Ayman Al Baz

Numpy is considered the go-to library for fast performant python scientifc computing, as it is a heavily optimized scientific computing library written in C. But is it really that fast (spoiler: yes, but not when used with python)? In this talk, I will be covering a real-world example in the domain of genetics, and demonstrating how I was able to out-perform numpy code running on my intel-7700k machine with a raspberry pi running code from the Nim programming language!

I will also be providing optimization tips for speeding up your code, with a focus on comparing python with the Nim programming language.

The talk can be roughly divided into the following sections:

Part 1 (32 min) - Introduction to myself. Describing a computationally expensive problem that I have experienced when working with Python. - My background is in data science and biology - Biologists tend to work with large datasets - One common problem we face is preprocessing of large amounts of data. A common technique used here using distance calculations - Why I decided to learn a different language Part 2 (1 min) - High level introduction to Nim - Nim is a statically compiled programming language - Etc... Part 3 (10 min) - Going through step-by-step on how to write performance optimal code with Nim. Describing common performance pitfalls that Python users may not be aware of. - Inlining - Using the type system to your advantage - Taking advantage of domain knowledge - Multi-threading (openmp) - Bounds checking - Compiler flags - Mathematical and algorithmic optimizations Part 4 (3 min) - Benchmarks and closing statement - Show benchmark comparing the naive python numpy implementation with the highly optimized Nim one - Top tips to keep in mind when writing high performance code.

Please note: the purpose of this talk isn't to be an exploration of some new state-of-the-art way of doing performance optimization or some relatively unknown optimization techniques. This is a general introduction for people hoping to use Nim for HPC that are coming from Python. Lots of emphasis will be put comparing the two languages.