On: Blitsort is an in-place stable adaptive rotate merge sort

[acmj]

I am thinking more about sorting, say, 20-byte or 24-byte custom structs. Directly sorting arrays is probably faster than sorting pointers to arrays. At least in my applications, I rarely just sort numbers; I more often sort numbers along with other data associated with the numbers.

[scandum → acmj]

Hard to say which would be faster, I've made a mental note to benchmark sorting 16 byte long doubles through pointers as well as directly.

I never tried, but it should be possible (and relatively easy) to add custom sizes in the .h file.