Seminar Series - The SpiceC Parallel Programming System

Speaker: Min Feng, University of California at Riverside
Date: Monday, February 20, 2012
Time: 11:15am-12:15pm
Location: 110 McBryde

Abstract:
This talk provides an overview of my work on SpiceC, a system that simplifies the task of program parallelization through a combination of SpiceC directives and an intuitive computation model. To support various application types, the SpiceC directives allow programmers to express different forms of parallelism. Speculation is also supported to enable exploitation of dynamic parallelism. Additional directives enable efficient parallelization in the presence of dynamic linked data structures and I/O operations. The SpiceC computation model offers software-managed memory isolation and data transfer between threads. Therefore, it can be adapted to a wide range of parallel architectures, including shared-memory systems (e.g., multicores and manycores), distributed-memory systems (e.g., clusters), and heterogeneous systems (e.g., GPGPUs). I have successfully parallelized substantial applications with SpiceC such as Velvet, Delaunay Refinement, and Incremental Tree Inducer. The base SpiceC implementation achieves speedups ranging from 2x to 18x by parallelizing computational loops for seven benchmarks on a 24-core system. With loop parallelization in the presence of I/O, SpiceC further improves the performance of eight applications by 68% on average in comparison to the parallelized versions of the same applications where only purely computational loops are parallelized. With data partitioning, SpiceC also achieves 1.3x--6.9x speedups for a set of benchmarks that use dynamic data structures on an 8-core system.

Bio:
Min Feng is currently a Ph.D. student in the Department of Computer Science and Engineering at the University of California at Riverside. His advisor is Professor Rajiv Gupta. His areas of research interest include compilers and architectures for parallel systems, and software tools for debugging. He received a Bachelor's degree from Tongji University, China, and a Master's degree from City University of Hong Kong.