Performance work in scientific computing is a discipline of measurement, not guesswork. This section covers the two halves of that discipline.
- Optimization Concepts — what an optimizing compiler does with a program: inlining, loop transformations, vectorization, and how optimization levels trade compile time for speed.
- A Measurement-First Workflow — timing, profiling, finding hot spots, and keeping a baseline so every change is judged against evidence.
For a gentler introduction, start with Fortran Performance: The Basics. The Parallel Programming section continues the story where a single core stops being enough.