def __init__(self, kernel=None, dat_dict=None):
self._dat_dict = access.DatArgStore(self._get_allowed_types(),
dat_dict)
self._cc = build.TMPCC
self._temp_dir = runtime.BUILD_DIR
if not os.path.exists(self._temp_dir):
os.mkdir(self._temp_dir)
self._kernel = kernel
self.loop_timer = ppmd.modules.code_timer.LoopTimer()
self.wrapper_timer = ppmd.opt.Timer(runtime.TIMER)
self._components = None
self._generate()
self._lib = build.simple_lib_creator(self._generate_header_source(),
self._components['LIB_SRC'],
self._kernel.name,
CC=self._cc)
self._group = None
for pd in self._dat_dict.items():
if issubclass(type(pd[1][0]), data.ParticleDat):
if pd[1][0].group is not None:
self._group = pd[1][0].group
break
def __init__(self, kernel, dat_dict, n=None, types_map=None):
self._types_map = types_map
self._kernel = kernel
self._dat_dict = access.DatArgStore(self._get_allowed_types(),
dat_dict)
# set compiler as NVCC default
self._cc = cuda_build.NVCC
self.loop_timer = ppmd.modules.code_timer.LoopTimer()
self._components = {
'LIB_PAIR_INDEX_0': '_i',
'LIB_NAME': str(self._kernel.name) + '_wrapper'
}
self._group = None
for pd in self._dat_dict.items():
if issubclass(type(pd[1][0]), cuda_data.ParticleDat):
if pd[1][0].group is not None:
self._group = pd[1][0].group
break
#start code creation
self._generate()
# Create library
self._lib = cuda_build.simple_lib_creator(
self._generate_header_source(), self._components['LIB_SRC'],
self._components['LIB_NAME'])
def __init__(self, kernel=None, dat_dict=None):
self._dat_dict = access.DatArgStore(
self._get_allowed_types(),
dat_dict
)
self._cc = build.TMPCC
self._temp_dir = runtime.BUILD_DIR
if not os.path.exists(self._temp_dir):
os.mkdir(self._temp_dir)
self._kernel = kernel
self.loop_timer = ppmd.modules.code_timer.LoopTimer()
self.wrapper_timer = ppmd.opt.Timer(runtime.TIMER)
self._components = {'LIB_PAIR_INDEX_0': '_i',
'LIB_PAIR_INDEX_1': '_j',
'OMP_THREAD_INDEX_SYM': '_threadid',
'LIB_NAME': str(self._kernel.name) + '_wrapper',
'LIB_HEADERS': [cgen.Include('omp.h', system=True),],
'OMP_SHARED_SYMS': []}
self._gather_size_limit = 8
self._generate()
self._group = None
for pd in self._dat_dict.items():
if issubclass(type(pd[1][0]), data.PositionDat):
self._group = pd[1][0].group
break
self._lib = build.simple_lib_creator(self._generate_header_source(),
self._components['LIB_SRC'],
self._kernel.name,
CC=self._cc)
def __init__(self, kernel=None, dat_dict=None, shell_cutoff=None):
self._dat_dict = access.DatArgStore(self._get_allowed_types(),
dat_dict)
self._cc = build.TMPCC
self._kernel = kernel
self.shell_cutoff = shell_cutoff
self.loop_timer = modules.code_timer.LoopTimer()
self.wrapper_timer = opt.Timer(runtime.TIMER)
self.list_timer = opt.Timer(runtime.TIMER)
self._gather_space = host.ThreadSpace(100, ctypes.c_uint8)
self._generate()
self._offset_list = host.Array(ncomp=27, dtype=ctypes.c_int)
self._lib = build.simple_lib_creator(self._generate_header_source(),
self._components['LIB_SRC'],
self._kernel.name,
CC=self._cc)
self._group = None
for pd in self._dat_dict.items():
if issubclass(type(pd[1][0]), data.PositionDat):
self._group = pd[1][0].group
break
#assert self._group is not None, "No cell to particle map found"
if self._group is not None:
self._make_cell_list(self._group)
self._kernel_execution_count = INT64(0)
self._invocations = 0
self._jstore = [host.Array(ncomp=100, dtype=ctypes.c_int) for tx in \
range(runtime.NUM_THREADS)]
def __init__(self, kernel=None, dat_dict=None, shell_cutoff=None, sub_divide=None):
self._dat_dict = access.DatArgStore(
self._get_allowed_types(), dat_dict)
self._cc = cuda_build.NVCC
self._kernel = kernel
self.shell_cutoff = shell_cutoff
if sub_divide is None:
rs_default = 5.
else:
rs_default = sub_divide
self.sub_divide_size = rs_default
#print "ACTUAL SUB CELL WIDTH", self.sub_divide_size
self.loop_timer = ppmd.modules.code_timer.LoopTimer()
self.wrapper_timer = opt.SynchronizedTimer(runtime.TIMER)
self._components = {'LIB_PAIR_INDEX_0': '_i',
'LIB_PAIR_INDEX_1': '_j',
'LIB_NAME': str(self._kernel.name) + '_wrapper'}
self._gather_size_limit = 4
self._generate()
self._lib = cuda_build.simple_lib_creator(
self._generate_header_source(),
self._components['LIB_SRC'],
self._kernel.name,
)[self._components['LIB_NAME']]
self._group = None
for pd in self._dat_dict.items():
if issubclass(type(pd[1][0]), cuda_data.PositionDat):
self._group = pd[1][0].group
break
assert self._group is not None, "No cell to particle map found"
new_decomp_flag = self._group.domain.cell_decompose(
self.shell_cutoff
)
if new_decomp_flag:
self._group.get_cell_to_particle_map().create()
self._key = (self.shell_cutoff,
self._group.domain,
self._group.get_position_dat())
_nd = PairLoopCellByCell._cell_lists
if not self._key in _nd.keys() or new_decomp_flag:
_nd[self._key] = cuda_cell.SubCellOccupancyMatrix(
domain=self._group.domain,
cell_width=self.sub_divide_size,
positions=self._group.get_position_dat(),
)
self.cell_list = _nd[self._key]
self._cell_list_count = 0
self._invocations = 0
# get the offset list
oslist = cell.convert_offset_tuples(
cell.radius_cell_decompose(shell_cutoff, self.cell_list.cell_sizes),
self.cell_list.cell_array,
remove_zero=True
)
self.offset_list = cuda_base.Array(ncomp=len(oslist), dtype=ctypes.c_int)
self.offset_list[:] = oslist[:]
def __init__(self, kernel=None, dat_dict=None, shell_cutoff=None):
self._dat_dict = access.DatArgStore(
self._get_allowed_types(), dat_dict)
self._cc = cuda_build.NVCC
self._kernel = kernel
'''
if type(shell_cutoff) is not logic.Distance:
shell_cutoff = logic.Distance(shell_cutoff)
'''
self.shell_cutoff = shell_cutoff
self.loop_timer = ppmd.modules.code_timer.LoopTimer()
self.wrapper_timer = opt.SynchronizedTimer(runtime.TIMER)
self._components = {'LIB_PAIR_INDEX_0': '_i',
'LIB_PAIR_INDEX_1': '_j',
'LIB_NAME': str(self._kernel.name) + '_wrapper'}
self._gather_size_limit = 4
self._generate()
self._lib = cuda_build.simple_lib_creator(
self._generate_header_source(),
self._components['LIB_SRC'],
self._kernel.name,
)[self._components['LIB_NAME']]
self._group = None
for pd in self._dat_dict.items():
if issubclass(type(pd[1][0]), cuda_data.PositionDat):
self._group = pd[1][0].group
break
assert self._group is not None, "No cell to particle map found"
new_decomp_flag = self._group.domain.cell_decompose(
self.shell_cutoff
)
if new_decomp_flag:
self._group.get_cell_to_particle_map().create()
self._key = (self.shell_cutoff,
self._group.domain,
self._group.get_position_dat())
_nd = PairLoopNeighbourListNSSplit._neighbour_list_dict_PNLNS_split
if not self._key in _nd.keys() or new_decomp_flag:
_nd[self._key] = cuda_cell.NeighbourListLayerSplit(
occ_matrix=self._group.get_cell_to_particle_map(),
cutoff=self.shell_cutoff
)
self.neighbour_list = _nd[self._key]
self._neighbourlist_count = 0
self._invocations = 0