Parallel Programming in OpenMP

Aimed at researchers who work or scientific C / C + + or Fortran programmers, Parallel Programming in OpenMP two standards describe what and how to use them to create software that takes full advantage of parallel computing. At its core, OpenMP is very simple. Techniques explored include how to tweak the default direction paralleled to a certain situation, how to use the parallel region (beyond simple loop), and the dos and don'ts of effective synchronization (with critical sections and barriers). The book is completed with some advice very good for how to work with the current system cache mechanism OpenMP-compliant. Overall, Parallel Programming in OpenMP introduce appropriate research programmer for a new vocabulary of idioms and techniques for parallelizing software using OpenMP -. Richard Dragan Topics covered: Overview of OpenMP programming standard for shared-memory multiprocessors Description OpenMP directives for Fortran OpenMP parallel hardware and pragmas for C / C + + Parallelizing simple do-loop parallel / parallel to the direction of shared and private scoping for variable reduction surgery thread data dependence and how to remove them OpenMP performance problems (work fairly, balancing the load in the loop, the scheduling option) Parallel mode area parallelize arbitrary blocks of code (master and slave threads, and paragraph copyin threadprivate direction) parallel work queue task number Divide by sharing thread Restrictions on work-sharing Noniterative Orphaning work Nested parallel regions in OpenMP parallelism control, including controlling the number of threads, dynamic threads, and OpenMP library calls for synchronization of data OpenMP threads Avoiding Critical race landing part (nested critical sections and atomic directive name and OpenMP Runtime library lock routines Event synchronization (barrier landing and part ordered) Custom synchronization, including the directive flush Programming tips for the performance of synchronization problems with the Act load balancing's Amdahl OpenMP for manual code parallelized to write code paralleled that matches the processor cache Avoiding false synchronization instructions Performance sharing problem for bus-based and Non-Uniform Memory Access (NUMA) machine quick reference OpenMP.

Download Link