Struct MultiGpuScheduler 

pub struct MultiGpuScheduler { /* private fields */ }

Multi-GPU frame scheduler with configurable load-balancing.

Thread Safety

MultiGpuScheduler is Send + Sync and can be shared across threads via Arc. The internal round-robin counter is protected by a Mutex.

Implementations

impl MultiGpuScheduler

pub fn new( devices: Vec<(Arc<GpuDevice>, f32)>, strategy: LoadBalanceStrategy, ) -> Result<Self>

Create a new scheduler from a list of (device, weight) pairs.

Errors

Returns GpuError::NotSupported if devices is empty.

pub fn equal_weight(devices: Vec<Arc<GpuDevice>>) -> Result<Self>

Create a scheduler from a list of devices with equal weights using the default LeastLoaded strategy.

Errors

Returns an error if devices is empty.

pub fn device_count(&self) -> usize

Number of devices managed by this scheduler.

pub fn select_device(&self) -> usize

Select the best device slot index for the next frame according to the current strategy.

pub fn dispatch<F, T>(&self, work_fn: F) -> Result<(T, usize)>

where F: FnOnce(&GpuDevice) -> Result<T>,

Dispatch a frame to the best available device.

Returns the index of the selected device slot.

The work_fn closure receives the selected GpuDevice and performs the actual GPU work. On success the measured latency (in seconds) is reported via on_complete; on failure on_failure is called.

pub fn device_stats(&self) -> Vec<DeviceStats>

Get a snapshot of per-device statistics.

pub fn total_dispatched(&self) -> u64

Total frames dispatched across all devices.

pub fn total_completed(&self) -> u64

Total frames completed (successfully) across all devices.

pub fn slot(&self, index: usize) -> Option<&DeviceSlot>

Get a reference to the device slot at index.

Returns None if the index is out of range.