git.saurik.com Git - apple/libc.git/blob

1 /*-

4 *

5 * This code is derived from software contributed to Berkeley by

6 * Peter McIlroy.

7 *

8 * Redistribution and use in source and binary forms, with or without

9 * modification, are permitted provided that the following conditions

10 * are met:

11 * 1. Redistributions of source code must retain the above copyright

12 * notice, this list of conditions and the following disclaimer.

13 * 2. Redistributions in binary form must reproduce the above copyright

14 * notice, this list of conditions and the following disclaimer in the

15 * documentation and/or other materials provided with the distribution.

16 * 4. Neither the name of the University nor the names of its contributors

17 * may be used to endorse or promote products derived from this software

18 * without specific prior written permission.

19 *

20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND

21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE

22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE

23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE

24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL

25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS

26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)

27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT

28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY

29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF

30 * SUCH DAMAGE.

31 */

33 #pragma clang diagnostic push

34 #pragma clang diagnostic ignored "-Wcomma"

36 #if defined(LIBC_SCCS) && !defined(lint)

37 static char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94";

38 #endif /* LIBC_SCCS and not lint */

39 #include <sys/cdefs.h>

40 __FBSDID("$FreeBSD: src/lib/libc/stdlib/merge.c,v 1.8 2007/01/09 00:28:10 imp Exp $");

42 /*

43 * Hybrid exponential search/linear search merge sort with hybrid

44 * natural/pairwise first pass. Requires about .3% more comparisons

45 * for random data than LSMS with pairwise first pass alone.

46 * It works for objects as small as two bytes.

47 */

49 #define NATURAL

50 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */

52 /* #define NATURAL to get hybrid natural merge.

53 * (The default is pairwise merging.)

54 */

56 #include <sys/types.h>

58 #include <errno.h>

59 #include <stdlib.h>

60 #include <string.h>

 static void setup(u_char *, u_char *, size_t, size_t,

     int (*)(const void *, const void *));

 static void insertionsort(u_char *, size_t, size_t,

     int (*)(const void *, const void *));

67 #define ISIZE sizeof(int)

68 #define PSIZE sizeof(u_char *)

69 #define ICOPY_LIST(src, dst, last) \

70 do \

71 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \

72 while(src < last)

73 #define ICOPY_ELT(src, dst, i) \

74 do \

75 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \

76 while (i -= ISIZE)

78 #define CCOPY_LIST(src, dst, last) \

79 do \

80 *dst++ = *src++; \

81 while (src < last)

82 #define CCOPY_ELT(src, dst, i) \

83 do \

84 *dst++ = *src++; \

85 while (i -= 1)

87 /*

88 * Find the next possible pointer head. (Trickery for forcing an array

89 * to do double duty as a linked list when objects do not align with word

90 * boundaries.

91 */

92 /* Assumption: PSIZE is a power of 2. */

93 #define EVAL(p) (u_char **) \

94 ((u_char *)0 + \

95 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))

97 /*

98 * Arguments are as for qsort.

99 */

100 int

 mergesort(base, nmemb, size, cmp)

102 void *base;

103 size_t nmemb;

104 size_t size;

         int (*cmp)(const void *, const void *);

106 {

107 size_t i;

108 int sense;

109 int big, iflag;

         u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;

         u_char *list2, *list1, *p2, *p, *last, **p1;

112

         if (size < PSIZE / 2) {         /* Pointers must fit into 2 * size. */

114 errno = EINVAL;

115 return (-1);

116 }

117

         if (nmemb == 0)

119 return (0);

120

121 /*

122 * XXX

123 * Stupid subtraction for the Cray.

124 */

125 iflag = 0;

         if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))

127 iflag = 1;

128

         if ((list2 = malloc(nmemb * size + PSIZE)) == NULL)

130 return (-1);

131

132 list1 = base;

         setup(list1, list2, nmemb, size, cmp);

134 last = list2 + nmemb * size;

135 i = big = 0;

         while (*EVAL(list2) != last) {

137 l2 = list1;

138 p1 = EVAL(list1);

             for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {

140 p2 = *EVAL(p2);

141 f1 = l2;

                 f2 = l1 = list1 + (p2 - list2);

143 if (p2 != last)

144 p2 = *EVAL(p2);

145 l2 = list1 + (p2 - list2);

                 while (f1 < l1 && f2 < l2) {

                         if ((*cmp)(f1, f2) <= 0) {

148 q = f2;

149 b = f1, t = l1;

150 sense = -1;

151 } else {

152 q = f1;

153 b = f2, t = l2;

154 sense = 0;

155 }

156 if (!big) { /* here i = 0 */

                                 while ((b += size) < t && cmp(q, b) >sense)

                                         if (++i == 6) {

159 big = 1;

160 goto EXPONENTIAL;

161 }

162 } else {

 EXPONENTIAL:                    for (i = size; ; i <<= 1)

                                         if ((p = (b + i)) >= t) {

                                                 if ((p = t - size) > b &&

                                                     (*cmp)(q, p) <= sense)

167 t = p;

168 else

169 b = p;

170 break;

                                         } else if ((*cmp)(q, p) <= sense) {

172 t = p;

173 if (i == size)

174 big = 0;

175 goto FASTCASE;

176 } else

177 b = p;

                                 while (t > b+size) {

                                         i = (((t - b) / size) >> 1) * size;

                                         if ((*cmp)(q, p = b + i) <= sense)

181 t = p;

182 else

183 b = p;

184 }

185 goto COPY;

 FASTCASE:                       while (i > size)

187 if ((*cmp)(q,

                                                 p = b + (i >>= 1)) <= sense)

189 t = p;

190 else

191 b = p;

192 COPY: b = t;

193 }

194 i = size;

195 if (q == f1) {

196 if (iflag) {

                                         ICOPY_LIST(f2, tp2, b);

                                         ICOPY_ELT(f1, tp2, i);

199 } else {

                                         CCOPY_LIST(f2, tp2, b);

                                         CCOPY_ELT(f1, tp2, i);

202 }

203 } else {

204 if (iflag) {

                                         ICOPY_LIST(f1, tp2, b);

                                         ICOPY_ELT(f2, tp2, i);

207 } else {

                                         CCOPY_LIST(f1, tp2, b);

                                         CCOPY_ELT(f2, tp2, i);

210 }

211 }

212 }

213 if (f2 < l2) {

214 if (iflag)

                                 ICOPY_LIST(f2, tp2, l2);

216 else

                                 CCOPY_LIST(f2, tp2, l2);

                 } else if (f1 < l1) {

219 if (iflag)

                                 ICOPY_LIST(f1, tp2, l1);

221 else

                                 CCOPY_LIST(f1, tp2, l1);

223 }

224 *p1 = l2;

225 }

226 tp2 = list1; /* swap list1, list2 */

227 list1 = list2;

228 list2 = tp2;

229 last = list2 + nmemb*size;

230 }

231 if (base == list2) {

                 memmove(list2, list1, nmemb*size);

233 list2 = list1;

234 }

235 free(list2);

236 return (0);

237 }

238

239 #define swap(a, b) { \

240 s = b; \

241 i = size; \

242 do { \

243 tmp = *a; *a++ = *s; *s++ = tmp; \

244 } while (--i); \

245 a -= size; \

246 }

247 #define reverse(bot, top) { \

248 s = top; \

249 do { \

250 i = size; \

251 do { \

252 tmp = *bot; *bot++ = *s; *s++ = tmp; \

253 } while (--i); \

254 s -= size2; \

255 } while(bot < s); \

256 }

257

258 /*

259 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of

260 * increasing order, list2 in a corresponding linked list. Checks for runs

261 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL

262 * is defined. Otherwise simple pairwise merging is used.)

263 */

264 void

 setup(list1, list2, n, size, cmp)

266 size_t n, size;

         int (*cmp)(const void *, const void *);

268 u_char *list1, *list2;

269 {

270 size_t i, size2;

271 int length, tmp, sense;

         u_char *f1, *f2, *s, *l2, *last, *p2;

273

274 size2 = size*2;

         if (n <= 5) {

                 insertionsort(list1, n, size, cmp);

                 *EVAL(list2) = (u_char*) list2 + n*size;

278 return;

279 }

280 /*

281 * Avoid running pointers out of bounds; limit n to evens

282 * for simplicity.

283 */

         i = 4 + (n & 1);

         insertionsort(list1 + (n - i) * size, i, size, cmp);

         last = list1 + size * (n - i);

         *EVAL(list2 + (last - list1)) = list2 + n * size;

288

289 #ifdef NATURAL

290 p2 = list2;

291 f1 = list1;

         sense = (cmp(f1, f1 + size) > 0);

         for (; f1 < last; sense = !sense) {

294 length = 2;

295 /* Find pairs with same sense. */

                 for (f2 = f1 + size2; f2 < last; f2 += size2) {

                         if ((cmp(f2, f2+ size) > 0) != sense)

298 break;

299 length += 2;

300 }

                 if (length < THRESHOLD) {               /* Pairwise merge */

302 do {

                                 p2 = *EVAL(p2) = f1 + size2 - list1 + list2;

                                 if (sense > 0)

                                         swap (f1, f1 + size);

                         } while ((f1 += size2) < f2);

307 } else { /* Natural merge */

308 l2 = f2;

                         for (f2 = f1 + size2; f2 < l2; f2 += size2) {

                                 if ((cmp(f2-size, f2) > 0) != sense) {

                                         p2 = *EVAL(p2) = f2 - list1 + list2;

                                         if (sense > 0)

                                                 reverse(f1, f2-size);

314 f1 = f2;

315 }

316 }

                         if (sense > 0)

                                 reverse (f1, f2-size);

319 f1 = f2;

                         if (f2 < last || cmp(f2 - size, f2) > 0)

                                 p2 = *EVAL(p2) = f2 - list1 + list2;

322 else

                                 p2 = *EVAL(p2) = list2 + n*size;

324 }

325 }

326 #else /* pairwise merge only. */

         for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {

                 p2 = *EVAL(p2) = p2 + size2;

                 if (cmp (f1, f1 + size) > 0)

                         swap(f1, f1 + size);

331 }

332 #endif /* NATURAL */

333 }

334

335 /*

336 * This is to avoid out-of-bounds addresses in sorting the

337 * last 4 elements.

338 */

339 static void

 insertionsort(a, n, size, cmp)

341 u_char *a;

342 size_t n, size;

         int (*cmp)(const void *, const void *);

344 {

         u_char *ai, *s, *t, *u, tmp;

346 int i;

347

         for (ai = a+size; --n >= 1; ai += size)

                 for (t = ai; t > a; t -= size) {

350 u = t - size;

                         if (cmp(u, t) <= 0)

352 break;

353 swap(u, t);

354 }

355 }

356 #pragma clang diagnostic pop