mergesort.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361
  1. /*
  2. * mergesort() implementation for systems that don't have it.
  3. *
  4. * Copyright (c) 1992, 1993
  5. * The Regents of the University of California. All rights reserved.
  6. *
  7. * This code is derived from software contributed to Berkeley by
  8. * Peter McIlroy.
  9. *
  10. * Redistribution and use in source and binary forms, with or without
  11. * modification, are permitted provided that the following conditions
  12. * are met:
  13. * 1. Redistributions of source code must retain the above copyright
  14. * notice, this list of conditions and the following disclaimer.
  15. * 2. Redistributions in binary form must reproduce the above copyright
  16. * notice, this list of conditions and the following disclaimer in the
  17. * documentation and/or other materials provided with the distribution.
  18. * 3. Neither the name of the University nor the names of its contributors
  19. * may be used to endorse or promote products derived from this software
  20. * without specific prior written permission.
  21. *
  22. * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  23. * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  24. * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  25. * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  26. * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  27. * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  28. * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  29. * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  30. * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  31. * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32. * SUCH DAMAGE.
  33. */
  34. #include "util.h"
  35. #if defined(LIBC_SCCS) && !defined(lint)
  36. static char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94";
  37. #endif /* LIBC_SCCS and not lint */
  38. #ifdef HAVE_MERGESORT
  39. #undef yasm__mergesort
  40. #endif
  41. #ifndef HAVE_MERGESORT
  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. */
  48. #define NATURAL
  49. #define THRESHOLD 16 /* Best choice for natural merge cut-off. */
  50. /* #define NATURAL to get hybrid natural merge.
  51. * (The default is pairwise merging.)
  52. */
  53. #include <errno.h>
  54. #include <string.h>
  55. static void setup(unsigned char *, unsigned char *, size_t, size_t,
  56. int (*)(const void *, const void *));
  57. static void insertionsort(unsigned char *, size_t, size_t,
  58. int (*)(const void *, const void *));
  59. #define ISIZE sizeof(int)
  60. #define PSIZE sizeof(unsigned char *)
  61. #define ICOPY_LIST(src, dst, last) \
  62. do \
  63. *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \
  64. while(src < last)
  65. #define ICOPY_ELT(src, dst, i) \
  66. do \
  67. *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \
  68. while (i -= ISIZE)
  69. #define CCOPY_LIST(src, dst, last) \
  70. do \
  71. *dst++ = *src++; \
  72. while (src < last)
  73. #define CCOPY_ELT(src, dst, i) \
  74. do \
  75. *dst++ = *src++; \
  76. while (i -= 1)
  77. /*
  78. * Find the next possible pointer head. (Trickery for forcing an array
  79. * to do double duty as a linked list when objects do not align with word
  80. * boundaries.
  81. */
  82. /* Assumption: PSIZE is a power of 2. */
  83. #define EVAL(p) (unsigned char **) \
  84. ((unsigned char *)0 + \
  85. (((unsigned char *)p + PSIZE - 1 - (unsigned char *) 0) & ~(PSIZE - 1)))
  86. #endif /*HAVE_MERGESORT*/
  87. /*
  88. * Arguments are as for qsort.
  89. */
  90. int
  91. yasm__mergesort(void *base, size_t nmemb, size_t size,
  92. int (*cmp)(const void *, const void *))
  93. {
  94. #ifdef HAVE_MERGESORT
  95. return mergesort(base, nmemb, size, cmp);
  96. #else
  97. size_t i;
  98. int sense;
  99. int big, iflag;
  100. unsigned char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2;
  101. unsigned char *list2, *list1, *p2, *p, *last, **p1;
  102. if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */
  103. #ifdef EINVAL
  104. errno = EINVAL;
  105. #endif
  106. return (-1);
  107. }
  108. if (nmemb == 0)
  109. return (0);
  110. /*
  111. * XXX
  112. * Stupid subtraction for the Cray.
  113. */
  114. iflag = 0;
  115. if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE))
  116. iflag = 1;
  117. if ((list2 = yasm_xmalloc(nmemb * size + PSIZE)) == NULL)
  118. return (-1);
  119. list1 = base;
  120. setup(list1, list2, nmemb, size, cmp);
  121. last = list2 + nmemb * size;
  122. i = 0;
  123. big = 0;
  124. while (*EVAL(list2) != last) {
  125. l2 = list1;
  126. p1 = EVAL(list1);
  127. for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) {
  128. p2 = *EVAL(p2);
  129. f1 = l2;
  130. f2 = l1 = list1 + (p2 - list2);
  131. if (p2 != last)
  132. p2 = *EVAL(p2);
  133. l2 = list1 + (p2 - list2);
  134. while (f1 < l1 && f2 < l2) {
  135. if ((*cmp)(f1, f2) <= 0) {
  136. q = f2;
  137. b = f1, t = l1;
  138. sense = -1;
  139. } else {
  140. q = f1;
  141. b = f2, t = l2;
  142. sense = 0;
  143. }
  144. if (!big) { /* here i = 0 */
  145. while ((b += size) < t && cmp(q, b) >sense)
  146. if (++i == 6) {
  147. big = 1;
  148. goto EXPONENTIAL;
  149. }
  150. } else {
  151. EXPONENTIAL: for (i = size; ; i <<= 1)
  152. if ((p = (b + i)) >= t) {
  153. if ((p = t - size) > b &&
  154. (*cmp)(q, p) <= sense)
  155. t = p;
  156. else
  157. b = p;
  158. break;
  159. } else if ((*cmp)(q, p) <= sense) {
  160. t = p;
  161. if (i == size)
  162. big = 0;
  163. goto FASTCASE;
  164. } else
  165. b = p;
  166. while (t > b+size) {
  167. i = (((t - b) / size) >> 1) * size;
  168. if ((*cmp)(q, p = b + i) <= sense)
  169. t = p;
  170. else
  171. b = p;
  172. }
  173. goto COPY;
  174. FASTCASE: while (i > size)
  175. if ((*cmp)(q,
  176. p = b + (i >>= 1)) <= sense)
  177. t = p;
  178. else
  179. b = p;
  180. COPY: b = t;
  181. }
  182. i = size;
  183. if (q == f1) {
  184. if (iflag) {
  185. ICOPY_LIST(f2, tp2, b);
  186. ICOPY_ELT(f1, tp2, i);
  187. } else {
  188. CCOPY_LIST(f2, tp2, b);
  189. CCOPY_ELT(f1, tp2, i);
  190. }
  191. } else {
  192. if (iflag) {
  193. ICOPY_LIST(f1, tp2, b);
  194. ICOPY_ELT(f2, tp2, i);
  195. } else {
  196. CCOPY_LIST(f1, tp2, b);
  197. CCOPY_ELT(f2, tp2, i);
  198. }
  199. }
  200. }
  201. if (f2 < l2) {
  202. if (iflag)
  203. ICOPY_LIST(f2, tp2, l2);
  204. else
  205. CCOPY_LIST(f2, tp2, l2);
  206. } else if (f1 < l1) {
  207. if (iflag)
  208. ICOPY_LIST(f1, tp2, l1);
  209. else
  210. CCOPY_LIST(f1, tp2, l1);
  211. }
  212. *p1 = l2;
  213. }
  214. tp2 = list1; /* swap list1, list2 */
  215. list1 = list2;
  216. list2 = tp2;
  217. last = list2 + nmemb*size;
  218. }
  219. if (base == list2) {
  220. memmove(list2, list1, nmemb*size);
  221. list2 = list1;
  222. }
  223. yasm_xfree(list2);
  224. return (0);
  225. #endif /*HAVE_MERGESORT*/
  226. }
  227. #ifndef HAVE_MERGESORT
  228. #define swap(a, b) { \
  229. s = b; \
  230. i = size; \
  231. do { \
  232. tmp = *a; *a++ = *s; *s++ = tmp; \
  233. } while (--i); \
  234. a -= size; \
  235. }
  236. #define reverse(bot, top) { \
  237. s = top; \
  238. do { \
  239. i = size; \
  240. do { \
  241. tmp = *bot; *bot++ = *s; *s++ = tmp; \
  242. } while (--i); \
  243. s -= size2; \
  244. } while(bot < s); \
  245. }
  246. /*
  247. * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of
  248. * increasing order, list2 in a corresponding linked list. Checks for runs
  249. * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL
  250. * is defined. Otherwise simple pairwise merging is used.)
  251. */
  252. void
  253. setup(unsigned char *list1, unsigned char *list2, size_t n, size_t size,
  254. int (*cmp)(const void *, const void *))
  255. {
  256. size_t i;
  257. unsigned int tmp;
  258. int length, sense;
  259. size_t size2;
  260. unsigned char *f1, *f2, *s, *l2, *last, *p2;
  261. size2 = size*2;
  262. if (n <= 5) {
  263. insertionsort(list1, n, size, cmp);
  264. *EVAL(list2) = (unsigned char*) list2 + n*size;
  265. return;
  266. }
  267. /*
  268. * Avoid running pointers out of bounds; limit n to evens
  269. * for simplicity.
  270. */
  271. i = 4 + (n & 1);
  272. insertionsort(list1 + (n - i) * size, i, size, cmp);
  273. last = list1 + size * (n - i);
  274. *EVAL(list2 + (last - list1)) = list2 + n * size;
  275. #ifdef NATURAL
  276. p2 = list2;
  277. f1 = list1;
  278. sense = (cmp(f1, f1 + size) > 0);
  279. for (; f1 < last; sense = !sense) {
  280. length = 2;
  281. /* Find pairs with same sense. */
  282. for (f2 = f1 + size2; f2 < last; f2 += size2) {
  283. if ((cmp(f2, f2+ size) > 0) != sense)
  284. break;
  285. length += 2;
  286. }
  287. if (length < THRESHOLD) { /* Pairwise merge */
  288. do {
  289. p2 = *EVAL(p2) = f1 + size2 - list1 + list2;
  290. if (sense > 0)
  291. swap (f1, f1 + size);
  292. } while ((f1 += size2) < f2);
  293. } else { /* Natural merge */
  294. l2 = f2;
  295. for (f2 = f1 + size2; f2 < l2; f2 += size2) {
  296. if ((cmp(f2-size, f2) > 0) != sense) {
  297. p2 = *EVAL(p2) = f2 - list1 + list2;
  298. if (sense > 0)
  299. reverse(f1, f2-size);
  300. f1 = f2;
  301. }
  302. }
  303. if (sense > 0)
  304. reverse (f1, f2-size);
  305. f1 = f2;
  306. if (f2 < last || cmp(f2 - size, f2) > 0)
  307. p2 = *EVAL(p2) = f2 - list1 + list2;
  308. else
  309. p2 = *EVAL(p2) = list2 + n*size;
  310. }
  311. }
  312. #else /* pairwise merge only. */
  313. for (f1 = list1, p2 = list2; f1 < last; f1 += size2) {
  314. p2 = *EVAL(p2) = p2 + size2;
  315. if (cmp (f1, f1 + size) > 0)
  316. swap(f1, f1 + size);
  317. }
  318. #endif /* NATURAL */
  319. }
  320. /*
  321. * This is to avoid out-of-bounds addresses in sorting the
  322. * last 4 elements.
  323. */
  324. static void
  325. insertionsort(unsigned char *a, size_t n, size_t size,
  326. int (*cmp)(const void *, const void *))
  327. {
  328. unsigned char *ai, *s, *t, *u, tmp;
  329. size_t i;
  330. for (ai = a+size; --n >= 1; ai += size)
  331. for (t = ai; t > a; t -= size) {
  332. u = t - size;
  333. if (cmp(u, t) <= 0)
  334. break;
  335. swap(u, t);
  336. }
  337. }
  338. #endif /*HAVE_MERGESORT*/