Short Course on OpenMP in Practice

Contents Organizer and Instructor Rationale Description Level of Material Course Objectives Who Should Attend Recommended Background Instructor Program Locations How to Register Questions? E-mail [email protected]   SIAM Conferences

Organizer and Instructor Charles Koelbel CRPC, Rice University and Advanced Computational Research Program, National Science Foundation Rationale OpenMP is an emerging standard for programming parallel computers, particularly those with shared-memory architectures. It includes many features intended to simplify the programming task and allow high-level parallelism, including: Parallel loop directives Task parallel constructs PRIVATE and SHARED data declarations Because most features of OpenMP are extensions of standard sequential language constructs, OpenMP is particularly valuable for incrementally converting existing sequential codes for parallel execution. Because it is so new, however, there is little experience in getting the most out of OpenMP on practical problems. OpenMP will be an important system for constructing parallel scientific programs for the near-term future. Applied mathematicians will, however, need training to use this exciting system. This course emphasizes practical techniques that can be used in those programs. Description This course introduces OpenMP, an emerging standard for programming shared-memory parallel computers. Most of the time will be devoted to the key features of the language: PARALLEL DO - allows parallel loops over data PARALLEL SECTIONS - allows task parallelism PRIVATE / SHARED - control access to data Approximately half the course will be devoted to extended practical examples. One example will be the conversion of a sequential code to parallel form, while others will show how efficient parallel programs can be constructed from first principles. Level of Material Introductory 50% Intermediate 25% Advanced 25% Course Objectives Attendees will learn: Generally useful parallel programming concepts Specific constructs for programming shared-memory machines Useful techniques for optimizing parallel programs for machines with complex memory hierarchies Who Should Attend Computational scientists who need to begin writing scientific programs for shared-memory machines.The course may also be of interest to computer scientists who need information on current trends in parallel processing. Recommended Background Essential: Programming experience in Fortran or C Useful: Knowledge of parallel computing concepts (architecture, data- and task-parallelism, compilers) Instructor Chuck Koelbel has been working at Rice University on compilers and languages for parallel machines since graduating from Purdue in 1990. He is best-known as one of the creators of High Performance Fortran. He is currently a rotator at NSF as the program director for Advanced Computational Research. Program Morning 8:00 AM Registration 9:00 AM-10:00 AM I. Introduction to Parallel Computing   A. Parallel Computers B. Parallel Languages & Systems C. Why OpenMP? II. Basic Parallel Features   A. Control structures B. Data environment C. Synchronization D. Run-time environment and library 10:00 AM-10:30 Coffee 10:30 AM-12:30 PM III. Designing Programs in OpenMP   A. Basics B. Developing a Parallel Code I C. Converting a Sequential Code D. Optimization issues IV. (Time Permitting) Recent OpenMP Developments   A. Combining with HPF B. Combining with MPI C. Available Debuggers and Tools 12:30 PM-2:00 PM Lunch (For short course registrants only) 2:00 PM Short course adjourns Location The short course on OpenMP in Practice will take place in Executive Salon 3 on the third floor. Coffee breaks will be in Lobby, third floor and lunch will be in Rose Garden Room. How to Register Seats are limited. We urge short course participants to register in advance. To register, please complete the Preregistration Form online and send it with your payment to reach SIAM office on or before March 3, 1999. Registration fee for the short course includes: course notes, final program and abstracts, coffee breaks, and lunch on Sunday, March 21.