Preprocessor and compiler utilization

There are several levels of optimization that give the compiler different degrees of freedom in instruction rearrangement.

The programmer who wants to obtain the highest possible performance from a given program running on a given machine must deal with several considerations:

• There are preprocessors that can rearrange some source code structures to form a functionally equivalent source module that can be compiled into more efficient executable code.
• Just as there are several variants of the architecture, there are several compiler options to allow optimal compilation for a specific variant or set of variants.
• The programmer can use the #pragma feature to inform the C compiler of certain aspects of the program that will allow the compiler to generate more efficient code by relaxing some of its worst-case assumptions.

Programmers who are unable to experiment, should always optimize. The difference in performance between optimized and unoptimized code is almost always so large that basic optimization (the -O option of the compiler commands) should always be used. The only exceptions are testing situations in which there is a specific need for straightforward code generation, such as statement-level performance analysis using the tprof tool.

These techniques yield additional performance improvement for some programs, but the determination of which combination yields the best performance for a specific program might require considerable recompilation and measurement.

For an extensive discussion of the techniques for efficient use of compilers, see Optimization and Tuning Guide for XL Fortran, XL C and XL C++.