cuda::experimental::graph

Defined in include/cuda/experimental/__graph/graph.cuh

struct graph

An owning wrapper for a cudaGraphExec_t handle.

The graph class provides a safe and convenient interface for managing the lifecycle of a cudaGraphExec_t object, ensuring proper cleanup and resource management. It supports move semantics, resource release, and launch of the CUDA graph.

See also

cuda::experimental::graph_builder

Note

The graph object is not directly constructible. One is obtained by calling the instantiate() method on a graph_builder object.

Public Functions

inline constexpr graph(graph &&__other) noexcept

Move constructor for graph.

Note

After the move, the source object is left in the empty state.

Parameters

__other – The graph object to move from.

Post

__other.get() == nullptr

inline constexpr ~graph()

Destructor for graph.

Ensures proper cleanup of the CUDA graph execution object.

Throws

None –

inline constexpr auto operator=(graph &&__other) noexcept -> graph&

Move assignment operator for graph.

Note

After the move, the source object is left in the empty state.

Parameters

__other – The graph object to move from.

Throws

None –

Returns

A reference to the current object.

Post

__other.get() == nullptr

inline constexpr void swap(graph &__other) noexcept

Swaps the contents of this graph with another.

Parameters

__other – The graph object to swap with.

Throws

None –

inline constexpr auto get() const noexcept -> cudaGraphExec_t

Retrieves the underlying CUDA graph execution object.

Throws

None –

Returns

The cudaGraphExec_t handle.

inline constexpr auto release() noexcept -> cudaGraphExec_t

Releases ownership of the CUDA graph execution object.

Throws

None –

Returns

The cudaGraphExec_t handle, leaving this object in a null state.

Post

get() == nullptr

inline constexpr void reset() noexcept

Resets the graph object, destroying the underlying CUDA graph execution object.

Throws

cuda::std::cuda_error – if cudaGraphExecDestroy fails.

Post

get() == nullptr

inline void launch(stream_ref __stream)

Launches the CUDA graph execution object on the specified stream.

Parameters

__stream – The stream on which to launch the graph.

Throws

cuda::std::cuda_error – if cudaGraphLaunch fails.

Public Static Functions

static inline constexpr auto from_native_handle(cudaGraphExec_t __exec) noexcept -> graph

Constructs a graph object from a native CUDA graph execution handle.

Parameters

__exec – The native CUDA graph execution handle to construct the graph object from.

Throws

None –

Post

get() == __exec