The GPU Research Center at University of Torino (UniTo) was established to support a variety of research activities employing the CUDA technology. In particular, two key research thrusts are:
1) the abstraction of GPGPU programming features and their integration into the FastFlow pattern-based/lock-free parallel programming library.
2) The evaluation of the expressiveness and of the effectiveness of parallel pattern-based languages for the porting of MonteCarlo simulations for high-energy physics onto GPGPU, and high-frequency streaming applications (e.g. signal processing, video filtering, high frequency trading).
Aldinucci believes that, developers of GPU-enabled application may benefit from a further abstraction of the state-of-the-art GPU programming tools, such as CUDA, and advocates parallel patterns, e.g. farm, pipeline, map, reduce, mapreduce, as vehicles able to bridge this abstraction gap. These patterns are not currently available as first-class abstractions neither in CUDA nor in OpenCL, and have the potentiality to improve the portability of parallel code onto GPGPU and its performance portability.
Website: GPU Research Center
Principal Investigator: Marco Aldinucci
Membri del Dipartimento: Marco Aldinucci, Maurizio Drocco, Fabio Tordini, Claudia Misale