Struct WorkgroupAutoTuner 

pub struct WorkgroupAutoTuner { /* private fields */ }

Auto-tuner that selects optimal workgroup sizes based on device limits.

Implementations

impl WorkgroupAutoTuner

pub fn new(limits: DeviceLimits) -> Self

Create a new auto-tuner with the given device limits.

pub fn with_defaults() -> Self

Create a new auto-tuner with default limits.

pub fn limits(&self) -> &DeviceLimits

Get the device limits.

pub fn tune_1d( &self, total_elements: u32, ) -> (WorkgroupSize, DispatchDimensions)

Auto-tune a 1D workgroup size for a linear problem.

Picks the largest warp-aligned size that fits within device limits.

pub fn tune_2d( &self, width: u32, height: u32, ) -> (WorkgroupSize, DispatchDimensions)

Auto-tune a 2D workgroup size for an image-like problem.

Balances squareness with warp alignment and device limits.

pub fn tune_2d_with_shared_memory( &self, width: u32, height: u32, shared_bytes_per_pixel: u32, ) -> (WorkgroupSize, DispatchDimensions)

Auto-tune for a 2D problem with shared memory requirements.

Takes into account the per-pixel shared memory usage and ensures the workgroup’s shared memory fits within device limits.

pub fn estimate_efficiency( &self, problem_width: u32, problem_height: u32, workgroup: &WorkgroupSize, ) -> f64

Estimate the efficiency of a given configuration.