Module queue 

Hardware command queue abstraction for parallel execution.

This module provides a device-agnostic interface for command queues, abstracting over CUDA streams, Metal command buffers, CPU task queues, etc.

Design

The HardwareQueue trait uses a builder pattern for chaining operations:

queue
    .wait(&signal, 1)       // Wait for dependency
    .exec(&kernel, &bufs)   // Execute kernel
    .signal(&signal, 2)     // Signal completion
    .submit()?;             // Submit to hardware

Queue Types

Most devices have two queue types:

• Compute queue: For kernel execution
• Copy queue: For DMA transfers (optional, some devices share)

Structs

DynQueue

Type-erased queue for use in the unified executor.

ExecParams

Kernel execution parameters.

Traits

HardwareQueue

Hardware command queue for submitting operations to a device.

Program

Compiled program that can be executed on a queue.

QueueFactory

Factory for creating hardware queues.