2 * A fast, small, non-recursive O(nlog n) sort for the Linux kernel
4 * Jan 23 2005 Matt Mackall <mpm@selenic.com>
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/types.h>
10 #include <linux/export.h>
11 #include <linux/sort.h>
13 static int alignment_ok(const void *base, int align)
15 return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
16 ((unsigned long)base & (align - 1)) == 0;
19 static void u32_swap(void *a, void *b, int size)
22 *(u32 *)a = *(u32 *)b;
26 static void u64_swap(void *a, void *b, int size)
29 *(u64 *)a = *(u64 *)b;
33 static void generic_swap(void *a, void *b, int size)
39 *(char *)a++ = *(char *)b;
45 * sort - sort an array of elements
46 * @base: pointer to data to sort
47 * @num: number of elements
48 * @size: size of each element
49 * @cmp_func: pointer to comparison function
50 * @swap_func: pointer to swap function or NULL
52 * This function does a heapsort on the given array. You may provide a
53 * swap_func function optimized to your element type.
55 * Sorting time is O(n log n) both on average and worst-case. While
56 * qsort is about 20% faster on average, it suffers from exploitable
57 * O(n*n) worst-case behavior and extra memory requirements that make
58 * it less suitable for kernel use.
61 void sort(void *base, size_t num, size_t size,
62 int (*cmp_func)(const void *, const void *),
63 void (*swap_func)(void *, void *, int size))
65 /* pre-scale counters for performance */
66 int i = (num/2 - 1) * size, n = num * size, c, r;
69 if (size == 4 && alignment_ok(base, 4))
71 else if (size == 8 && alignment_ok(base, 8))
74 swap_func = generic_swap;
78 for ( ; i >= 0; i -= size) {
79 for (r = i; r * 2 + size < n; r = c) {
82 cmp_func(base + c, base + c + size) < 0)
84 if (cmp_func(base + r, base + c) >= 0)
86 swap_func(base + r, base + c, size);
91 for (i = n - size; i > 0; i -= size) {
92 swap_func(base, base + i, size);
93 for (r = 0; r * 2 + size < i; r = c) {
96 cmp_func(base + c, base + c + size) < 0)
98 if (cmp_func(base + r, base + c) >= 0)
100 swap_func(base + r, base + c, size);