Struct SimulatedParallelExecutor 

pub struct SimulatedParallelExecutor { /* private fields */ }

Simulated multi-GPU executor for development/testing

Uses threading to simulate multiple GPUs within a single process. Useful for testing parallel algorithms without actual multi-GPU hardware.

Implementations

impl SimulatedParallelExecutor

pub fn new( config: ParallelConfig, rank: usize, shared_buffers: Arc<RwLock<Vec<Vec<f32>>>>, barrier: Arc<Barrier>, ) -> Self

Create a simulated parallel executor

Trait Implementations

impl ParallelExecutor for SimulatedParallelExecutor

fn config(&self) -> &ParallelConfig

Get the parallelism configuration

fn rank(&self) -> usize

Get local rank

fn world_size(&self) -> usize

Get world size

fn device(&self) -> Device

Get the local device

fn barrier<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Barrier synchronization across all ranks

fn all_reduce<'life0, 'life1, 'async_trait>( &'life0 self, data: &'life1 mut [f32], ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

All-reduce operation (sum by default)

fn all_gather<'life0, 'life1, 'async_trait>( &'life0 self, local_data: &'life1 [f32], ) -> Pin<Box<dyn Future<Output = Result<Vec<f32>>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

All-gather operation

fn broadcast<'life0, 'life1, 'async_trait>( &'life0 self, data: &'life1 mut [f32], ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Broadcast from master to all ranks

fn is_master(&self) -> bool

Check if this is the master rank