There's a sorting algorithm who really knows where his towel is

FroodSort

FroodSort is an adaptive sorting algorithm that automatically selects between various sorting strategies based on input characteristics, with a O(1) algorithm selection overhead:

• O(n) counting sort as a sane default and for dense integer ranges
• Best-case O(n) insertion sort for small arrays and nearly-sorted data
• O(n) radix sort for very large ranges of integers
• O(n log n) std::sort as fallback

This algorithm is optimized for integer arrays with small ranges relative to size, nearly-sorted sequences, quirks of various datatypes, and modern CPU architectures. It's still a work in progress as I'm not familiar with library development and type gymnastics in c++. Also, there's probably a better implementation out there I haven't seen; this is mostly for my learning purposes anyway.

Benchmarks

Compiled with g++ -O3 -Wall .\froodsort.cpp, the current version outputs something similar to this:

Nearly sorted array of 10000000 ints:

Choosing counting sort
Adaptive sort: 300.75ms
std::sort:    1498.94ms

Random array of 10000000 ints, range 1-1M):

Choosing counting sort
Adaptive sort: 264.94ms
std::sort:    2601.44ms

BigType1 array of 100000 elements

Falling back to std::sort
Adaptive sort: 42.22ms
std::sort:    36.35ms
1 1 1 1 1

Nearly sorted array of 10000000 floats

Falling back to std::sort
Adaptive sort: 1411.46ms
std::sort:    1336.72ms

Small range array of 10000000 ints, range 1-100):

Choosing counting sort
Adaptive sort: 209.65ms
std::sort:    778.12ms

Large array with wide range (20000000 elements, range 0 to 1073741823):

Choosing radix sort
Adaptive sort: 2788.92ms
std::sort:    6291.99ms
First few elements after sorting: 43 106 290 308 324 

• These benchmarks seem too good to be true, more testing is required.

• Large datatypes are pretty slow and fall back to std::sort, need to look into this.