tf::cudaScopedPerThreadEvent class
#include <src/taskflow/cuda/cuda_stream.hpp>

class that provides RAII-styled guard of event acquisition

Contents

Sample usage:

{
  tf::cudaScopedPerThreadEvent event(1);  // acquires a event on device 1

  // use event as a normal cuda event (cudaEvent_t)
  cudaStreamWaitEvent(stream, event);

}  // leaving the scope releases the event back to the pool on device 1

The scoped per-thread event is primarily used by tf::Executor to execute CUDA tasks (e.g., tf::cudaFlow, tf::cudaFlowCapturer).

cudaScopedPerThreadEvent is non-copyable.

Constructors, destructors, conversion operators

cudaScopedPerThreadEvent(int device) explicit
constructs a scoped event under the given device
cudaScopedPerThreadEvent()
constructs a scoped event under the current device.
~cudaScopedPerThreadEvent()
destructs the scoped event guard
operator cudaEvent_t() const
implicit conversion to the native CUDA event (cudaEvent_t)
cudaScopedPerThreadEvent(const cudaScopedPerThreadEvent&) deleted
disabled copy constructor
cudaScopedPerThreadEvent(cudaScopedPerThreadEvent&&) defaulted
default move constructor

Public functions

auto operator=(const cudaScopedPerThreadEvent&) -> cudaScopedPerThreadEvent& deleted
disabled copy assignment
auto operator=(cudaScopedPerThreadEvent&&) -> cudaScopedPerThreadEvent& deleted
default move assignment

Function documentation

tf::cudaScopedPerThreadEvent::cudaScopedPerThreadEvent(int device) explicit

constructs a scoped event under the given device

Parameters
device device context of the requested event

The constructor acquires a event from a per-thread event pool.

tf::cudaScopedPerThreadEvent::cudaScopedPerThreadEvent()

constructs a scoped event under the current device.

The constructor acquires a event from a per-thread event pool.

tf::cudaScopedPerThreadEvent::~cudaScopedPerThreadEvent()

destructs the scoped event guard

The destructor releases the event to the per-thread event pool.