本文整理汇总了Python中pycuda.driver.Event方法的典型用法代码示例。如果您正苦于以下问题：Python driver.Event方法的具体用法？Python driver.Event怎么用？Python driver.Event使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类pycuda.driver的用法示例。
在下文中一共展示了driver.Event方法的5个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: _get_events
# 需要导入模块: from pycuda import driver [as 别名]
# 或者: from pycuda.driver import Event [as 别名]
def _get_events():
return (drv.Event(), drv.Event()) 示例2: __init__
# 需要导入模块: from pycuda import driver [as 别名]
# 或者: from pycuda.driver import Event [as 别名]
def __init__(self, transformer, comm, op):
super(CudaAllReduceKernel, self).__init__(transformer)
self.op = op
self.tensor = op.tensor_description()
self.device_id = int(transformer.device_id)
self.device_ids = list(map(int, self.op.device_ids))
self.event = drv.Event(flags=event_flags.INTERPROCESS | event_flags.DISABLE_TIMING)
self.stream = drv.Stream()
self.output_buff_dict = {}
self.scratch_buff_dict = {}
self.event_buff_dict = {}
self.comm = comm
self.init_buffers() 示例3: init_buffers
# 需要导入模块: from pycuda import driver [as 别名]
# 或者: from pycuda.driver import Event [as 别名]
def init_buffers(self):
shape = self.op.args[0].tensor_description().shape
dtype = self.op.args[0].tensor_description().dtype
n_devs = len(self.op.device_ids)
size = self.op.args[0].tensor_description().axes.size
segment_size = calculate_segment_size(size, n_devs)
# Allocate output and scratch buffers
self.output_buff = gpuarray.zeros(shape, dtype)
self.scratch_buff = gpuarray.zeros(segment_size * n_devs, dtype)
self.output_buff_dict[self.device_id] = self.output_buff.gpudata
self.scratch_buff_dict[self.device_id] = self.scratch_buff.gpudata
# Allocate IPC handles
output_ipc_hdl = drv.mem_get_ipc_handle(self.output_buff.gpudata)
scratch_ipc_hdl = drv.mem_get_ipc_handle(self.scratch_buff.gpudata)
event_ipc_hdl = self.event.ipc_handle()
# Broadcast handles to others
msg = (self.device_id, output_ipc_hdl, scratch_ipc_hdl, event_ipc_hdl)
for i in self.device_ids:
if i == self.device_id:
self.comm.bcast(msg, root=i)
else:
(peer_id,
output_ipc_hdl,
scratch_ipc_hdl,
event_ipc_hdl) = self.comm.bcast(None, root=i)
output_hdl = drv.IPCMemoryHandle(output_ipc_hdl)
scratch_hdl = drv.IPCMemoryHandle(scratch_ipc_hdl)
event_hdl = drv.Event.from_ipc_handle(event_ipc_hdl)
self.output_buff_dict[peer_id] = output_hdl
self.scratch_buff_dict[peer_id] = scratch_hdl
self.event_buff_dict[peer_id] = event_hdl 示例4: init_mark
# 需要导入模块: from pycuda import driver [as 别名]
# 或者: from pycuda.driver import Event [as 别名]
def init_mark(self):
"""
Generate a timing mark object.
Returns:
timing mark (pycude driver event)
"""
return drv.Event() 示例5: _get_events
# 需要导入模块: from pycuda import driver [as 别名]
# 或者: from pycuda.driver import Event [as 别名]
def _get_events():
return (drv.Event(), drv.Event())
# debugging tool
# import re
# import traceback as tb
# nrv_re = re.compile(r'nervanagpu\.py$')
# def print_trace():
# caller = None
# for frame in tb.extract_stack():
# if GPUTensor.nrv_re.search(frame[0]):
# break
# caller = (frame[0],frame[1])
# print caller