtype=numpy.float32)
gaussian_gpu = pyopencl.array.to_device(self.queue, gaus)
self.cl_mem[name] = gaussian_gpu
return gaussian_gpu
def _init_gaussian(self, sigma):
"""Create a buffer of the right size according to the width of the gaussian ...
:param sigma: width of the gaussian, the length of the function will be 8*sigma + 1
Same calculation done on CPU
x = numpy.arange(size) - (size - 1.0) / 2.0
gaussian = numpy.exp(-(x / sigma) ** 2 / 2.0).astype(numpy.float32)
gaussian /= gaussian.sum(dtype=numpy.float32)
"""
pyopencl.enqueue_barrier(self.queue).wait()
name = "gaussian_%s" % sigma
size = kernel_size(sigma, True)
wg_size = nextpower(size)
logger.info("Allocating %s float for blur sigma: %s. wg=%s max_wg=%s", size, sigma, wg_size, self.block_size)
wg1 = self.kernels_wg["gaussian"]
if wg1 >= wg_size:
gaussian_gpu = pyopencl.array.empty(self.queue, size, dtype=numpy.float32)
pyopencl.enqueue_barrier(self.queue).wait()
kernel = self.kernels.get_kernel("gaussian")
shm1 = pyopencl.LocalMemory(4 * wg_size)
shm2 = pyopencl.LocalMemory(4 * wg_size)
evt = kernel(self.queue, (wg_size,), (wg_size,),
gaussian_gpu.data,
numpy.float32(sigma), # const float sigma,
numpy.int32(size), # const int SIZE
shm1, shm2) # some shared memory
pyopencl.enqueue_barrier(self.queue).wait()
if self.profile:
self.events.append(("gaussian %s" % sigma, evt))
else:
logger.info("Workgroup size error: gaussian wg: %s < max_work_group_size: %s",
wg1, self.block_size)
# common bug on OSX when running on CPU
x = numpy.arange(size) - (size - 1.0) / 2.0
gaus = numpy.exp(-(x / sigma) ** 2 / 2.0).astype(numpy.float32)
gaus /= gaus.sum(dtype=numpy.float32)
gaussian_gpu = pyopencl.array.to_device(self.queue, gaus)
self.cl_mem[name] = gaussian_gpu
return gaussian_gpu