Module multi_gpu 

Multi-GPU context management with per-device context pools.

When working with multiple GPUs, it is common to maintain one CUDA context per device and dispatch work across them. DevicePool automates context lifecycle management and provides scheduling helpers (round-robin, best-available) for multi-GPU workloads.

Thread safety

DevicePool is Send + Sync. Each context is wrapped in an Arc<Context> so it can be shared across threads. The caller is responsible for calling Context::set_current on the appropriate thread before issuing driver calls.

Example

use oxicuda_driver::multi_gpu::DevicePool;

oxicuda_driver::init()?;
let pool = DevicePool::new()?;
println!("managing {} devices", pool.device_count());

for (dev, ctx) in pool.iter() {
    ctx.set_current()?;
    println!("device {}: {}", dev.ordinal(), dev.name()?);
}

Structs

DevicePool

Per-device context pool for multi-GPU management.