Parallel Programming with Intel® Parallel Studio XE

Chapter 6

Where to Parallelize

What's in This Chapter?

Hotspot analysis using the Intel compiler

Hotspot analysis using the auto-parallelizer

Hotspot analysis using Amplifier XE

The purpose of parallelization is to improve the performance of an application. Performance can be measured either by how much time a program takes to run or by how much work a program can do per second. Within a program, it is the busy sections, or hotspots, that should be made parallel. The more the hotspots contribute to the overall run time of the program, the better the performance improvement you will obtain by parallelizing them.

Hotspot analysis is an important first step in the parallelism process. This chapter shows three different ways to identify hotspots in your code using Parallel Studio XE. Without carrying out Hotspot analysis, there is a danger that you will end up making little or no difference to your program's performance. The section “Hotspot Analysis Using the Auto-Parallelizer” includes some tips on how to help the auto-parallelizer do its job better.

A Note for Linux Users

Most of the text of this chapter uses the Windows version of the compiler options. You can use the option-mapping tool to find the equivalent Linux option. The following example finds the Linux equivalent of /Oy-:

map_opts -tl -lc -opts /Oy- Intel(R) Compiler option mapping tool mapping Windows options to Linux for C++ ‘-Oy-’ Windows option maps to --> ‘-fomit-frame-pointer-’ option on Linux --> ‘-fno-omit-frame-pointer’ ...