SYCL Implementation of Nvidia's OneSweep

Welcome to the SYCL implementation of Nvidia's OneSweep! This project is an open-source, high-performance implementation designed to efficiently sort 1 billion 32-bit key/value pairs using Nvidia's RTX 4090 GPU.

Page: https://m-gjerde.github.io/SYCLOneSweep

Highlights

• Blazing Fast Performance: Achieves sorting of 1 billion (2^30) 32-bit key/value pairs in just 0.17 seconds (5.88 GKeys/s) on Nvidia RTX 4090. (Slightly slower than cuda implementation, but optimizations can still be done)
• Open Source: Completely free to use and modify under the MIT license.
• SYCL Compatibility: Leverages SYCL for cross-platform compatibility and future-proofing.
• Efficient Resource Usage: Optimized to make the most of GPU resources without requiring proprietary CUDA libraries.

Getting Started

Prerequisites

• SYCL-compatible compiler (e.g., Intel DPC++, Codeplay ComputeCpp)
• CMake (for building the project)

Installation

1. Clone the repository:

   git clone https://github.com/M-Gjerde/SYCLOneSweep
   cd SYCLOneSweep
   mkdir build
   cd build
   cmake ..
   make
   ./onesweep_sycl
   
   

License

This project is licensed under the MIT License. See the LICENSE file for more details.

Special Thanks

A special thanks to the b0nes164/OneSweep repository for providing a CUDA library-less implementation that inspired and facilitated this project. https://github.com/b0nes164/OneSweep