Conference site ยป Proceedings

PyStream: Compiling Python onto the GPU

Nick Bray
Google

Abstract

PyStream is a static compiler that can radically transform Python code and run it on a Graphics Processing Unit (GPU). Python compiled to run on the GPU is \textasciitilde{}100,000x faster than when interpreted on the CPU. The PyStream compiler is specially designed to simplify the development of real-time rendering systems by allowing the entire rendering system to be written in a single, highly productive language. Without PyStream, GPU-accelerated real-time rendering systems must contain two separate code bases written in two separate languages: one for the CPU and one for the GPU. Functions and data structures are not shared between the code bases, and any common functionality must be redundantly written in both languages. PyStream unifies a rendering system into a single, Python code base, allowing functions and data structures to be transparently shared between the CPU and the GPU. A single, unified code base makes it easy to create, maintain, and evolve a high-performance GPU-accelerated application.

Keywords

pystream, compiling python, gpu

Bibtex entry

Full text PDF