1 /* $NetBSD: merge.c,v 1.14 2012/03/13 21:13:48 christos Exp $ */ 2 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 35 #include <sys/cdefs.h> 36 #if defined(LIBC_SCCS) && !defined(lint) 37 #if 0 38 static char sccsid[] = "from: @(#)merge.c 8.2 (Berkeley) 2/14/94"; 39 #else 40 __RCSID("$NetBSD: merge.c,v 1.14 2012/03/13 21:13:48 christos Exp $"); 41 #endif 42 #endif /* LIBC_SCCS and not lint */ 43 44 /* 45 * Hybrid exponential search/linear search merge sort with hybrid 46 * natural/pairwise first pass. Requires about .3% more comparisons 47 * for random data than LSMS with pairwise first pass alone. 48 * It works for objects as small as two bytes. 49 */ 50 51 #define NATURAL 52 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */ 53 54 /* #define NATURAL to get hybrid natural merge. 55 * (The default is pairwise merging.) 56 */ 57 58 #include "namespace.h" 59 #include <sys/types.h> 60 61 #include <assert.h> 62 #include <errno.h> 63 #include <stdlib.h> 64 #include <string.h> 65 66 #ifdef __weak_alias 67 __weak_alias(mergesort,_mergesort) 68 #endif 69 70 static void setup(u_char *, u_char *, size_t, size_t, 71 int (*)(const void *, const void *)); 72 static void insertionsort(u_char *, size_t, size_t, 73 int (*)(const void *, const void *)); 74 75 #define ISIZE sizeof(int) 76 #define PSIZE sizeof(u_char *) 77 #define ICOPY_LIST(src, dst, last) \ 78 do \ 79 *(int*)(void *)dst = *(int*)(void *)src, \ 80 src += ISIZE, dst += ISIZE; \ 81 while(src < last) 82 #define ICOPY_ELT(src, dst, i) \ 83 do \ 84 *(int*)(void *)dst = *(int*)(void *)src, \ 85 src += ISIZE, dst += ISIZE; \ 86 while (i -= ISIZE) 87 88 #define CCOPY_LIST(src, dst, last) \ 89 do \ 90 *dst++ = *src++; \ 91 while (src < last) 92 #define CCOPY_ELT(src, dst, i) \ 93 do \ 94 *dst++ = *src++; \ 95 while (i -= 1) 96 97 /* 98 * Find the next possible pointer head. (Trickery for forcing an array 99 * to do double duty as a linked list when objects do not align with word 100 * boundaries. 101 */ 102 /* Assumption: PSIZE is a power of 2. */ 103 #define EVAL(p) ((u_char **)(void *) \ 104 ((u_char *)0 + \ 105 (((u_char *)(void *)(p) + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))) 106 107 /* 108 * Arguments are as for qsort. 109 */ 110 int 111 mergesort(void *base, size_t nmemb, size_t size, 112 int (*cmp)(const void *, const void *)) 113 { 114 size_t i; 115 int sense; 116 int big, iflag; 117 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; 118 u_char *list2, *list1, *p2, *p, *last, **p1; 119 120 _DIAGASSERT(base != NULL); 121 _DIAGASSERT(cmp != NULL); 122 123 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ 124 errno = EINVAL; 125 return (-1); 126 } 127 128 /* 129 * XXX 130 * Stupid subtraction for the Cray. 131 */ 132 iflag = 0; 133 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) 134 iflag = 1; 135 136 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) 137 return (-1); 138 139 list1 = base; 140 setup(list1, list2, nmemb, size, cmp); 141 last = list2 + nmemb * size; 142 i = big = 0; 143 while (*EVAL(list2) != last) { 144 l2 = list1; 145 p1 = EVAL(list1); 146 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { 147 p2 = *EVAL(p2); 148 f1 = l2; 149 f2 = l1 = list1 + (p2 - list2); 150 if (p2 != last) 151 p2 = *EVAL(p2); 152 l2 = list1 + (p2 - list2); 153 while (f1 < l1 && f2 < l2) { 154 if ((*cmp)(f1, f2) <= 0) { 155 q = f2; 156 b = f1, t = l1; 157 sense = -1; 158 } else { 159 q = f1; 160 b = f2, t = l2; 161 sense = 0; 162 } 163 if (!big) { /* here i = 0 */ 164 #if 0 165 LINEAR: 166 #endif 167 while ((b += size) < t && cmp(q, b) >sense) 168 if (++i == 6) { 169 big = 1; 170 goto EXPONENTIAL; 171 } 172 } else { 173 EXPONENTIAL: for (i = size; ; i <<= 1) 174 if ((p = (b + i)) >= t) { 175 if ((p = t - size) > b && 176 (*cmp)(q, p) <= sense) 177 t = p; 178 else 179 b = p; 180 break; 181 } else if ((*cmp)(q, p) <= sense) { 182 t = p; 183 if (i == size) 184 big = 0; 185 goto FASTCASE; 186 } else 187 b = p; 188 #if 0 189 SLOWCASE: 190 #endif 191 while (t > b+size) { 192 i = (((t - b) / size) >> 1) * size; 193 if ((*cmp)(q, p = b + i) <= sense) 194 t = p; 195 else 196 b = p; 197 } 198 goto COPY; 199 FASTCASE: while (i > size) 200 if ((*cmp)(q, 201 p = b + 202 (i = (unsigned int) i >> 1)) <= sense) 203 t = p; 204 else 205 b = p; 206 COPY: b = t; 207 } 208 i = size; 209 if (q == f1) { 210 if (iflag) { 211 ICOPY_LIST(f2, tp2, b); 212 ICOPY_ELT(f1, tp2, i); 213 } else { 214 CCOPY_LIST(f2, tp2, b); 215 CCOPY_ELT(f1, tp2, i); 216 } 217 } else { 218 if (iflag) { 219 ICOPY_LIST(f1, tp2, b); 220 ICOPY_ELT(f2, tp2, i); 221 } else { 222 CCOPY_LIST(f1, tp2, b); 223 CCOPY_ELT(f2, tp2, i); 224 } 225 } 226 } 227 if (f2 < l2) { 228 if (iflag) 229 ICOPY_LIST(f2, tp2, l2); 230 else 231 CCOPY_LIST(f2, tp2, l2); 232 } else if (f1 < l1) { 233 if (iflag) 234 ICOPY_LIST(f1, tp2, l1); 235 else 236 CCOPY_LIST(f1, tp2, l1); 237 } 238 *p1 = l2; 239 } 240 tp2 = list1; /* swap list1, list2 */ 241 list1 = list2; 242 list2 = tp2; 243 last = list2 + nmemb*size; 244 } 245 if (base == list2) { 246 memmove(list2, list1, nmemb*size); 247 list2 = list1; 248 } 249 free(list2); 250 return (0); 251 } 252 253 #define swap(a, b) { \ 254 s = b; \ 255 i = size; \ 256 do { \ 257 tmp = *a; *a++ = *s; *s++ = tmp; \ 258 } while (--i); \ 259 a -= size; \ 260 } 261 #define reverse(bot, top) { \ 262 s = top; \ 263 do { \ 264 i = size; \ 265 do { \ 266 tmp = *bot; *bot++ = *s; *s++ = tmp; \ 267 } while (--i); \ 268 s -= size2; \ 269 } while(bot < s); \ 270 } 271 272 /* 273 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of 274 * increasing order, list2 in a corresponding linked list. Checks for runs 275 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL 276 * is defined. Otherwise simple pairwise merging is used.) 277 */ 278 279 /* XXX: shouldn't this function be static? - lukem 990810 */ 280 void 281 setup(u_char *list1, u_char *list2, size_t n, size_t size, 282 int (*cmp)(const void *, const void *)) 283 { 284 int length, tmp, sense; 285 u_char *f1, *f2, *s, *l2, *last, *p2; 286 size_t size2, i; 287 288 _DIAGASSERT(cmp != NULL); 289 _DIAGASSERT(list1 != NULL); 290 _DIAGASSERT(list2 != NULL); 291 292 size2 = size*2; 293 if (n <= 5) { 294 insertionsort(list1, n, size, cmp); 295 *EVAL(list2) = list2 + n*size; 296 return; 297 } 298 /* 299 * Avoid running pointers out of bounds; limit n to evens 300 * for simplicity. 301 */ 302 i = 4 + (n & 1); 303 insertionsort(list1 + (n - i) * size, i, size, cmp); 304 last = list1 + size * (n - i); 305 *EVAL(list2 + (last - list1)) = list2 + n * size; 306 307 #ifdef NATURAL 308 p2 = list2; 309 f1 = list1; 310 sense = (cmp(f1, f1 + size) > 0); 311 for (; f1 < last; sense = !sense) { 312 length = 2; 313 /* Find pairs with same sense. */ 314 for (f2 = f1 + size2; f2 < last; f2 += size2) { 315 if ((cmp(f2, f2+ size) > 0) != sense) 316 break; 317 length += 2; 318 } 319 if (length < THRESHOLD) { /* Pairwise merge */ 320 do { 321 p2 = *EVAL(p2) = f1 + size2 - list1 + list2; 322 if (sense > 0) 323 swap (f1, f1 + size); 324 } while ((f1 += size2) < f2); 325 } else { /* Natural merge */ 326 l2 = f2; 327 for (f2 = f1 + size2; f2 < l2; f2 += size2) { 328 if ((cmp(f2-size, f2) > 0) != sense) { 329 p2 = *EVAL(p2) = f2 - list1 + list2; 330 if (sense > 0) 331 reverse(f1, f2 - size); 332 f1 = f2; 333 } 334 } 335 if (sense > 0) 336 reverse(f1, f2 - size); 337 f1 = f2; 338 if (f2 < last || cmp(f2 - size, f2) > 0) 339 p2 = *EVAL(p2) = f2 - list1 + list2; 340 else 341 p2 = *EVAL(p2) = list2 + n*size; 342 } 343 } 344 #else /* pairwise merge only. */ 345 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { 346 p2 = *EVAL(p2) = p2 + size2; 347 if (cmp (f1, f1 + size) > 0) 348 swap(f1, f1 + size); 349 } 350 #endif /* NATURAL */ 351 } 352 353 /* 354 * This is to avoid out-of-bounds addresses in sorting the 355 * last 4 elements. 356 */ 357 static void 358 insertionsort(u_char *a, size_t n, size_t size, 359 int (*cmp)(const void *, const void *)) 360 { 361 u_char *ai, *s, *t, *u, tmp; 362 size_t i; 363 364 _DIAGASSERT(a != NULL); 365 _DIAGASSERT(cmp != NULL); 366 367 for (ai = a+size; --n >= 1; ai += size) 368 for (t = ai; t > a; t -= size) { 369 u = t - size; 370 if (cmp(u, t) <= 0) 371 break; 372 swap(u, t); 373 } 374 } 375