I consider the problem of the domain-specific optimization of programs. I review different approaches, discuss their potential, and sketch instances of them from the practice of high-performance parallelism. Readers need not be familiar with high-performance computing.