13 - 18 OCTOBER 2013, ATLANTA, GEORGIA, USA

UberShadie: A Domain-Specific Language for General Volume Processing and Visualization on Heterogeneous Parallel Systems

Contributors: 
Hyungseok Choi, Hanspeter Pfister, Won-Ki Jeong
Description
In this paper, we introduce our on-going work on developing a domain-specific language specifically designed for 3D data processing and visualization on heterogeneous parallel computing systems. Our method is inspired by the previous work by Hasan et al., (Shadie [5]). We observed that the language design resembling Python is important for novice users, and the high-level abstraction of GPU programming is another advantage of Shadie. However, we also observed that the shader-like framework in Shadie becomes a major obstacle that significantly impairs the flexibility of the system. Based on these observations, we propose to develop a more advanced, flexible, and easy-to-use programming language, compiler, and runtime system, called UeberShadie, specifically designed to easily write a research code handling 3D volume data while providing superior computing performance by leveraging the state-of-the-art heterogeneous parallel computing technology, such as multicore CPUs and GPUs (Graphics Processing Units).