Struct GpuGraph 

pub struct GpuGraph { /* private fields */ }

A captured CUDA graph that can be replayed with minimal dispatch overhead.

Created via gpu_graph_capture or manually with new() + capture_begin() / capture_end().

Implementations

impl GpuGraph

pub fn new() -> Result<Self>

Create a new empty CUDA graph. Errors on CPU-only builds.

pub fn capture_begin( &mut self, pool: Option<MemPoolId>, mode: CaptureMode, ) -> Result<()>

Begin capturing GPU operations into this graph.

All CUDA kernel launches on the current stream after this call will be recorded until capture_end.

pub fn capture_end(&mut self) -> Result<()>

End capture and finalize the graph.

pub fn replay(&self) -> Result<()>

Replay the captured graph. All kernels are launched with a single CUDA API call, eliminating per-kernel dispatch overhead.

pub fn reset(&mut self) -> Result<()>

Reset the graph, allowing recapture.

pub fn pool(&self) -> MemPoolId

Get the memory pool ID for this graph.

Trait Implementations

impl Drop for GpuGraph

fn drop(&mut self)

Executes the destructor for this type.

fn pin_drop(self: Pin<&mut Self>)

🔬This is a nightly-only experimental API. (pin_ergonomics)

Execute the destructor for this type, but different to Drop::drop, it requires self to be pinned.