There was a C library - posted here but I cannot remember the name - which used the preprocessor to avoid invoking a function call per comparison. You'd use it something like this:

#define SORT_ELEMENT_TYPE int

#define SORT_COMPARE(a, b) sign(a - b)

#include "specialized_sort.h"

And the header would define a function with the following signature:

int sort_int(int[] a, size_t n);

Why is this limited to the preprocessor? I imagine the compiler proper could inline the compare function parameter into the qsort implementation body in the typical case, at the qsort() call site. (Assuming qsort's implementation was in a header, as the original comment posited).

If both the comparison function as well as the sort function which the comparison function is passed to are declared 'static inline' and available in a header, the compiler is smart enough to figure it out and inline the code, provided optimizations are turned on.

Does the necessary pointer casting in the implementations also make it slower than it should be (the specialized versions generated by compilers that support generics wouldn't have any casting)?