Struct KernelProfile 

pub struct KernelProfile {
    pub kernel: Arc<CachedKernel>,
    pub device: DeviceSpec,
    pub num_buffers: usize,
    pub elapsed: Duration,
}

Per-kernel timing from a profiled execution.

Holds an Arc<CachedKernel> for zero-copy access to kernel metadata (entry point, generated code, global/local size, variable names).

Example

let plan = tensor.prepare()?;
let profiles = plan.execute_profiled()?;

for (i, p) in profiles.iter().enumerate() {
    println!("{:4} {:>8.3}ms  {}  ({} bufs, {:?})",
        i, p.elapsed.as_secs_f64() * 1000.0,
        p.kernel.entry_point, p.num_buffers, p.device);
}

Fields

kernel: Arc<CachedKernel>

Compiled kernel (entry_point, code, global_size, local_size, var_names).

device: DeviceSpec

Device this kernel executed on.

num_buffers: usize

Number of buffer arguments.

elapsed: Duration

Wall-clock execution time.