1 |
|
|
/* $OpenBSD: qsort.c,v 1.18 2017/05/30 14:54:09 millert Exp $ */ |
2 |
|
|
/*- |
3 |
|
|
* Copyright (c) 1992, 1993 |
4 |
|
|
* The Regents of the University of California. All rights reserved. |
5 |
|
|
* |
6 |
|
|
* Redistribution and use in source and binary forms, with or without |
7 |
|
|
* modification, are permitted provided that the following conditions |
8 |
|
|
* are met: |
9 |
|
|
* 1. Redistributions of source code must retain the above copyright |
10 |
|
|
* notice, this list of conditions and the following disclaimer. |
11 |
|
|
* 2. Redistributions in binary form must reproduce the above copyright |
12 |
|
|
* notice, this list of conditions and the following disclaimer in the |
13 |
|
|
* documentation and/or other materials provided with the distribution. |
14 |
|
|
* 3. Neither the name of the University nor the names of its contributors |
15 |
|
|
* may be used to endorse or promote products derived from this software |
16 |
|
|
* without specific prior written permission. |
17 |
|
|
* |
18 |
|
|
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
19 |
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
20 |
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
21 |
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
22 |
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
23 |
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
24 |
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
25 |
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
26 |
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
27 |
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
28 |
|
|
* SUCH DAMAGE. |
29 |
|
|
*/ |
30 |
|
|
|
31 |
|
|
#include <sys/types.h> |
32 |
|
|
#include <stdlib.h> |
33 |
|
|
|
34 |
|
|
static __inline char *med3(char *, char *, char *, int (*)(const void *, const void *)); |
35 |
|
|
static __inline void swapfunc(char *, char *, size_t, int); |
36 |
|
|
|
37 |
|
|
#define min(a, b) (a) < (b) ? a : b |
38 |
|
|
|
39 |
|
|
/* |
40 |
|
|
* Qsort routine from Bentley & McIlroy's "Engineering a Sort Function". |
41 |
|
|
* |
42 |
|
|
* This version differs from Bentley & McIlroy in the following ways: |
43 |
|
|
* 1. The partition value is swapped into a[0] instead of being |
44 |
|
|
* stored out of line. |
45 |
|
|
* |
46 |
|
|
* 2. The swap function can swap 32-bit aligned elements on 64-bit |
47 |
|
|
* platforms instead of swapping them as byte-aligned. |
48 |
|
|
* |
49 |
|
|
* 3. It uses David Musser's introsort algorithm to fall back to |
50 |
|
|
* heapsort(3) when the recursion depth reaches 2*lg(n + 1). |
51 |
|
|
* This avoids quicksort's quadratic behavior for pathological |
52 |
|
|
* input without appreciably changing the average run time. |
53 |
|
|
* |
54 |
|
|
* 4. Tail recursion is eliminated when sorting the larger of two |
55 |
|
|
* subpartitions to save stack space. |
56 |
|
|
*/ |
57 |
|
|
#define SWAPTYPE_BYTEV 1 |
58 |
|
|
#define SWAPTYPE_INTV 2 |
59 |
|
|
#define SWAPTYPE_LONGV 3 |
60 |
|
|
#define SWAPTYPE_INT 4 |
61 |
|
|
#define SWAPTYPE_LONG 5 |
62 |
|
|
|
63 |
|
|
#define TYPE_ALIGNED(TYPE, a, es) \ |
64 |
|
|
(((char *)a - (char *)0) % sizeof(TYPE) == 0 && es % sizeof(TYPE) == 0) |
65 |
|
|
|
66 |
|
|
#define swapcode(TYPE, parmi, parmj, n) { \ |
67 |
|
|
size_t i = (n) / sizeof (TYPE); \ |
68 |
|
|
TYPE *pi = (TYPE *) (parmi); \ |
69 |
|
|
TYPE *pj = (TYPE *) (parmj); \ |
70 |
|
|
do { \ |
71 |
|
|
TYPE t = *pi; \ |
72 |
|
|
*pi++ = *pj; \ |
73 |
|
|
*pj++ = t; \ |
74 |
|
|
} while (--i > 0); \ |
75 |
|
|
} |
76 |
|
|
|
77 |
|
|
static __inline void |
78 |
|
|
swapfunc(char *a, char *b, size_t n, int swaptype) |
79 |
|
|
{ |
80 |
✗✗✗✓ ✗ |
1338582 |
switch (swaptype) { |
81 |
|
|
case SWAPTYPE_INT: |
82 |
|
|
case SWAPTYPE_INTV: |
83 |
|
|
swapcode(int, a, b, n); |
84 |
|
|
break; |
85 |
|
|
case SWAPTYPE_LONG: |
86 |
|
|
case SWAPTYPE_LONGV: |
87 |
✓✓ |
1338638 |
swapcode(long, a, b, n); |
88 |
|
669291 |
break; |
89 |
|
|
default: |
90 |
|
|
swapcode(char, a, b, n); |
91 |
|
|
break; |
92 |
|
|
} |
93 |
|
669291 |
} |
94 |
|
|
|
95 |
|
|
#define swap(a, b) do { \ |
96 |
|
|
switch (swaptype) { \ |
97 |
|
|
case SWAPTYPE_INT: { \ |
98 |
|
|
int t = *(int *)(a); \ |
99 |
|
|
*(int *)(a) = *(int *)(b); \ |
100 |
|
|
*(int *)(b) = t; \ |
101 |
|
|
break; \ |
102 |
|
|
} \ |
103 |
|
|
case SWAPTYPE_LONG: { \ |
104 |
|
|
long t = *(long *)(a); \ |
105 |
|
|
*(long *)(a) = *(long *)(b); \ |
106 |
|
|
*(long *)(b) = t; \ |
107 |
|
|
break; \ |
108 |
|
|
} \ |
109 |
|
|
default: \ |
110 |
|
|
swapfunc(a, b, es, swaptype); \ |
111 |
|
|
} \ |
112 |
|
|
} while (0) |
113 |
|
|
|
114 |
|
|
#define vecswap(a, b, n) if ((n) > 0) swapfunc(a, b, n, swaptype) |
115 |
|
|
|
116 |
|
|
static __inline char * |
117 |
|
|
med3(char *a, char *b, char *c, int (*cmp)(const void *, const void *)) |
118 |
|
|
{ |
119 |
✓✓ |
3709944 |
return cmp(a, b) < 0 ? |
120 |
✓✓ |
1895792 |
(cmp(b, c) < 0 ? b : (cmp(a, c) < 0 ? c : a )) |
121 |
✓✓ |
886666 |
:(cmp(b, c) > 0 ? b : (cmp(a, c) < 0 ? a : c )); |
122 |
|
|
} |
123 |
|
|
|
124 |
|
|
static void |
125 |
|
|
introsort(char *a, size_t n, size_t es, size_t maxdepth, int swaptype, |
126 |
|
|
int (*cmp)(const void *, const void *)) |
127 |
|
|
{ |
128 |
|
|
char *pa, *pb, *pc, *pd, *pl, *pm, *pn; |
129 |
|
|
int cmp_result; |
130 |
|
2075452 |
size_t r, s; |
131 |
|
|
|
132 |
✓✓ |
1707013 |
loop: if (n < 7) { |
133 |
✓✓ |
4028042 |
for (pm = a + es; pm < a + n * es; pm += es) |
134 |
✓✓✓✓
|
3563850 |
for (pl = pm; pl > a && cmp(pl - es, pl) > 0; |
135 |
|
|
pl -= es) |
136 |
✗✓✗ |
481328 |
swap(pl, pl - es); |
137 |
|
1037432 |
return; |
138 |
|
|
} |
139 |
✓✓ |
669581 |
if (maxdepth == 0) { |
140 |
✓✗ |
292 |
if (heapsort(a, n, es, cmp) == 0) |
141 |
|
292 |
return; |
142 |
|
|
} |
143 |
|
669289 |
maxdepth--; |
144 |
|
669289 |
pm = a + (n / 2) * es; |
145 |
✓✓ |
669289 |
if (n > 7) { |
146 |
|
|
pl = a; |
147 |
|
636810 |
pn = a + (n - 1) * es; |
148 |
✓✓ |
636810 |
if (n > 40) { |
149 |
|
96892 |
s = (n / 8) * es; |
150 |
|
96892 |
pl = med3(pl, pl + s, pl + 2 * s, cmp); |
151 |
|
96892 |
pm = med3(pm - s, pm, pm + s, cmp); |
152 |
|
96892 |
pn = med3(pn - 2 * s, pn - s, pn, cmp); |
153 |
|
96892 |
} |
154 |
|
636810 |
pm = med3(pl, pm, pn, cmp); |
155 |
|
636810 |
} |
156 |
✗✓✗ |
1338578 |
swap(a, pm); |
157 |
|
669289 |
pa = pb = a + es; |
158 |
|
669289 |
pc = pd = a + (n - 1) * es; |
159 |
|
1270081 |
for (;;) { |
160 |
✓✓✓✓
|
37333798 |
while (pb <= pc && (cmp_result = cmp(pb, a)) <= 0) { |
161 |
✓✓ |
11176233 |
if (cmp_result == 0) { |
162 |
✗✓✗ |
234 |
swap(pa, pb); |
163 |
|
117 |
pa += es; |
164 |
|
117 |
} |
165 |
|
11176233 |
pb += es; |
166 |
|
|
} |
167 |
✓✓✓✓
|
21209314 |
while (pb <= pc && (cmp_result = cmp(pc, a)) >= 0) { |
168 |
✓✓ |
6446147 |
if (cmp_result == 0) { |
169 |
✗✓✗ |
50 |
swap(pc, pd); |
170 |
|
25 |
pd -= es; |
171 |
|
25 |
} |
172 |
|
6446147 |
pc -= es; |
173 |
|
|
} |
174 |
✓✓ |
1270081 |
if (pb > pc) |
175 |
|
|
break; |
176 |
✗✓✗ |
1201584 |
swap(pb, pc); |
177 |
|
600792 |
pb += es; |
178 |
|
600792 |
pc -= es; |
179 |
|
|
} |
180 |
|
|
|
181 |
|
669289 |
pn = a + n * es; |
182 |
|
669289 |
r = min(pa - a, pb - pa); |
183 |
✓✓ |
1338564 |
vecswap(a, pb - r, r); |
184 |
✗✓ |
2007867 |
r = min(pd - pc, pn - pd - es); |
185 |
✓✓ |
669305 |
vecswap(pb, pn - r, r); |
186 |
|
|
/* |
187 |
|
|
* To save stack space we sort the smaller side of the partition first |
188 |
|
|
* using recursion and eliminate tail recursion for the larger side. |
189 |
|
|
*/ |
190 |
|
|
r = pb - pa; |
191 |
|
|
s = pd - pc; |
192 |
✓✓ |
669289 |
if (r < s) { |
193 |
|
|
/* Recurse for 1st side, iterate for 2nd side. */ |
194 |
✓✗ |
82687 |
if (s > es) { |
195 |
✓✓ |
82687 |
if (r > es) { |
196 |
|
81534 |
introsort(a, r / es, es, maxdepth, |
197 |
|
|
swaptype, cmp); |
198 |
|
81534 |
} |
199 |
|
82687 |
a = pn - s; |
200 |
|
82687 |
n = s / es; |
201 |
|
82687 |
goto loop; |
202 |
|
|
} |
203 |
|
|
} else { |
204 |
|
|
/* Recurse for 2nd side, iterate for 1st side. */ |
205 |
✓✓ |
586602 |
if (r > es) { |
206 |
✓✓ |
586600 |
if (s > es) { |
207 |
|
200913 |
introsort(pn - s, s / es, es, maxdepth, |
208 |
|
|
swaptype, cmp); |
209 |
|
200913 |
} |
210 |
|
586600 |
n = r / es; |
211 |
|
586600 |
goto loop; |
212 |
|
|
} |
213 |
|
|
} |
214 |
|
1037728 |
} |
215 |
|
|
|
216 |
|
|
void |
217 |
|
|
qsort(void *a, size_t n, size_t es, int (*cmp)(const void *, const void *)) |
218 |
|
|
{ |
219 |
|
|
size_t i, maxdepth = 0; |
220 |
|
|
int swaptype; |
221 |
|
|
|
222 |
|
|
/* Approximate 2*ceil(lg(n + 1)) */ |
223 |
✓✓ |
3846597 |
for (i = n; i > 0; i >>= 1) |
224 |
|
790380 |
maxdepth++; |
225 |
|
755279 |
maxdepth *= 2; |
226 |
|
|
|
227 |
✓✗✓✗
|
1510558 |
if (TYPE_ALIGNED(long, a, es)) |
228 |
|
755279 |
swaptype = es == sizeof(long) ? SWAPTYPE_LONG : SWAPTYPE_LONGV; |
229 |
|
|
else if (sizeof(int) != sizeof(long) && TYPE_ALIGNED(int, a, es)) |
230 |
|
|
swaptype = es == sizeof(int) ? SWAPTYPE_INT : SWAPTYPE_INTV; |
231 |
|
|
else |
232 |
|
|
swaptype = SWAPTYPE_BYTEV; |
233 |
|
|
|
234 |
|
755279 |
introsort(a, n, es, maxdepth, swaptype, cmp); |
235 |
|
|
|
236 |
|
755279 |
} |
237 |
|
|
|
238 |
|
|
DEF_STRONG(qsort); |