blob: fc20df42aa6f2945aae8586c3c4d3599ca6d67f0
1 | /* |
2 | * A fast, small, non-recursive O(nlog n) sort for the Linux kernel |
3 | * |
4 | * Jan 23 2005 Matt Mackall <mpm@selenic.com> |
5 | */ |
6 | |
7 | #include <linux/types.h> |
8 | #include <linux/export.h> |
9 | #include <linux/sort.h> |
10 | |
11 | static int alignment_ok(const void *base, int align) |
12 | { |
13 | return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || |
14 | ((unsigned long)base & (align - 1)) == 0; |
15 | } |
16 | |
17 | static void u32_swap(void *a, void *b, int size) |
18 | { |
19 | u32 t = *(u32 *)a; |
20 | *(u32 *)a = *(u32 *)b; |
21 | *(u32 *)b = t; |
22 | } |
23 | |
24 | static void u64_swap(void *a, void *b, int size) |
25 | { |
26 | u64 t = *(u64 *)a; |
27 | *(u64 *)a = *(u64 *)b; |
28 | *(u64 *)b = t; |
29 | } |
30 | |
31 | static void generic_swap(void *a, void *b, int size) |
32 | { |
33 | char t; |
34 | |
35 | do { |
36 | t = *(char *)a; |
37 | *(char *)a++ = *(char *)b; |
38 | *(char *)b++ = t; |
39 | } while (--size > 0); |
40 | } |
41 | |
42 | /** |
43 | * sort - sort an array of elements |
44 | * @base: pointer to data to sort |
45 | * @num: number of elements |
46 | * @size: size of each element |
47 | * @cmp_func: pointer to comparison function |
48 | * @swap_func: pointer to swap function or NULL |
49 | * |
50 | * This function does a heapsort on the given array. You may provide a |
51 | * swap_func function optimized to your element type. |
52 | * |
53 | * Sorting time is O(n log n) both on average and worst-case. While |
54 | * qsort is about 20% faster on average, it suffers from exploitable |
55 | * O(n*n) worst-case behavior and extra memory requirements that make |
56 | * it less suitable for kernel use. |
57 | */ |
58 | |
59 | void sort(void *base, size_t num, size_t size, |
60 | int (*cmp_func)(const void *, const void *), |
61 | void (*swap_func)(void *, void *, int size)) |
62 | { |
63 | /* pre-scale counters for performance */ |
64 | int i = (num/2 - 1) * size, n = num * size, c, r; |
65 | |
66 | if (!swap_func) { |
67 | if (size == 4 && alignment_ok(base, 4)) |
68 | swap_func = u32_swap; |
69 | else if (size == 8 && alignment_ok(base, 8)) |
70 | swap_func = u64_swap; |
71 | else |
72 | swap_func = generic_swap; |
73 | } |
74 | |
75 | /* heapify */ |
76 | for ( ; i >= 0; i -= size) { |
77 | for (r = i; r * 2 + size < n; r = c) { |
78 | c = r * 2 + size; |
79 | if (c < n - size && |
80 | cmp_func(base + c, base + c + size) < 0) |
81 | c += size; |
82 | if (cmp_func(base + r, base + c) >= 0) |
83 | break; |
84 | swap_func(base + r, base + c, size); |
85 | } |
86 | } |
87 | |
88 | /* sort */ |
89 | for (i = n - size; i > 0; i -= size) { |
90 | swap_func(base, base + i, size); |
91 | for (r = 0; r * 2 + size < i; r = c) { |
92 | c = r * 2 + size; |
93 | if (c < i - size && |
94 | cmp_func(base + c, base + c + size) < 0) |
95 | c += size; |
96 | if (cmp_func(base + r, base + c) >= 0) |
97 | break; |
98 | swap_func(base + r, base + c, size); |
99 | } |
100 | } |
101 | } |
102 | |
103 | EXPORT_SYMBOL(sort); |
104 | |
105 | #if 0 |
106 | #include <linux/slab.h> |
107 | /* a simple boot-time regression test */ |
108 | |
109 | int cmpint(const void *a, const void *b) |
110 | { |
111 | return *(int *)a - *(int *)b; |
112 | } |
113 | |
114 | static int sort_test(void) |
115 | { |
116 | int *a, i, r = 1; |
117 | |
118 | a = kmalloc(1000 * sizeof(int), GFP_KERNEL); |
119 | BUG_ON(!a); |
120 | |
121 | printk("testing sort()\n"); |
122 | |
123 | for (i = 0; i < 1000; i++) { |
124 | r = (r * 725861) % 6599; |
125 | a[i] = r; |
126 | } |
127 | |
128 | sort(a, 1000, sizeof(int), cmpint, NULL); |
129 | |
130 | for (i = 0; i < 999; i++) |
131 | if (a[i] > a[i+1]) { |
132 | printk("sort() failed!\n"); |
133 | break; |
134 | } |
135 | |
136 | kfree(a); |
137 | |
138 | return 0; |
139 | } |
140 | |
141 | module_init(sort_test); |
142 | #endif |
143 |