Rust GPU Chimera Demo

A cross-platform demo of a single Rust codebase running on both the CPU and GPU via CUDA, Vulkan, Metal, and DirectX. There are no shader or kernel languages used, only Rust.

Supported Configurations

Running the Demo

The demo runs a bitonic sort on various data types (u32, i32, f32) with different sizes and configurations.

Linux

# CPU execution
cargo run --release

# Vulkan via wgpu
cargo run --release --features wgpu

# Vulkan via ash
cargo run --release --features ash

# CUDA (NVIDIA GPU required)
cargo run --release --features cuda

macOS

# CPU execution
cargo run --release

# Metal via wgpu (SPIR-V → MSL translation)
cargo run --release --features wgpu

# Vulkan via wgpu (requires MoltenVK)
cargo run --release --features wgpu,vulkan

# Vulkan via ash (requires MoltenVK)
cargo run --release --features ash

Windows

# CPU execution
cargo run --release

# DirectX 12 via wgpu (SPIR-V → HLSL translation)
cargo run --release --features wgpu

# Vulkan via wgpu
cargo run --release --features wgpu,vulkan

# Vulkan via ash
cargo run --release --features ash

# CUDA (NVIDIA GPU required)
cargo run --release --features cuda

Instead of cargo run you can replace it with cargo test to run unit tests for the same configuration.

Project Structure

rust-gpu-chimera-demo/
├── kernel/           # Compute kernel logic and entrypoints
│   └── src/
│       └── lib.rs
├── shared/           # Code that runs on both the CPU and GPU
│   └── src/
│       └── lib.rs
├── src/
│   ├── runners/      # Code that runs on the CPU/host and interfaces with the GPU
│   │   ├── cpu.rs
│   │   ├── cuda.rs
│   │   ├── wgpu.rs
│   │   └── ash.rs
│   ├── lib.rs
│   └── main.rs       # Demo application binary
└── build.rs          # Kernel compilation orchestration

Footnotes

1. CUDA is not supported on macOS/iOS/tvOS/visionOS.
   ZLUDA could potentially enable CUDA on these platforms in the future. ↩ ↩² ↩³ ↩⁴

2. CUDA is not supported on Android.
   ZLUDA could potentially enable CUDA on Android in the future. ↩