Struct PendingBatchStep 

pub struct PendingBatchStep { /* private fields */ }

A one-row decode whose device work has been enqueued but whose result has not crossed back to the host yet. The worker owns the CUDA context, so this is deliberately a poll-at-the-next scheduler-boundary handoff rather than a background CUDA thread. Keeping the completion event and output buffers alive prevents the async-pool from recycling them while the next step runs.

Implementations

impl PendingBatchStep

pub fn wait(self) -> Result<(Vec<Vec<f32>>, Vec<Option<u32>>), Box<dyn Error>>

Wait for this step only, then perform one ordered readback of its host-visible results. The compute stream may already be carrying the following step when this runs.