Struct MegakernelConfig 

pub struct MegakernelConfig {
    pub worker_count: u32,
    pub max_wall_time: Duration,
    pub expected_items_per_worker: u32,
    pub workload: MegakernelWorkloadHints,
}

Configuration for one megakernel dispatch invocation.

Fields

worker_count: u32

Number of persistent worker workgroups.

max_wall_time: Duration

Maximum wall-clock time the megakernel runs before draining queued work and exiting.

expected_items_per_worker: u32

Hint to the scheduler about expected items per worker.

workload: MegakernelWorkloadHints

Optional workload-shape hints consumed by the launch policy.

Implementations

impl MegakernelConfig

pub fn validate(&self) -> Result<(), BackendError>

Validate the config and surface actionable errors.

Errors

Returns an error when the worker count is zero or the wall-clock budget is empty, because either condition would make persistent dispatch unschedulable.

pub fn dispatch_grid( &self, queue_len: u32, max_workgroup_size_x: u32, ) -> [u32; 3]

Compute the direct-dispatch grid for queue_len logical work slots.

worker_count is the caller’s persistent worker-workgroup ceiling; the returned grid never launches more workgroups than that ceiling or the backend occupancy cap.

pub const fn launch_request( &self, queue_len: u32, max_workgroup_size_x: u32, max_compute_workgroups_per_dimension: u32, max_compute_invocations_per_workgroup: u32, ) -> MegakernelLaunchRequest

Build a policy request from this config and adapter limits.

pub fn launch_request_for_tasks( &self, tasks: &[TaskWorkItem], max_workgroup_size_x: u32, max_compute_workgroups_per_dimension: u32, max_compute_invocations_per_workgroup: u32, ) -> Result<MegakernelLaunchRequest, BackendError>

Build a policy request from device-visible continuation task slots.

Paused, completed, empty, running, and faulted tasks do not add launch lanes. Yielded and requeued tasks stay schedulable so the GPU can resume them without a CPU-side republish loop.

Errors

Returns BackendError when a task slot contains an invalid state word.

pub fn launch_recommendation( &self, queue_len: u32, max_workgroup_size_x: u32, max_compute_workgroups_per_dimension: u32, max_compute_invocations_per_workgroup: u32, ) -> Result<MegakernelLaunchRecommendation, BackendError>

Recommend one launch shape through the shared megakernel policy.

Errors

Returns BackendError when adapter limits are malformed.

pub fn launch_recommendation_for_tasks( &self, tasks: &[TaskWorkItem], max_workgroup_size_x: u32, max_compute_workgroups_per_dimension: u32, max_compute_invocations_per_workgroup: u32, ) -> Result<MegakernelLaunchRecommendation, BackendError>

Recommend one launch shape for a continuation task queue.

Errors

Returns BackendError when adapter limits are malformed or any task slot contains an invalid state word.

Trait Implementations

impl Clone for MegakernelConfig

fn clone(&self) -> MegakernelConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MegakernelConfig

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

Formats the value using the given formatter.

impl Default for MegakernelConfig

fn default() -> Self

Returns the “default value” for a type.