Trait KernelTrace 

pub trait KernelTrace:
    Send
    + Sync
    + 'static {
    // Provided methods
    fn before_enqueue(&self, info: &KernelInfo<'_>) { ... }
    fn after_enqueue(
        &self,
        info: &KernelInfo<'_>,
        result: Result<(), &GpuError>,
    ) { ... }
    fn before_complete(&self, info: &KernelInfo<'_>) { ... }
    fn after_complete(
        &self,
        info: &KernelInfo<'_>,
        result: Result<(), &GpuError>,
        latency: Duration,
    ) { ... }
}

Lifecycle hook receiver. All four methods have empty default bodies, so a custom trace can override only the events it cares about.

The trait is always compiled; no feature flag is required. When no trace is attached to a KernelEnvelope, the envelope skips every call and adds zero runtime cost beyond a single Option::is_some check (which the optimizer typically folds away).

Provided Methods

fn before_enqueue(&self, info: &KernelInfo<'_>)

Fires immediately before the synchronous enqueue closure runs.

fn after_enqueue(&self, info: &KernelInfo<'_>, result: Result<(), &GpuError>)

Fires immediately after the synchronous enqueue closure returns. result is Ok(()) on success or Err(&GpuError) if the enqueue body failed.

fn before_complete(&self, info: &KernelInfo<'_>)

Fires just before the completion future is awaited (i.e. after a successful enqueue, on the spawned Tokio task).

fn after_complete( &self, info: &KernelInfo<'_>, result: Result<(), &GpuError>, latency: Duration, )

Fires after the completion future resolves. latency is the wall-clock duration between before_complete and the resolved completion (i.e. host-observed completion latency, not GPU time).

Implementors