gpu.cpp is a lightweight library that simplifies portable GPU compute with C++ by leveraging the WebGPU specification. It allows for dropping in GPU code into C++ projects to run on various GPUs. The library emphasizes lightweight, fast iteration, and low boilerplate code, aiming to enable individual developers and researchers to easily incorporate GPU computation using standard C++ compilers. With a small implementation surface area and prebuilt native WebGPU implementation, gpu.cpp enables near-instantaneous C++ builds. Featured examples include a GELU kernel for neural networks, matrix multiplication, physics simulations, and SDF rendering. While gpu.cpp is not a high-level numerical computing framework, it enables portable, low-level GPU computation without the need for extensive tooling. The library is targeted towards developers familiar with C++ and GPU programming and is not meant for high-level machine learning frameworks or GPU graphics rendering.
https://github.com/AnswerDotAI/gpu.cpp