Introduction to OpenMP programming

Parallel

Article from ADMIN 12/2012

By Jeff Layton

Parallel programming is not easy, but one tool you can use to help parallelize your application is OpenMP. Most compilers are compatible with OpenMP and allow you to parallelize your code on a single node.

Almost any developer of HPC programs will tell you that parallel programming is not easy. Unfortunately, no magic -parallel option exists, so typically you have to dive into the details with the goal of improving the performance of your application. This means really understanding the algorithm, the code that performs the algorithm, and the hardware on which you are going to run the final application.

Today, it's very easy to get laptops with at least two, if not four, cores. Desktops can easily have eight cores with lots of memory. You can also get x86 servers with 64 cores that access all of the memory on the system. The seductive call of "more cores" is very attractive for improving performance or even running larger problems (or larger games), but it can take a great deal of work to get there.

That magic -parallel compiler option that would automagically parallelize your applications is so elusive because any arbitrary application can be parallelized in a number of ways – including such methods as defining how variables relate to one another, what data pointers represent, and the level of loop parallelism – leaving the compiler with no idea which of the many possible parallelizations is correct. What is needed is a way to guide the compiler by giving it hints about what it can parallelize and how, without having to write lots of extra code. The goal is to keep the number of changes to the source code to a minimum, so they do not affect the building of serial code, and let the compiler parallelize the application for you. This is exactly what OpenMP does.

Introduction to OpenMP

OpenMP [1], short for Open Multi-Processing, is an Application Program Interface (API) that can be used to direct explicitly multithreaded, shared memory parallelism in applications. It is a set of standard

...

Use Express-Checkout link below to read the full article (PDF).