Moirai GPU Compute
Cross-platform GPU compute integration for the Moirai concurrency library.
This crate provides backend-independent GPU launch planning by default and,
with the wgpu-backend feature, GPU acceleration using wgpu-rs.
Features
- Backend-independent launch planning through [
occupancy] - Optional cross-platform GPU compute using wgpu-rs
- Zero-copy buffer management with automatic memory mapping
- Async compute pipeline integration
- SIMD-fallback for unsupported hardware
- Buffer pooling and resource management
- Automatic device selection and capability detection
Architecture
The GPU integration follows these principles:
- Composable: GPU tasks integrate seamlessly with CPU tasks
- Unix Philosophy: Single responsibility for GPU compute
- Predictable: Consistent performance characteristics
- Idiomatic: Follows Rust and wgpu best practices
- Domain-centric: Designed specifically for concurrent GPU compute