Contents

• Motivation
• The Julia programming language
• Getting started with Julia

Motivation

• An abstract FE framework for cell morphogenesis
  • To accommodate all the FE simulation needs in the lab
  • Easily import or prototype new FE technology
  • In Julia, as a subpackage of Gridap

Borja da Rocha, Bleyer et Turlier, submitted

• Since 2012
• General purpose
• Focus in high-performance
• Well-suited for numerical analysis and computational science
• v1.7.2 (Feb 8 2022)
• v1.8 (very soon)

https://julialang.org/

Why Julia?

To answer this, we need to review
high-level compiled and interpreted programming languages

medium.com, Young Coder, The Difference Between Compiled and Interpreted Languages, Sep 14, 2020

medium.com, Young Coder, The Difference Between Compiled and Interpreted Languages, Sep 14, 2020

medium.com, Young Coder, The Difference Between Compiled and Interpreted Languages, Sep 14, 2020

For computationally-demanding applications (e.g., FEs):

• "Hybrid" software applications:
  • E.g., python front-end + c++ back-end
  • two-language problem > limited extensibility and prototyping
• Julia potentially provides better balance
  • performance and productivity in a single environment

Main features of Julia

• Optional typing
• Multiple dispatch (generalisation of polymorphism)

• Multi-paradigm
  • Procedural, functional, meta, object-oriented...

Main features of Julia

• Performance through
  • type inference (deduce output types from input types)
    • generate fast, optimized code
  • just-in-time (JIT) or run-time compilation
    • user experience as an intepreted one
• The only JIT-compiled language in the petaflop $\footnotesize{\left(\mathsf{10^{15}}\right)}$ club
  • Since 2017; along with C, C++ and Fortran

Main features of Julia

• Time to first plot or compiler latency
  • JIT compilation time on first run

• Two minor releases per year (+3-5 patch releases per minor)
  • Beware of breaking changes and compiler latency regressions

Julia vs Python

• Python has never been so popular (in Feb 2022)
  • #1 in TIOBE index (search engine results)
  • #1 in IEEE Spectrum index
  • #1 in PYPL index
  • ...
• Julia way behind (#20-30). Growing, but not emerging
• Often performance not needed or enough with GPU parallelism

Julia vs Python in machine learning

Numbers speak for themselves....

Julia vs Python in machine learning

Numbers speak for themselves.... BUT this is not the whole story!

TensorFlow	PyTorch	Flux.jl

My case for Julia

• It risks becoming a niche language for scientific computing
• But it could become THE standard in this domain
  • The only petascale-capable JIT-compiled language
  • Improved balance of productivity and performance
    • e.g. in basic research on finite elements
• Further reading: independent benchmarks
  • Julia vs C gcc -- similar performance, C is best --
  • Julia vs Python 3 -- Julia much faster --

• Download and install
• Differences from Python/C/C++
• The Julia package manager
• Introductory tutorials
• Style and performance tips
• Noteworthy packages
• VS Code as Julia IDE
• The Julia Community

Download and install Julia

Download Julia

• MAC specific instructions
• Linux specific instructions

Check supported platforms

Julia comes with a full-featured interactive command-line REPL (read-eval-print loop) built into the julia executable

Documentation

Julia is well-documented

Julia documentation

It comprises an excellent manual

Manual / Getting started USE IT!

Fast-track to Julia from Python/C/C++

• Noteworthy differences from Python and C/C++
  • Array indexing in Julia is 1-based not 0-based
  • Julia arrays are column major (Fortran ordered)
  • Julia has no classes (data + methods) and no multiple inheritance
  • ...
• Julia Cheat Sheet
• Matlab-Python-Julia Cheat Sheet (outdated)

The Julia package manager, Pkg.jl

• Integrated into the REPL (Pkg documentation)
• Manages reproducible environments, not only packages
  • An environment is a record of
    • Package dependencies and compatibilities in Project.toml
    • Package state (pkg version, git revision, ...) in Manifest.toml
• If a Julia project ($\approx$ library) contains a manifest, this will install the packages in the same state that is given by that manifest

Set up the working environment

• Work in projects or packages
• Creating packages
  • With Pkg.jl
  • With PkgTemplates.jl For software projects
    • e.g., set up Github Actions (continuous integration)
  • With DrWatson.jl For scientific projects
    • e.g., manage and reproduce parametric simulations

Introductory tutorials

• JuliaTutorials
  • Jupyter notebooks (e.g., 00, 01, 06, 10, 10.1 and 09)
• Julia track in Exercism
  • Solve coding exercises and get mentored
• General and domain-specific courses
  • Introduction to Julia ➔ Videos for JuliaTutorials
  • Foundations of Machine Learning

How to develop your code in Julia?

• Follow test-driven development
• Option 1: Combine your editor of choice and the REPL
  • REPL-based workflows Use Revise.jl
  • Add Revise.jl in the global (default) environment
• Option 2: Use VS Code with Julia extension (THE Julia IDE)
  • VS Code workflows
  • Matlab-ish experience (e.g., workspace browser, debugger...)

How to write your code in Julia?

• Style guide
  • Append ! to names of functions that modify their arguments
    • e.g., sort does not mutate, sort! mutates
  • Prefer exported methods over direct field access
    • data members of a type are not hidden
  • Apply naming and argument ordering similar to Julia base
  • ...

How to write your code in Julia?

• Performance tips
  • Performance critical code should be inside a function
    • Julia's compiler works better on code inside functions
    • Avoid global variables
  • Write "type-stable" functions (results in faster code)
  • Measure performance and memory footprint with @time

• Use BenchmarkTools.jl and ProfileView.jl

More on VS Code as Julia IDE

• Quite powerful
  • Built-in debugger
  • Native support for Jupyter notebooks (e.g., JuliaTutorials)
  • Embedded results (e.g., plots, profiler, databases)
  • New features every year (announced in JuliaCon)
• Collaborative sessions
  • A "Google Docs" experience with integrated audio and text chat

• Scientific machine learning
  • FluxML (GitHub)
    • Several JuliaCon vids like this one
  • SciML (GitHub)
    • Julia SciML Symposium at SIAM CSE 2021
• Image processing
  • JuliaImages (GitHub)
  • DeconvOptim.jl (incipient)

Parallel computing in Julia

• Multi-threading is "Open-MP"-style and built in the core library
• Multi-processing and distributed computing
  • With standard module Distributed or non-standard MPI.jl, ...
• GPU computing
  • Workshop on GPU programming (e.g., CUDA.jl)
• Julia joins petaflop club $\footnotesize{\mathsf{1.54}}$ petaflops / $\footnotesize{\mathsf{1.3·10^6}}$ threads / $\footnotesize{\mathsf{9.3}}$k cores
• Poisson FE problem up to $\footnotesize{\mathsf{15}}$k cores in GridapDistributed.jl

Other interesting packages

• DataFrames.jl: fast tabular data manipulation
• Pluto.jl: reactive Jupyter notebooks
• Makie.jl: interactive plots
• DrWatson.jl: scientific projects management

Community

• 35 million downloads as of Jan 2022
• Full statistics Jan 2022
• JuliaCon: free and virtual
  • 2022 edition in July 27-29
  • 2021 edition
    • 43,000 unique viewers
    • > 300 presentations on YouTube

Gridap.jl in JuliaCon 2021

Getting help

• Julia Discourse
  • Go-to place
  • Read this before posting
  • Stack Exchange and Stack Overflow not so active
• Julia Slack
  • Dedicated channels (#gridap, #machine-learning, ...)
• Julia Manual ALWAYS

Announcements

• SciMLCon 2022
  • free and virtual on Mar 23 15:00-23:00 CET
• Google Summer Of Code 2022
  • Funded and mentored online "interships" in an open source organization
  • Applications open April 4 - 19

And that's all! Thank you!