murphyqm / swd6_hpp

Training materials for the SWD 6 High Performance Python workshop

High Performance Python

• What are the motivations for profiling and optimising code?
• The perceived problems with Python; interpreted vs. compiled languages
• The Python language vs. the Python environment: Python, CPython and PyPy

First session

• Profiling code- how to identify the 'problem areas' or potential slow-running areas in code
• Speeding up code 1: Ahead-Of-Time (AOT) compilation with Cython
• Speeding up code 2: Just-In-Time (JIT) compilation with numba
• Speeding up code 3: parallel programming with multiprocessing and moving code to clusters
• Speeding up code 4: An alternative Python- pypy

Second session

• Speeding up code 5: The benefits of numpy
• Performance differences between Python lists, arrays and Numpy arrays
• Introduction to Computer architecture: why caches, threads and processes are important
• The Python GIL (Global Interpreter Lock) and why this is important
• Design choices- choosing the right algorithm for the job (big O notation)
• Final thoughts and other ideas:
  • Why not just write it in C?
  • Use a specialist language to write native functions, eg. Chapel for parallel programming or Julia for general purpose performant code.