Struct OccupancyModel 

pub struct OccupancyModel {
    pub compute_units: u32,
    pub max_warps_per_cu: u32,
    pub warp_size: u32,
    pub shared_mem_per_cu: u32,
    pub registers_per_cu: u32,
}

GPU occupancy model (simplified).

Models occupancy as the ratio of active warps to maximum concurrent warps.

Fields

compute_units: u32

Total number of compute units (SMs / CUs).

max_warps_per_cu: u32

Maximum warps per compute unit.

warp_size: u32

Warp size (threads per warp, typically 32 for NVIDIA or 64 for AMD).

shared_mem_per_cu: u32

Shared memory per compute unit (bytes).

registers_per_cu: u32

Registers per compute unit.

Implementations

impl OccupancyModel

pub fn mid_range() -> Self

Create a model resembling a mid-range discrete GPU.

pub fn estimate_occupancy( &self, workgroup_size: u32, shared_mem_bytes: u32, registers_per_thread: u32, ) -> f64

Estimate the theoretical occupancy (0.0–1.0) for a kernel.

Occupancy is limited by:

1. Workgroup size (must not exceed warp_size * max_warps_per_cu).
2. Shared memory usage.
3. Register usage.

pub fn peak_gflops(&self, clock_mhz: f64) -> f64

Total theoretical peak throughput in GFLOP/s (mock model).

Assumes 2 FP32 ops per clock per SIMD unit.

Trait Implementations

impl Clone for OccupancyModel

fn clone(&self) -> OccupancyModel

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for OccupancyModel

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

Formats the value using the given formatter.