gpu-rng

gpu-rng is a Rust project that leverages GPU compute shaders via wgpu to generate random numbers efficiently. It uses a Linear Congruential Generator (LCG) implemented in a compute shader written in WGSL.

This project was made purely as a learning endeavor to try and get more familiar with wgpu, WGSL shaders, random number generation processes, and rust.

Features

• GPU-Accelerated Random Number Generation: Generates random numbers in parallel using a compute shader.
• Linear Congruential Generator (LCG): Implements a simple yet effective algorithm for generating pseudorandom numbers.
• Scalable Workload: Configurable number of random numbers to generate, divided into GPU workgroups.
• Cross-Platform: Built on wgpu, ensuring support across Vulkan, Metal, DirectX, and WebGPU.

Prerequisites

Before building and running this project, ensure the following are installed:

• Rust: Install via rustup.
• ... tbh, i think that is all you need?

Project Structure

.
├── Cargo.lock
├── Cargo.toml
├── README.md
└── src
    ├── main.rs    # Main Rust program
    └── rng.wgsl   # Compute shader

• main.rs: Contains the Rust code to set up the GPU, load the shader, and retrieve random numbers.
• rng.wgsl: Defines the compute shader for random number generation.

Installation

   git clone https://github.com/evanwmart/gpu-rng.git
   cd gpu-rng
   cargo build
   cargo run

How It Works

1. Compute Shader: The shader (rng.wgsl) is executed on the GPU. It uses the global invocation ID to seed a Linear Congruential Generator (LCG), generating one random number per thread.

2. Buffers:

   • An output buffer stores the random numbers generated by the shader.
   • A staging buffer is used to transfer data back from the GPU to the CPU.

3. Workgroups: The shader divides work into groups of 64 threads. The number of threads is determined by the NUM_RANDOMS constant in main.rs.

4. Data Transfer:

   • The GPU writes random numbers to the output buffer.
   • The output is copied to a staging buffer and mapped back to the CPU for reading.

5. Rust and wgpu: The Rust code initializes wgpu, compiles the shader, and handles buffer mapping for GPU-CPU communication.

Configuration

You can configure the number of random numbers to generate by modifying NUM_RANDOMS in main.rs:

const NUM_RANDOMS: u32 = 1024; // Adjust as needed

Example Output

When you run the program, it generates and prints a list of random numbers:

$ cargo run
Random Numbers: [2891336453, 3639132858, 91961967, 839758372, ...]

Dependencies

This project uses the following crates:

• wgpu: Safe and portable GPU abstraction.
• bytemuck: Utilities for zero-cost casting of slices and arrays.
• pollster: Blocking executor for async functions.

Learning Opportunities

gpu-rng is a great project for exploring:

• GPU compute shaders.
• Parallel programming concepts.
• Rust's ecosystem for GPU programming.

Contributing

Contributions are welcome! Feel free to:

• Report issues.
• Suggest improvements.
• Submit pull requests.

License

This project is licensed under the MIT License.

Acknowledgments

Shoutout wgpu and Rust for making cool stuff.