#include <taskflow/cuda/cuda_stream.hpp>
template<typename Creator, typename Deleter>
tf::cudaStreamBase class

class to create a smart pointer wrapper for managing cudaStream_t

Template parameters
Creator functor to create the stream (used in constructor)
Deleter functor to delete the stream (used in destructor)

The cudaStream class encapsulates a cudaStream_t using std::unique_ptr, ensuring that CUDA events are properly created and destroyed with a unique ownership.

Public types

using base_type = std::unique_ptr<std::remove_pointer_t<cudaStream_t>, Deleter>
base type for the underlying unique pointer

Constructors, destructors, conversion operators

template<typename... ArgsT>
cudaStreamBase(ArgsT && ... args) explicit
constructs a cudaStream object by passing the given arguments to the stream creator
operator cudaStream_t() const noexcept
implicit conversion to the underlying cudaStream_t object

Public functions

void synchronize() const
synchronizes the associated stream
void begin_capture(cudaStreamCaptureMode m = cudaStreamCaptureModeGlobal) const
begins graph capturing on the stream
auto end_capture() const -> cudaGraph_t
ends graph capturing on the stream
void record(cudaEvent_t event) const
records an event on the stream
void wait(cudaEvent_t event) const
waits on an event

Typedef documentation

template<typename Creator, typename Deleter>
using tf::cudaStreamBase<Creator, Deleter>::base_type = std::unique_ptr<std::remove_pointer_t<cudaStream_t>, Deleter>

base type for the underlying unique pointer

This alias provides a shorthand for the underlying std::unique_ptr type that manages CUDA stream resources with an associated deleter.

Function documentation

template<typename Creator, typename Deleter> template<typename... ArgsT>
tf::cudaStreamBase<Creator, Deleter>::cudaStreamBase(ArgsT && ... args) explicit

constructs a cudaStream object by passing the given arguments to the stream creator

Parameters
args arguments to pass to the stream creator

Constructs a cudaStream object by passing the given arguments to the stream creator

template<typename Creator, typename Deleter>
tf::cudaStreamBase<Creator, Deleter>::operator cudaStream_t() const noexcept

implicit conversion to the underlying cudaStream_t object

Returns the underlying cudaStream_t object, equivalently calling base_type::get().

template<typename Creator, typename Deleter>
void tf::cudaStreamBase<Creator, Deleter>::synchronize() const

synchronizes the associated stream

Equivalently calling cudaStreamSynchronize to block until this stream has completed all operations.

template<typename Creator, typename Deleter>
void tf::cudaStreamBase<Creator, Deleter>::begin_capture(cudaStreamCaptureMode m = cudaStreamCaptureModeGlobal) const

begins graph capturing on the stream

When a stream is in capture mode, all operations pushed into the stream will not be executed, but will instead be captured into a graph, which will be returned via cudaStream::end_capture.

A thread's mode can be one of the following:

  • cudaStreamCaptureModeGlobal: This is the default mode. If the local thread has an ongoing capture sequence that was not initiated with cudaStreamCaptureModeRelaxed at cuStreamBeginCapture, or if any other thread has a concurrent capture sequence initiated with cudaStreamCaptureModeGlobal, this thread is prohibited from potentially unsafe API calls.
  • cudaStreamCaptureModeThreadLocal: If the local thread has an ongoing capture sequence not initiated with cudaStreamCaptureModeRelaxed, it is prohibited from potentially unsafe API calls. Concurrent capture sequences in other threads are ignored.
  • cudaStreamCaptureModeRelaxed: The local thread is not prohibited from potentially unsafe API calls. Note that the thread is still prohibited from API calls which necessarily conflict with stream capture, for example, attempting cudaEventQuery on an event that was last recorded inside a capture sequence.

template<typename Creator, typename Deleter>
cudaGraph_t tf::cudaStreamBase<Creator, Deleter>::end_capture() const

ends graph capturing on the stream

Equivalently calling cudaStreamEndCapture to end capture on stream and returning the captured graph. Capture must have been initiated on stream via a call to cudaStream::begin_capture. If capture was invalidated, due to a violation of the rules of stream capture, then a NULL graph will be returned.

template<typename Creator, typename Deleter>
void tf::cudaStreamBase<Creator, Deleter>::record(cudaEvent_t event) const

records an event on the stream

Equivalently calling cudaEventRecord to record an event on this stream, both of which must be on the same CUDA context.

template<typename Creator, typename Deleter>
void tf::cudaStreamBase<Creator, Deleter>::wait(cudaEvent_t event) const

waits on an event

Equivalently calling cudaStreamWaitEvent to make all future work submitted to stream wait for all work captured in event.