def reduce(self, context, x, out=None, initial=0.0, queue=None):
if queue is None:
queue = x.queue
if not isinstance(x, cl.DeviceMemoryView):
x = cl.from_host(queue.context, x)
#output, input, shared, group_size, initial=0.0
size = x.size
shared = cl.local_memory(x.ctype, ndim=1, shape=[size])
group_size = size // 2
for item in [2, 4, 8, 16, 32, 64, 128, 256, 512]:
if group_size < item:
group_size = item // 2
break
else:
group_size = 512
if out is None:
out = cl.empty(queue.context, [1], x.format)
kernel = reduce_kernel.compile(queue.context,
function=self.device_func,
output=cl.global_memory(out.ctype,
flat=True),
array=cl.global_memory(x.ctype,
flat=True),
shared=shared,
group_size=cl.cl_uint,
cly_meta=self.device_func.func_name)
max_wgsize = kernel.work_group_size(queue.device)
group_size = min(max_wgsize, group_size)
kernel(queue, out, out.array_info, x, x.array_info, shared,
shared.local_info, group_size)
# reduce_kernel(queue, self.device_func, out, x, shared, group_size)
# reduce_kernel(queue, self.device_func, out, x, shared, group_size)
array = CLArray._view_as_this(out)
array.__array_init__(context, queue)
return array
def reduce(self, context, x, out=None, initial=0.0, queue=None):
if queue is None:
queue = x.queue
if not isinstance(x, cl.DeviceMemoryView):
x = cl.from_host(queue.context, x)
#output, input, shared, group_size, initial=0.0
size = x.size
shared = cl.local_memory(x.ctype, ndim=1, shape=[size])
group_size = size // 2
for item in [2, 4, 8, 16, 32, 64, 128, 256, 512]:
if group_size < item:
group_size = item // 2
break
else:
group_size = 512
if out is None:
out = cl.empty(queue.context, [1], x.format)
kernel = reduce_kernel.compile(queue.context,
function=self.device_func,
output=cl.global_memory(out.ctype, flat=True),
array=cl.global_memory(x.ctype, flat=True),
shared=shared,
group_size=cl.cl_uint,
cly_meta=self.device_func.func_name)
max_wgsize = kernel.work_group_size(queue.device)
group_size = min(max_wgsize, group_size)
kernel(queue, out, out.array_info, x, x.array_info, shared, shared.local_info, group_size)
# reduce_kernel(queue, self.device_func, out, x, shared, group_size)
# reduce_kernel(queue, self.device_func, out, x, shared, group_size)
array = CLArray._view_as_this(out)
array.__array_init__(context, queue)
return array
def reduce(queue, function, input, initial=0.0):
'''
reduce(queue, function, sequence[, initial]) -> value
Apply a function of two arguments cumulatively to the items of a sequence,
from left to right, so as to reduce the sequence to a single value.
For example, reduce(lambda x, y: x+y, [1, 2, 3, 4, 5]) calculates
((((1+2)+3)+4)+5). If initial is present, it is placed before the items
of the sequence in the calculation, and serves as a default when the
sequence is empty.
'''
size = input.size
shared = cl.local_memory(input.format, [size])
output = cl.empty(queue.context, [1], input.format)
group_size = size // 2
cl_reduce(queue, function, output, input, shared, group_size, initial)
return output
def reduce(queue, function, input, initial=0.0):
'''
reduce(queue, function, sequence[, initial]) -> value
Apply a function of two arguments cumulatively to the items of a sequence,
from left to right, so as to reduce the sequence to a single value.
For example, reduce(lambda x, y: x+y, [1, 2, 3, 4, 5]) calculates
((((1+2)+3)+4)+5). If initial is present, it is placed before the items
of the sequence in the calculation, and serves as a default when the
sequence is empty.
'''
size = input.size
shared = cl.local_memory(input.format, [size])
output = cl.empty(queue.context, [1], input.format)
group_size = size // 2
cl_reduce(queue, function, output, input , shared, group_size, initial)
return output
