Parallel Programming with Intel® Parallel Studio XE

Chapter 16

Parallelizing Legacy Code

What's in This Chapter?

Introducing the Dhrystone benchmark

Parallelizing the C version

Parallelizing the C++ version

One of the biggest challenges programmers face is making existing code parallel. The older and bigger the code, the more difficult the task is. Successful parallelism should lead to improved performance and scalability without having to make wholesale changes to the code.

Legacy code represents the huge investment of time and effort that programmers have made. For most programmers, maintaining and modifying legacy code is a significant headache, because it often includes the following characteristics:

• Large, monolithic code base
• Unknown or misunderstood content
• Old-style programming
• Ubiquitous use of global variables

This case study begins by introducing the Dhrystone benchmark, and then explores adding parallelism to the code using OpenMP and Cilk Plus. The hands-on activities give you an opportunity to try out the different approaches. You explore the following ways of addingparallelism to the benchmark:

• Synchronizing shared variable access
• Duplicating global variables
• Wrapping the application in a C++ class
• Using Cilk Plus holders

Taking the original serial version of the Dhrystone benchmark, the first three attempts at parallelization use OpenMP, and the last attempt uses Cilk Plus. In all these attempts, the following questions are addressed:

• Will adding parallelism improve the performance of the code?
• Can the parallelism ...