A collaborative repository dedicated to providing a curated collection of high-performance GPU kernels across various computing frameworks. The goal is to create a readily available resource for developers to accelerate their applications by reusing optimized kernel implementations for common tasks.
Writing efficient GPU kernels from scratch can be time-consuming and complex. This repository aims to simplify that process by offering a growing library of pre-written, tested, and functioning kernels. Whether you're working with OpenCL, CUDA, or other GPU programming paradigms, you should be able to find and contribute reusable code snippets here.
The kernels are organized by their respective GPU computing framework:
OpenCL/: Contains OpenCL C kernels (.clfiles) and their associated Python-based test suite (using PyOpenCL).OpenCL/tests/python/: Houses the Python testing framework for OpenCL kernels. Refer to its dedicatedREADME.mdfor details on creating and running tests.
CUDA/: (Future/Planned) Will contain CUDA C/C++ kernels (.cufiles) and their respective testing frameworks.[Other Frameworks]/: (Future/Planned) Additional directories for other GPU computing frameworks like SYCL, Vulkan Compute, etc., as contributions grow.
Each framework directory will contain the kernel files themselves and, where applicable, a tests/ subdirectory with language-specific test runners and examples.
- Diverse Kernel Collection: A growing set of GPU kernels for common computational tasks (e.g., element-wise operations, matrix multiplication, reductions).
- Framework Agnostic Organization: Kernels are categorized by their programming framework for easy navigation.
- Reusable Code: Designed for direct integration into your projects.
- Comprehensive Testing: Each kernel framework aims to have its own robust testing infrastructure to ensure correctness and performance.
- Community-Driven: Open for contributions from developers worldwide.
For detailed instructions on how to create, run, and manage test cases for the OpenCL kernels, please refer to the specific README.md located at:
OpenCL Tests README
This separate README.md covers:
- Setting up the PyOpenCL testing environment.
- The structure of OpenCL kernel test files.
- How to use
generate_test_mapping.pyandtest_runner.py.
We welcome contributions from the community! If you have an optimized GPU kernel you'd like to share, or if you want to improve existing ones, please refer to our CONTRIBUTING.md for guidelines.
This project is licensed under the MIT License - see the LICENSE file for details.
Start exploring and contributing to faster GPU computing today!