krastanov/julia-perf

Julia Performance Optimization

Diagnose and fix performance issues in Julia code following the official Performance Tips.

Diagnosis Workflow

1. Profile first — identify the actual bottleneck before optimizing.
2. Check type stability — run @code_warntype on hot functions.
3. Measure allocations — use @time/@btime to find unexpected allocations.
4. Apply targeted fixes — use the references below for specific patterns.

using BenchmarkTools, Profile

# Step 1: Measure
@btime my_function($args)

# Step 2: Check type stability
@code_warntype my_function(args)

# Step 3: Profile
@profile my_function(args)
Profile.print(noisefloor=2.0)

# Step 4: Track allocations
julia --track-allocation=user -e 'using MyPkg; my_function(args); Profile.clear_malloc_data(); my_function(args)'

Top Rules

1. Put hot code in functions — never at global/script scope.
2. Avoid untyped globals — use const or pass as arguments.
3. Use concrete types everywhere — in structs, containers, return values.
4. Pre-allocate and mutate — use ! functions to reuse buffers.
5. Access arrays in column-major order — first index varies fastest.
6. Fuse broadcasts — use @. or dot syntax to avoid temporaries.

Reference

• Type Stability — Type inference, stable returns, function barriers
• Structs & Dispatch — Concrete fields, parametric types, Val
• Memory & Arrays — Pre-allocation, views, column-major, broadcasting
• Annotations & Tweaks — @inbounds, @fastmath, @simd, misc tips
• Profiling & Diagnosis — @code_warntype, profiling, allocation tracking

Related Skills

• julia-jet - JET.jl static analysis overview
• julia-bench - Quick benchmarks, suites, and benchmark CI