Struct MemoryBandwidthModel 

pub struct MemoryBandwidthModel {
    pub peak_bandwidth_gbs: f64,
    pub peak_compute_gflops: f64,
}

GPU memory bandwidth model.

Estimates the effective bandwidth and the roofline-model bound for a kernel.

Fields

peak_bandwidth_gbs: f64

Peak memory bandwidth in GB/s.

peak_compute_gflops: f64

Peak compute throughput in GFLOP/s.

Implementations

impl MemoryBandwidthModel

pub fn mid_range() -> Self

Create a model for a mid-range discrete GPU.

pub fn arithmetic_intensity(flops: f64, bytes_accessed: f64) -> f64

Compute the arithmetic intensity (FLOP/byte) of a kernel.

flops – total floating-point operations. bytes_accessed – total bytes read/written.

pub fn roofline_performance(&self, arithmetic_intensity: f64) -> f64

Roofline performance estimate (GFLOP/s) given arithmetic intensity.

pub fn estimated_runtime_ms(&self, flops: f64, bytes_accessed: f64) -> f64

Estimated kernel execution time in milliseconds.

flops – total FLOPs in the kernel. bytes_accessed – total bytes transferred.

pub fn is_bandwidth_bound(&self, arithmetic_intensity: f64) -> bool

Whether the kernel is bandwidth-bound or compute-bound.

Returns true if bandwidth-bound (arithmetic intensity below the ridge point).

Trait Implementations

impl Clone for MemoryBandwidthModel

fn clone(&self) -> MemoryBandwidthModel

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MemoryBandwidthModel

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

Formats the value using the given formatter.