ilist-sorting-performance.md

IList Sorting Performance Benchmarks

Unity Helpers ships several custom sorting algorithms for IList<T> that cover different trade-offs between adaptability, allocation patterns, and stability. This page gathers context and benchmark snapshots so you can choose the right algorithm for your workload and compare results across operating systems.

Algorithm Cheatsheet

Algorithm	Stable?	Best For	Reference
Ghost Sort	No	Mixed workloads that benefit from adaptive gap sorting and few allocations	Upstream project by Will Stafford Parsons (public repository currently offline)
Meteor Sort	No	Almost-sorted data where gap shrinking beats plain insertion sort	Upstream project by Will Stafford Parsons (public repository currently offline)
Pattern-Defeating QuickSort	No	General-purpose quicksort with protections against worst-case inputs	pdqsort by Orson Peters
Grail Sort	Yes	Large datasets where stability + low allocations matter	GrailSort
Power Sort	Yes	Partially ordered data that benefits from adaptive run detection	PowerSort (Munro & Wild)
Tim Sort	Yes	General-purpose stable sorting with abundant natural runs	Wikipedia - Timsort
Jesse Sort	No	Data with long runs or duplicates where dual patience piles shine	JesseSort
Green Sort	Yes	Sustainable stable merges that trim ordered prefixes	greeNsort
Ska Sort	No	Branch-friendly partitioning on large unstable datasets	Ska Sort
Ipn Sort	No	In-place adaptive quicksort scenarios needing robust pivots	ipnsort write-up
Smooth Sort	No	Weak-heap hybrid that approaches O(n) for presorted data	Smoothsort - Wikipedia
Block Merge Sort	Yes	Stable merges with √n buffer (WikiSort style)	WikiSort
IPS⁴o Sort	No	Cache-aware samplesort with multiway partitioning	IPS⁴o paper
Power Sort Plus	Yes	Enhanced run-priority merges inspired by Wild & Nebel	PowerSort paper
Glide Sort	Yes	Stable galloping merges from the Rust glidesort research	sort-research-rs
Flux Sort	No	Dual-pivot quicksort tuned for modern CPUs	sort-research-rs
Insertion Sort	Yes	Tiny or nearly sorted collections where O(n²) is acceptable	Wikipedia - Insertion sort

What does “stable” mean? Stable sorting algorithms preserve the relative order of elements that compare as equal. This matters when items carry secondary keys (e.g., sorting people by last name but keeping first-name order deterministic). Unstable algorithms can reshuffle equal entries, which is usually fine for numeric keys but can break deterministic pipelines.

Heads up: The original Ghost Sort repository was formerly hosted on GitHub under wstaffordp/ghostsort, but it currently returns 404. The Unity Helpers implementation remains based on that source; we will relink if/when an official mirror returns.

Dataset Scenarios

• Sorted – ascending integers, verifying best-case behavior.
• Nearly Sorted (2% swaps) – deterministic neighbor swaps introduce light disorder to expose adaptive optimizations.
• Shuffled (deterministic) – Fisher–Yates shuffle using a fixed seed for reproducibility across runs and machines.

Each benchmark sorts a fresh copy of the dataset once and reports wall-clock duration. If a cell still shows pending, re-run the benchmark suite to collect fresh data for that algorithm/dataset size.

Run the IListSortingPerformanceTests.Benchmark test inside Unity’s Test Runner to refresh the tables below. Results automatically land in the section that matches the current operating system.

Windows (Editor/Player)

Last updated 2026-04-22 22:08 UTC on Windows 11 (10.0.26200) 64bit

Times are single-pass measurements in milliseconds (lower is better). n/a indicates the algorithm was skipped for the dataset size.

Sorted

List Size	Ghost	Meteor	Pattern-Defeating QuickSort	Grail	Power	Insertion	Tim	Jesse	Green	Ska	Ipn	Smooth	Block	IPS4o	Power+	Glide	Flux
100	0.004 ms	0.001 ms	0.001 ms	0.001 ms	0.001 ms	0.001 ms	0.001 ms	0.093 ms	0.001 ms	0.006 ms	0.001 ms	0.002 ms	0.001 ms	0.001 ms	0.001 ms	0.001 ms	0.002 ms
1,000	0.022 ms	0.024 ms	0.006 ms	0.006 ms	0.005 ms	0.005 ms	0.005 ms	1.16 ms	0.006 ms	0.095 ms	0.007 ms	0.016 ms	0.005 ms	0.031 ms	0.004 ms	0.005 ms	0.026 ms
10,000	0.261 ms	0.343 ms	0.063 ms	0.068 ms	0.040 ms	0.052 ms	0.038 ms	14.4 ms	0.059 ms	1.36 ms	0.063 ms	0.161 ms	0.045 ms	0.513 ms	0.040 ms	0.040 ms	0.341 ms
100,000	3.04 ms	4.46 ms	0.632 ms	0.639 ms	0.393 ms	n/a	0.379 ms	192 ms	0.591 ms	17.6 ms	0.654 ms	1.64 ms	0.438 ms	6.22 ms	0.390 ms	0.383 ms	4.38 ms

Nearly Sorted (2% swaps)

List Size	Ghost	Meteor	Pattern-Defeating QuickSort	Grail	Power	Insertion	Tim	Jesse	Green	Ska	Ipn	Smooth	Block	IPS4o	Power+	Glide	Flux
100	0.001 ms	0.002 ms	0.002 ms	0.001 ms	0.002 ms	0.001 ms	0.001 ms	0.118 ms	0.001 ms	0.006 ms	0.001 ms	0.002 ms	0.001 ms	0.002 ms	0.002 ms	0.002 ms	0.002 ms
1,000	0.020 ms	0.024 ms	0.028 ms	0.007 ms	0.015 ms	0.006 ms	0.012 ms	1.21 ms	0.006 ms	0.096 ms	0.025 ms	0.016 ms	0.006 ms	0.039 ms	0.022 ms	0.014 ms	0.027 ms
10,000	0.253 ms	0.342 ms	0.371 ms	0.069 ms	0.228 ms	0.055 ms	0.168 ms	14.3 ms	0.062 ms	1.38 ms	0.340 ms	0.167 ms	0.060 ms	0.595 ms	0.361 ms	0.159 ms	0.352 ms
100,000	3.10 ms	4.54 ms	4.47 ms	0.726 ms	3.27 ms	n/a	2.25 ms	186 ms	0.642 ms	17.3 ms	4.09 ms	1.69 ms	0.548 ms	7.03 ms	5.16 ms	2.14 ms	4.40 ms

Shuffled (deterministic)

List Size	Ghost	Meteor	Pattern-Defeating QuickSort	Grail	Power	Insertion	Tim	Jesse	Green	Ska	Ipn	Smooth	Block	IPS4o	Power+	Glide	Flux
100	0.008 ms	0.006 ms	0.005 ms	0.006 ms	0.007 ms	0.015 ms	0.009 ms	0.041 ms	0.006 ms	0.008 ms	0.006 ms	0.010 ms	0.005 ms	0.005 ms	0.022 ms	0.010 ms	0.006 ms
1,000	0.144 ms	0.120 ms	0.084 ms	0.099 ms	0.104 ms	1.28 ms	0.132 ms	0.442 ms	0.101 ms	0.127 ms	0.091 ms	0.179 ms	0.083 ms	0.101 ms	0.422 ms	0.143 ms	0.094 ms
10,000	1.95 ms	1.75 ms	1.14 ms	1.35 ms	1.43 ms	128 ms	1.48 ms	5.54 ms	1.36 ms	1.72 ms	1.23 ms	2.56 ms	1.13 ms	1.44 ms	5.94 ms	1.69 ms	1.38 ms
100,000	28.4 ms	23.8 ms	14.4 ms	17.3 ms	17.3 ms	n/a	19.1 ms	69.7 ms	17.1 ms	22.3 ms	15.1 ms	32.2 ms	14.3 ms	18.3 ms	75.0 ms	22.1 ms	17.1 ms

macOS

Pending — run the IList sorting benchmark suite on macOS to capture results.

Linux

Pending — run the IList sorting benchmark suite on Linux to capture results.

Other Platforms

Pending — run the IList sorting benchmark suite on the target platform to capture results.