Highly Abstracted Video Processing Language Meets Multi-Grain Auto-Parallelization

Multi-core and SIMD-supported processors have been popular and now can be highly utilized for image/video processing. However, to exploit theoretical performance of such a platform, developers must write programs, considering detailed specifications of the platform. We have proposed a video processing library RaVioli for solving this issue. RaVioli conceals two resolutions, frame rate and number of pixels, from developers for simplifying image/video processing programming. However, image/video processing programs could be written with more abstract syntax than that RaVioli provides, and image/video processing has much more potential parallelism than that RaVioli can exploit. To solve these problems, in this paper, we propose a multi-grain parallel video processing environment composed of a highly abstracted programming language and a dedicated compiler. The environment makes image/video processing programming easier and achieves higher performance.