high performance computing on graphics processing units: hgpu.org

Nowadays, the main advances in computational power are due to parallelism. However, most parallel languages have been designed with a focus on processors and threads. This makes dealing with data and memory in programs hard, which distances the implementation from its original algorithm. We propose a new paradigm for parallel programming, the data-autonomous paradigm, where computation is performed by autonomous data elements. Programs in this paradigm are focused on making the data collaborate in a highly parallel fashion. We furthermore present AuDaLa, the first data autonomous programming language, and include an operational semantics. Programming in AuDaLa is very natural, as illustrated by examples, albeit in a style very different from sequential and contemporary parallel programming.