Workshop-descriptions | Lemieux's Lab

Metaprogramming

Fri, 19 Dec 2025 00:00:00 +0000

Sometimes, we need to work with the code syntax directly, treating it as data. This comes up often for logging, benchmarking and monitoring (think of @time or @btime from BenchmarkTools and @showprogress from ProgressMeter), adding extra features to existing code without rewriting it (think of @threads, @async), changing the meaning of some code structures (with @assert to make something a unit test) or cheeky compiler tricks (@turbo, which promises the compiler your code doesn’t index out of bounds, iterate over empty collections, order doesn’t matter, etc…)

We want to write things like this: @log_step x + y * z

And get outputs like this: [LOG] x + y * z = 42

This is behaviour that’s difficult to emulate with a function.

Downloadable content

Notebook: julia | html | pdf

Nvidia Nsight

Fri, 14 Nov 2025 00:00:00 +0000

Serena Chan

Nsight Systems is a system-wide performance analysis that allow us to capture activity across the CPU, GPU, and OS! It uses low-overhead tracing/sampling to find bottlenecks: it can’t exactly tell you why there are slowdowns occurring, but where and when they occur.

Downloadable content

Notes: pdf | pptx

RTF report: nsys-rep

Model Optimization

Fri, 17 Oct 2025 00:00:00 +0000

Léa Kaufmann

Optimizing your model’s parameters with wandb on Compute Canada.

Downloadable content

Notes: pdf

Multithreading

Fri, 08 Aug 2025 00:00:00 +0000

Nicolas Jacquin

Multithreading allows a program to run multiple computations at the same time on separate CPU cores. This is especially useful for CPU-bound tasks where raw computing power is the bottleneck.

In Julia, multithreading is enabled with the environment variable JULIA_NUM_THREADS, and used via macros like Threads.@threads and Threads.@spawn.

The goal is to split work into smaller, independent parts that can be safely computed in parallel.

We’ll also talk about asynchronous programming, which defines coroutines running at the same time on a single thread, and able to yield control to each other while they wait for something.

Downloadable content

Notebook: julia | pdf | html

Debugging

Wed, 23 Apr 2025 00:00:00 +0000

Serena Chan

Testing and comparing debugging and profiling tools in Julia.

Downloadable content

Notes: pdf

Julia script: dbg.jl

Data files: features.jld2 | labels.jld2

Advance Typing

Wed, 12 Mar 2025 00:00:00 +0000

Nicolas Jacquin

Typing matters!

In general, type ambiguity forces the compiler to jump through many hoops to make sure it can accomodate the ambiguity. All of this adds extra overhead, which you should never pay unless you have absolutely no choice (or optimization is not a concern).

Downloadable content

Notebook: julia | html

GPU Kernels

Thu, 06 Feb 2025 00:00:00 +0000

Serena Chan | Carl Munoz

In Julia, GPU usage is already optimized for many processes through CUDA.jl. Simply by applying a function to a CuArray, operations (e.g. broadcasting and map-reducing) are executed on GPU-specialized code. Additionally, more complex tasks, such as operations in machine learning algorithms like self-attention, have optimized code through cuDNN.jl.

These are done through GPU kernels, which implement functions that exploit the CUDA architecture of GPUs.

Today, we will show how to write such kernels, for when already-optimized kernels do not already exist.

Downloadable content

Notebook: julia | pdf | html

Environement files: Manifest.toml | Project.toml

Julia scripts: demo_sandbox.jl | softmax_examples

Web Scrapping

Tue, 01 Oct 2024 00:00:00 +0000

Nicolas Jacquin

What is Web Scrapping?

Extracting data from website (usually what’s displayed on a webpage)
Used to automate repetitive processes that require checking or interacting with websites
Data is either extracted from the HTML (easy), but sometimes snooping around the requests done by a webpage in the background is required to get the data you need (Sometimes time consuming, consider if the time saved is worth it!)

Downloadable content

Notebook: html