Event
Ph.D. Dissertation Defense: Fady A. Ghanim
Wednesday, November 2, 2016
9:45 a.m.-12:15 p.m.
Room 2460 - AVW Bldg.
Maria Hoo
301 405 3681
mch@umd.edu
ANNOUNCEMENT: Ph.D. Dissertation Defense
Name: Fady A. Ghanim
Advisory Committee:
Professor Rajeev Barua, Chair/Advisor
Professor Uzi Vishkin, Co-Advisor
Professor Bruce Jacob
Professor Manoj Franklin
Professor Alan Sussman, Dean's Representative
Date/Time: Wednesday Nov, 2nd 2016 | 9:45am - 12:15pm
Advisory Committee:
Professor Rajeev Barua, Chair/Advisor
Professor Uzi Vishkin, Co-Advisor
Professor Bruce Jacob
Professor Manoj Franklin
Professor Alan Sussman, Dean's Representative
Date/Time: Wednesday Nov, 2nd 2016 | 9:45am - 12:15pm
Location: Room 2460 - AVW
Title: Easy PRAM-Based High-Performance Parallel Programming
Abstract:
Parallel machines have become more widely used. Unfortunately parallel programming technologies have advanced at a much slower pace except for regular programs. For irregular programs, this advancement is inhibited by high synchronization costs, non-loop parallelism, non-array data structures, recursively expressed parallelism and parallelism that is too fine-grained to be exploitable.
Title: Easy PRAM-Based High-Performance Parallel Programming
Abstract:
Parallel machines have become more widely used. Unfortunately parallel programming technologies have advanced at a much slower pace except for regular programs. For irregular programs, this advancement is inhibited by high synchronization costs, non-loop parallelism, non-array data structures, recursively expressed parallelism and parallelism that is too fine-grained to be exploitable.
This work introduces ICE, a new parallel programming language that is easy-to-program, since: (i) ICE is a synchronous, lock-step language; (ii) that allows direct transcription of PRAM algorithms, and (iii) the PRAM algorithmic theory offers unique wealth of parallel algorithms and techniques. This work suggests that ICE can be a part of an ecosystem consisting of the XMT architecture, the PRAM algorithmic model, and ICE itself, that together deliver on the twin goal of easy programming and efficient parallelization of irregular programs. The XMT architecture, developed at UMD, can exploit fine-grained parallelism in irregular programs. This work also introduces a compiler that translates the lock-step ICE language into the multithreaded XMTC language while maintaining comparable performance to highly optimized XMTC code, a significant contribution because multi-threading is a feature shared by practically all current scalable parallel programming languages. The work is evaluated by comparing ICE to manually optimized for performance XMTC programs. It shows that ICE allows easier programming of PRAM algorithms, as indicated by a 35.5% reduction in the number of lines of code on average observed over 11 out of 16 benchmarks, while maintaining comparable run-time performance across all benchmarks.
