By Samuel Belko, published on May 28, 2024.
I would like to share some good resources that helped me improve my coding skills and the overall experience with Julia. From what I know so far, it is a good idea to invest time into understanding the type system and multiple dispatch, since it is the way idiomatic & performant Julia code is written.
SciML Book: optimizing serial code
some of the topics covered
memory related speedups
avoid cache misses
column major indexing of matrices
stack vs heap
mutating a cache variable in place to avoid new allocations
using views instead of slicing an array
fusing loop operations with broadcasting to avoid allocations of temporary variables
function specialization related speedups
type inference inside functions & type stability
compiler is building a different function for different combinations of input types that is specialized and only works for that type combination
type information is compile time information and values are runtime information, types needs to provide enough information to the compiler to generate efficient, specialized lowered code
multiple dispatch as a natural extension of just-in-time compilation
parametric types can create a family of isbits types
abstract types don't help the compiler (but are useful for steering multiple dispatch)
function barriers for dealing with type instability (one dynamic dispatch call after which only static dispatch calls follow)
Hands-On Design Patterns and Best Practices with Julia
some of the topics I found helpful
semantic versioning
dependency inversion principle
function forwarding & extending types by wrapping them
rules for automatic conversion between data types
array of structs vs struct of arrays
the holy traits pattern, singleton type dispatch
type piracy anti-pattern (extending a third-party function without using own types in any of the function arguments)
contains many good pointers to libraries and setups for a better coding experience, in particular