Struct ConditionalCudaGraphBody 

pub struct ConditionalCudaGraphBody { /* private fields */ }

CUDA-owned body graph produced while adding one conditional-WHILE node.

The body graph and conditional handle are owned by the parent graph. They must not be destroyed independently. The numeric handle is intended to be passed by value to a device kernel that calls cudaGraphSetConditional.

Implementations

impl ConditionalCudaGraphBody

pub fn graph(&self) -> CUgraph

pub fn handle(&self) -> CUgraphConditionalHandle

pub fn context(&self) -> CUcontext

pub fn linear_chain_node_kinds( &self, ) -> Result<Vec<CudaGraphNodeKind>, CudaConditionalGraphUnavailable>

Return this body’s actual node kinds in dependency-chain order.

The body must be a single linear dependency chain. CUDA’s node-list enumeration order is not used as an execution-order signal.

pub fn capture_on_stream<F, E>( &self, stream: &CudaStream, record: F, ) -> Result<(), CudaConditionalGraphUnavailable>

where F: FnOnce() -> Result<(), E>, E: Display,

Capture graph-compatible work directly into this conditional body.

stream must be a non-default stream. The callback must not allocate or free CUDA memory, synchronize, or record/wait on events. Host-only suballocation from retained device storage is allowed if it cannot grow or remap that storage; captured addresses must remain valid and must not be recycled while graph consumers or saved backward values need them. CUDA conditional bodies allow only kernel, empty, child-graph, device memcpy/memset, and nested conditional nodes. Capture setup and cleanup may wait for deferred destruction under the module-level lifecycle contract.

Trait Implementations

impl Debug for ConditionalCudaGraphBody

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.