HeteroPar'2012 Program

Session I Chair: Rosa M. Badia
9:30-10:30 Invited Speaker: Enrique Quintana, Universitat Jaume I, Spain Unleashing CPU-GPU Acceleration for Control Theory Applications Abstract: Automatic control systems play critical roles in many fields, e.g. manufacturing, electronics, communication, transportation, biology, and medicine. However, the solution of real control problems typically implies very sophisticate methods, to meet the numerical precision and time restrictions. In this talk we will show how model order reduction, an important control theory application, which required the use of a moderate size cluster even for a moderate dynamical system only a few years ago, can nowadays be easily solved using an optimized algorithm for a hybrid CPU-GPU platform. Using dense matrix inversion, a key operation to the solution of the certain matrix equations arising in model order reduction, we will review techniques common to the optimization of dense linear algebra kernels as, e.g, concurrent computation on both CPU and GPU, overlapping computation with communication, reduction of PCI transfers, and static look-ahead. We will then move to more advanced strategies, including dynamic scheduling, mixed precision and iterative refinement, out-of-core computing, and energy-aware computing on this class of platforms.
10:30-11:00 Albano Alves, José Rufino, António Pina and Luís Santos dOpenCL - Supporting Distributed Heterogeneous Computing in HPC Clusters
Coffee Break
Session II Chair: Enrique Quintana
11:30-12:00 Richard Membarth, Frank Hannig, Jürgen Teich, Mario Körner and Wieland Eckert Mastering Software Variant Explosion for GPU Accelerators
12:00-12:30 Artur Podobas, Mats Brorsson and Vladimir Vlassov Exploring Heterogeneous Scheduling using the Task-Centric Programming model
12:30-13:00 Hartwig Anzt, Stanimire Tomov, Jack Dongarra and Vincent Heuveline Weighted Block-Asynchronous Iteration on GPU-Accelerated Systems BEST PAPER!
Lunch Break
Session III Chair: Alexey Lastovetsky
14:30-15:00 Biao Wang, Mauricio Alvarez-Mesa, Chi Ching Chi and Ben Juurlink An Optimized Parallel IDCT on Graphics Processing Units
15:00-15:30 Svetislav Momcilovic, Nuno Roma and Leonel Sousa Multi-level Parallelization of Advanced Video Coding on Hybrid CPU+GPU Platforms
15:30-16:00 Irina Demeshko, Satoshi Matsuoka, Naoya Maruyama and Hirofumi Tomita Multi-GPU implementation of the NICAM atmospheric model
Coffee Break
Session IV Chair: Rosa M. Badia
16:30-17:00 Kiril Dichev and Alexey Lastovetsky MPI vs BitTorrent : switching between large-message broadcast algorithms in the presence of bottleneck links
17:00-17:30 Hector Blanco De Frutos, Fernando Guirado, Josep Lluis Lérida and Victor M. Albornoz MIP Model Scheduling for Multi-Clusters
17:30-18:00 Ahmad Abdelfattah, David Keyes and Hatem Ltaief Systematic Approach in Optimizing Numerical Memory-Bound Kernels on GPU