Compiling Code for High-Performance Computing

Numpy and the compiled programs it depends on are extremely performant, if allowed to do their work; however, many performance issues in scientific Python code stem from high overhead between the Python interpreter and compiled scientific libraries during performance-critical calculations. In this unit, we look at how to reduce this overhead, how to compile both simple and complex algorithms into simple operations, and how to compile code for a variety of other processor types, including GPUs.