Struct KernelStats 

pub struct KernelStats {
    pub execution_time_ms: f64,
    pub memory_bandwidth_gb_s: Option<f64>,
    pub estimated_compute_throughput_gflops: Option<f64>,
    pub bytes_transferred_to_device: usize,
    pub bytes_transferred_from_device: usize,
    pub estimated_occupancy_percent: f64,
}

Kernel execution statistics.

Honesty of the reported values

This crate does not currently dispatch the sparse-FFT work to a real device runtime (the actual computation is performed on the host CPU). Consequently the fields below are split into two categories:

• Measured — derived from the real problem sizes and the real host wall-clock time: Self::execution_time_ms, Self::bytes_transferred_to_device and Self::bytes_transferred_from_device.
• Estimated — analytically modelled from the KernelConfig and the FFT operation count, not measured on hardware: Self::estimated_compute_throughput_gflops and Self::estimated_occupancy_percent. They are honest model outputs (see estimate_kernel_performance), never fabricated constants. When a value genuinely cannot be modelled it is reported as None.

Fields

execution_time_ms: f64

Measured wall-clock time of the host computation backing this kernel (ms).

memory_bandwidth_gb_s: Option<f64>

Effective memory bandwidth derived from real byte counts and the measured execution_time_ms (GB/s). None if no time was measured.

estimated_compute_throughput_gflops: Option<f64>

Analytically estimated compute throughput from the FFT operation count and the measured time (GFLOPS). None if it cannot be modelled.

bytes_transferred_to_device: usize

Memory transfers host->device (bytes), computed from real sizes.

bytes_transferred_from_device: usize

Memory transfers device->host (bytes), computed from real sizes.

estimated_occupancy_percent: f64

Analytically estimated occupancy from the launch configuration (percent). This is a model output (threads/registers vs. SM limits), not a hardware measurement.

Trait Implementations

impl Clone for KernelStats

fn clone(&self) -> KernelStats

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for KernelStats

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

Formats the value using the given formatter.