Compiler Optimization

When writing numerical codes, we often want the fastest performance we can get. Compilers can make optimizations to our code to get better performance. This can include eliminating common subexpressions, unrolling loops to get better cache performance, inlining functions to reduce the overhead of a call, and utilizing special hardware instructions.

Optimization often requires multiple passes from the compiler, which makes the compilation take longer, but the resulting code can be much faster.

Optimization flags

Most compilers use -O to specify optimization, and the amount of optimization can be controlled using -O1 to the least optimization to -O3 for aggressive optimization. At the highest level of optimization, the compiler may transform math in a way that is algebraically equivalent but has different roundoff.

Tip

You can see what the different optimization levels do by looking at man g++.

NDEBUG

By default, the C++ standard library has asserts that check at runtime that the library is being used as intended. These can be disabled by passing the flag -DNDEBUG to g++.