def sum(a, dtype=None, queue=None):
"""
.. versionadded:: 2011.1
"""
from pyopencl.reduction import get_sum_kernel
krnl = get_sum_kernel(a.context, dtype, a.dtype)
return krnl(a, queue=queue)
def sum(a, dtype=None, queue=None):
"""
.. versionadded:: 2011.1
"""
from pyopencl.reduction import get_sum_kernel
krnl = get_sum_kernel(a.context, dtype, a.dtype)
return krnl(a, queue=queue)
def init_OpenCL_decoding(self,msg_at_time_, context_=False):
if not context_:
self.context = cl.create_some_context()
else:
self.context = context_
print(self.context.get_info(cl.context_info.DEVICES))
path = os.path.split(os.path.abspath(__file__))
kernelsource = open(os.path.join(path[0], "kernels_min_and_BP.cl")).read()
tpl = Template(kernelsource)
rendered_tp = tpl.render(cn_degree=self.d_c_max, vn_degree=self.d_v_max, msg_at_time=msg_at_time_)
self.program = cl.Program(self.context, str(rendered_tp)).build()
self.queue = cl.CommandQueue(self.context)
self.inbox_memory_start_varnodes_buffer = cl_array.to_device(self.queue,
self.inbox_memory_start_varnodes.astype(np.int32))
self.inbox_memory_start_checknodes_buffer = cl_array.to_device(self.queue,
self.inbox_memory_start_checknodes.astype(np.int32))
self.degree_varnode_nr_buffer = cl_array.to_device(self.queue, self.degree_varnode_nr.astype(np.int32))
self.degree_checknode_nr_buffer = cl_array.to_device(self.queue, self.degree_checknode_nr.astype(np.int32))
self.target_memorycells_varnodes_buffer = cl_array.to_device(self.queue,
self.target_memory_cells_varnodes.astype(np.int32))
self.target_memorycells_checknodes_buffer = cl_array.to_device(self.queue,
self.target_memory_cells_checknodes.astype(np.int32))
self.checknode_inbox_buffer = cl_array.empty(self.queue, self.inbox_memory_checknodes.shape, dtype=np.float64)
self.varnode_inbox_buffer = cl_array.empty(self.queue, self.inbox_memory_varnodes.shape, dtype=np.float64)
self.syndrom_buffer = cl_array.empty(self.queue,
(self.degree_checknode_nr.shape[0], self.inbox_memory_varnodes.shape[-1]), dtype=np.int32)
self.krnl = get_sum_kernel(self.context, None,
dtype_in=self.varnode_inbox_buffer.dtype) # varnode_output_buffer.dtype )
# define programs
self.send_prog = self.program.send_channel_values_to_checknode_inbox
self.varnode_update_prog = self.program.varnode_update
self.checknode_update_prog = self.program.checknode_update
self.calc_syndrom_prog = self.program.calc_syndrome
self.varoutput_prog = self.program.calc_varnode_output
def sum(a, dtype=None, queue=None):
from pyopencl.reduction import get_sum_kernel
krnl = get_sum_kernel(dtype, a.dtype)
return krnl(a, queue=queue)
def sum(a, dtype=None, queue=None):
from pyopencl.reduction import get_sum_kernel
krnl = get_sum_kernel(a.context, dtype, a.dtype)
return krnl(a, queue=queue)
def init_OpenCL_decoding(self, msg_at_time_, context_=False):
if not context_:
self.context = cl.create_some_context()
else:
self.context = context_
path = os.path.split(os.path.abspath(__file__))
kernelsource = open(os.path.join(path[0],
"kernels_template.cl")).read()
tpl = Template(kernelsource)
rendered_tp = tpl.render(cn_degree=self.degree_checknode_nr[0],
vn_degree=self.degree_varnode_nr[0],
msg_at_time=msg_at_time_)
self.program = cl.Program(self.context, str(rendered_tp)).build()
self.queue = cl.CommandQueue(self.context)
mem_pool = cl.tools.MemoryPool(
cl.tools.ImmediateAllocator(self.queue, cl.mem_flags.READ_ONLY))
mem_pool2 = cl.tools.MemoryPool(cl.tools.ImmediateAllocator(
self.queue))
#mem_pool = None
self.inbox_memory_start_varnodes_buffer = cl_array.to_device(
self.queue,
self.inbox_memory_start_varnodes.astype(np.int32),
allocator=mem_pool)
self.inbox_memory_start_checknodes_buffer = cl_array.to_device(
self.queue,
self.inbox_memory_start_checknodes.astype(np.int32),
allocator=mem_pool)
self.degree_varnode_nr_buffer = cl_array.to_device(
self.queue,
self.degree_varnode_nr.astype(np.int32),
allocator=mem_pool)
self.degree_checknode_nr_buffer = cl_array.to_device(
self.queue,
self.degree_checknode_nr.astype(np.int32),
allocator=mem_pool)
self.target_memorycells_varnodes_buffer = cl_array.to_device(
self.queue,
self.target_memory_cells_varnodes.astype(np.int32),
allocator=mem_pool)
self.target_memorycells_checknodes_buffer = cl_array.to_device(
self.queue,
self.target_memory_cells_checknodes.astype(np.int32),
allocator=mem_pool)
self.Trellis_checknode_vector_a_buffer = cl_array.to_device(
self.queue,
self.Trellis_checknode_vector_a.astype(np.int32),
allocator=mem_pool)
self.Trellis_varnode_vector_a_buffer = cl_array.to_device(
self.queue,
self.Trellis_varnode_vector_a.astype(np.int32),
allocator=mem_pool)
self.checknode_inbox_buffer = cl_array.empty(
self.queue,
self.inbox_memory_checknodes.shape,
dtype=np.int32,
allocator=mem_pool2)
self.varnode_inbox_buffer = cl_array.empty(
self.queue,
self.inbox_memory_varnodes.shape,
dtype=np.int32,
allocator=mem_pool2)
self.syndrom_buffer = cl_array.empty(
self.queue, (self.degree_checknode_nr.shape[0],
self.inbox_memory_varnodes.shape[-1]),
dtype=np.int32,
allocator=mem_pool2)
self.krnl = get_sum_kernel(self.context,
None,
dtype_in=self.varnode_inbox_buffer.dtype
) # varnode_output_buffer.dtype )
# define programs
self.send_prog = self.program.send_channel_values_to_checknode_inbox
self.first_iter_prog = self.program.checknode_update_iter0
self.first_iter_prog.set_scalar_arg_dtypes(
[None, None, None, None, None, np.int32, np.int32, None])
self.varnode_update_prog = self.program.varnode_update
self.varnode_update_prog.set_scalar_arg_dtypes([
None, None, None, None, None, None, np.int32, np.int32, np.int32,
None
])
self.checknode_update_prog = self.program.checknode_update
self.checknode_update_prog.set_scalar_arg_dtypes(
[None, None, None, None, None, np.int32, np.int32, np.int32, None])
self.calc_syndrom_prog = self.program.calc_syndrome
self.calc_syndrom_prog.set_scalar_arg_dtypes(
[None, None, None, np.int32, None])
self.varoutput_prog = self.program.calc_varnode_output
self.varoutput_prog.set_scalar_arg_dtypes(
[None, None, None, None, np.int32, np.int32, np.int32, None, None])
