def load_symbol(name):
mem, sz = cufunc.module.get_global_symbol("%s__%s__" %
(cufunc.name,
name))
val = ctypes.c_int()
driver.device_to_host(ctypes.addressof(val), mem, sz)
return val.value
def load_symbol(name):
mem, sz = cufunc.module.get_global_symbol("%s__%s__" %
(cufunc.name,
name))
val = ctypes.c_int()
driver.device_to_host(ctypes.addressof(val), mem, sz)
return val.value
def _kernel_call(self, args, griddim, blockdim, stream=0, sharedmem=0):
# Prepare kernel
cufunc = self._func.get()
if self.debug:
excname = cufunc.name + "__errcode__"
excmem, excsz = cufunc.module.get_global_symbol(excname)
assert excsz == ctypes.sizeof(ctypes.c_int)
excval = ctypes.c_int()
excmem.memset(0, stream=stream)
# Prepare arguments
retr = [] # hold functors for writeback
kernelargs = []
for t, v in zip(self.argument_types, args):
self._prepare_args(t, v, stream, retr, kernelargs)
# Configure kernel
cu_func = cufunc.configure(griddim,
blockdim,
stream=stream,
sharedmem=sharedmem)
# Invoke kernel
cu_func(*kernelargs)
if self.debug:
driver.device_to_host(ctypes.addressof(excval), excmem, excsz)
if excval.value != 0:
# An error occurred
def load_symbol(name):
mem, sz = cufunc.module.get_global_symbol(
"%s__%s__" % (cufunc.name, name))
val = ctypes.c_int()
driver.device_to_host(ctypes.addressof(val), mem, sz)
return val.value
tid = [load_symbol("tid" + i) for i in 'zyx']
ctaid = [load_symbol("ctaid" + i) for i in 'zyx']
code = excval.value
exccls, exc_args = self.call_helper.get_exception(code)
# Prefix the exception message with the thread position
prefix = "tid=%s ctaid=%s" % (tid, ctaid)
if exc_args:
exc_args = ("%s: %s" %
(prefix, exc_args[0]), ) + exc_args[1:]
else:
exc_args = prefix,
raise exccls(*exc_args)
# retrieve auto converted arrays
for wb in retr:
wb()
def _kernel_call(self, args, griddim, blockdim, stream=0, sharedmem=0):
# Prepare kernel
cufunc = self._func.get()
if self.debug:
excname = cufunc.name + "__errcode__"
excmem, excsz = cufunc.module.get_global_symbol(excname)
assert excsz == ctypes.sizeof(ctypes.c_int)
excval = ctypes.c_int()
excmem.memset(0, stream=stream)
# Prepare arguments
retr = [] # hold functors for writeback
kernelargs = []
for t, v in zip(self.argument_types, args):
self._prepare_args(t, v, stream, retr, kernelargs)
# Configure kernel
cu_func = cufunc.configure(griddim, blockdim,
stream=stream,
sharedmem=sharedmem)
# Invoke kernel
cu_func(*kernelargs)
if self.debug:
driver.device_to_host(ctypes.addressof(excval), excmem, excsz)
if excval.value != 0:
# An error occurred
def load_symbol(name):
mem, sz = cufunc.module.get_global_symbol("%s__%s__" %
(cufunc.name,
name))
val = ctypes.c_int()
driver.device_to_host(ctypes.addressof(val), mem, sz)
return val.value
tid = [load_symbol("tid" + i) for i in 'zyx']
ctaid = [load_symbol("ctaid" + i) for i in 'zyx']
code = excval.value
exccls, exc_args = self.call_helper.get_exception(code)
# Prefix the exception message with the thread position
prefix = "tid=%s ctaid=%s" % (tid, ctaid)
if exc_args:
exc_args = ("%s: %s" % (prefix, exc_args[0]),) + exc_args[1:]
else:
exc_args = prefix,
raise exccls(*exc_args)
# retrieve auto converted arrays
for wb in retr:
wb()
