cuda::experimental::conditional_handle#

struct conditional_handle#

A thin, non-owning wrapper around a cudaGraphConditionalHandle.

A conditional handle is a graph-scoped token whose value at runtime controls whether the body of an if or while conditional node executes. The handle is owned by the graph — there is no destroy API — so this wrapper is trivially copyable and safe to pass by value into device kernels.

Users can construct a handle directly, or let make_if_node / make_while_node create one automatically.

Public Functions

inline explicit conditional_handle(
graph_builder_ref __graph,
bool __default_val = true
)#

Creates a conditional handle for the given graph.

Parameters:

  • __graph – Graph in which the conditional node will be inserted.

  • __default_val – Initial value of the handle (true = execute body, false = skip).

Throws:

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

inline void set_value(bool __value) const noexcept#

Sets the runtime value of the conditional handle from device code.

Parameters:

__value – Non-zero to execute the body, zero to skip.

inline void enable() const noexcept#

Convenience: enables execution of the conditional body (sets the handle to 1).

inline void disable() const noexcept#

Convenience: disables execution of the conditional body (sets the handle to 0).

inline ::cudaGraphConditionalHandle get() const noexcept#

Returns the underlying cudaGraphConditionalHandle.