1 /* $NetBSD: merge.c,v 1.10 2000/01/22 22:19:19 mycroft 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. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 */ 38 39 #include <sys/cdefs.h> 40 #if defined(LIBC_SCCS) && !defined(lint) 41 #if 0 42 static char sccsid[] = "from: @(#)merge.c 8.2 (Berkeley) 2/14/94"; 43 #else 44 __RCSID("$NetBSD: merge.c,v 1.10 2000/01/22 22:19:19 mycroft Exp $"); 45 #endif 46 #endif /* LIBC_SCCS and not lint */ 47 48 /* 49 * Hybrid exponential search/linear search merge sort with hybrid 50 * natural/pairwise first pass. Requires about .3% more comparisons 51 * for random data than LSMS with pairwise first pass alone. 52 * It works for objects as small as two bytes. 53 */ 54 55 #define NATURAL 56 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */ 57 58 /* #define NATURAL to get hybrid natural merge. 59 * (The default is pairwise merging.) 60 */ 61 62 #include "namespace.h" 63 #include <sys/types.h> 64 65 #include <assert.h> 66 #include <errno.h> 67 #include <stdlib.h> 68 #include <string.h> 69 70 #ifdef __weak_alias 71 __weak_alias(mergesort,_mergesort) 72 #endif 73 74 static void setup __P((u_char *, u_char *, size_t, size_t, 75 int (*)(const void *, const void *))); 76 static void insertionsort __P((u_char *, size_t, size_t, 77 int (*)(const void *, const void *))); 78 79 #define ISIZE sizeof(int) 80 #define PSIZE sizeof(u_char *) 81 #define ICOPY_LIST(src, dst, last) \ 82 do \ 83 *(int*)(void *)dst = *(int*)(void *)src, \ 84 src += ISIZE, dst += ISIZE; \ 85 while(src < last) 86 #define ICOPY_ELT(src, dst, i) \ 87 do \ 88 *(int*)(void *)dst = *(int*)(void *)src, \ 89 src += ISIZE, dst += ISIZE; \ 90 while (i -= ISIZE) 91 92 #define CCOPY_LIST(src, dst, last) \ 93 do \ 94 *dst++ = *src++; \ 95 while (src < last) 96 #define CCOPY_ELT(src, dst, i) \ 97 do \ 98 *dst++ = *src++; \ 99 while (i -= 1) 100 101 /* 102 * Find the next possible pointer head. (Trickery for forcing an array 103 * to do double duty as a linked list when objects do not align with word 104 * boundaries. 105 */ 106 /* Assumption: PSIZE is a power of 2. */ 107 #define EVAL(p) ((u_char **)(void *) \ 108 ((u_char *)0 + \ 109 (((u_char *)(void *)(p) + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1)))) 110 111 /* 112 * Arguments are as for qsort. 113 */ 114 int 115 mergesort(base, nmemb, size, cmp) 116 void *base; 117 size_t nmemb; 118 size_t size; 119 int (*cmp) __P((const void *, const void *)); 120 { 121 int i, sense; 122 int big, iflag; 123 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; 124 u_char *list2, *list1, *p2, *p, *last, **p1; 125 126 _DIAGASSERT(base != NULL); 127 _DIAGASSERT(cmp != NULL); 128 129 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ 130 errno = EINVAL; 131 return (-1); 132 } 133 134 /* 135 * XXX 136 * Stupid subtraction for the Cray. 137 */ 138 iflag = 0; 139 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) 140 iflag = 1; 141 142 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) 143 return (-1); 144 145 list1 = base; 146 setup(list1, list2, nmemb, size, cmp); 147 last = list2 + nmemb * size; 148 i = big = 0; 149 while (*EVAL(list2) != last) { 150 l2 = list1; 151 p1 = EVAL(list1); 152 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { 153 p2 = *EVAL(p2); 154 f1 = l2; 155 f2 = l1 = list1 + (p2 - list2); 156 if (p2 != last) 157 p2 = *EVAL(p2); 158 l2 = list1 + (p2 - list2); 159 while (f1 < l1 && f2 < l2) { 160 if ((*cmp)(f1, f2) <= 0) { 161 q = f2; 162 b = f1, t = l1; 163 sense = -1; 164 } else { 165 q = f1; 166 b = f2, t = l2; 167 sense = 0; 168 } 169 if (!big) { /* here i = 0 */ 170 #if 0 171 LINEAR: 172 #endif 173 while ((b += size) < t && cmp(q, b) >sense) 174 if (++i == 6) { 175 big = 1; 176 goto EXPONENTIAL; 177 } 178 } else { 179 EXPONENTIAL: for (i = size; ; i <<= 1) 180 if ((p = (b + i)) >= t) { 181 if ((p = t - size) > b && 182 (*cmp)(q, p) <= sense) 183 t = p; 184 else 185 b = p; 186 break; 187 } else if ((*cmp)(q, p) <= sense) { 188 t = p; 189 if (i == size) 190 big = 0; 191 goto FASTCASE; 192 } else 193 b = p; 194 #if 0 195 SLOWCASE: 196 #endif 197 while (t > b+size) { 198 i = (((t - b) / size) >> 1) * size; 199 if ((*cmp)(q, p = b + i) <= sense) 200 t = p; 201 else 202 b = p; 203 } 204 goto COPY; 205 FASTCASE: while (i > size) 206 if ((*cmp)(q, 207 p = b + 208 (i = (unsigned int) i >> 1)) <= sense) 209 t = p; 210 else 211 b = p; 212 COPY: b = t; 213 } 214 i = size; 215 if (q == f1) { 216 if (iflag) { 217 ICOPY_LIST(f2, tp2, b); 218 ICOPY_ELT(f1, tp2, i); 219 } else { 220 CCOPY_LIST(f2, tp2, b); 221 CCOPY_ELT(f1, tp2, i); 222 } 223 } else { 224 if (iflag) { 225 ICOPY_LIST(f1, tp2, b); 226 ICOPY_ELT(f2, tp2, i); 227 } else { 228 CCOPY_LIST(f1, tp2, b); 229 CCOPY_ELT(f2, tp2, i); 230 } 231 } 232 } 233 if (f2 < l2) { 234 if (iflag) 235 ICOPY_LIST(f2, tp2, l2); 236 else 237 CCOPY_LIST(f2, tp2, l2); 238 } else if (f1 < l1) { 239 if (iflag) 240 ICOPY_LIST(f1, tp2, l1); 241 else 242 CCOPY_LIST(f1, tp2, l1); 243 } 244 *p1 = l2; 245 } 246 tp2 = list1; /* swap list1, list2 */ 247 list1 = list2; 248 list2 = tp2; 249 last = list2 + nmemb*size; 250 } 251 if (base == list2) { 252 memmove(list2, list1, nmemb*size); 253 list2 = list1; 254 } 255 free(list2); 256 return (0); 257 } 258 259 #define swap(a, b) { \ 260 s = b; \ 261 i = size; \ 262 do { \ 263 tmp = *a; *a++ = *s; *s++ = tmp; \ 264 } while (--i); \ 265 a -= size; \ 266 } 267 #define reverse(bot, top) { \ 268 s = top; \ 269 do { \ 270 i = size; \ 271 do { \ 272 tmp = *bot; *bot++ = *s; *s++ = tmp; \ 273 } while (--i); \ 274 s -= size2; \ 275 } while(bot < s); \ 276 } 277 278 /* 279 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of 280 * increasing order, list2 in a corresponding linked list. Checks for runs 281 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL 282 * is defined. Otherwise simple pairwise merging is used.) 283 */ 284 285 /* XXX: shouldn't this function be static? - lukem 990810 */ 286 void 287 setup(list1, list2, n, size, cmp) 288 size_t n, size; 289 int (*cmp) __P((const void *, const void *)); 290 u_char *list1, *list2; 291 { 292 int i, length, size2, tmp, sense; 293 u_char *f1, *f2, *s, *l2, *last, *p2; 294 295 _DIAGASSERT(cmp != NULL); 296 _DIAGASSERT(list1 != NULL); 297 _DIAGASSERT(list2 != NULL); 298 299 size2 = size*2; 300 if (n <= 5) { 301 insertionsort(list1, n, size, cmp); 302 *EVAL(list2) = list2 + n*size; 303 return; 304 } 305 /* 306 * Avoid running pointers out of bounds; limit n to evens 307 * for simplicity. 308 */ 309 i = 4 + (n & 1); 310 insertionsort(list1 + (n - i) * size, (size_t)i, size, cmp); 311 last = list1 + size * (n - i); 312 *EVAL(list2 + (last - list1)) = list2 + n * size; 313 314 #ifdef NATURAL 315 p2 = list2; 316 f1 = list1; 317 sense = (cmp(f1, f1 + size) > 0); 318 for (; f1 < last; sense = !sense) { 319 length = 2; 320 /* Find pairs with same sense. */ 321 for (f2 = f1 + size2; f2 < last; f2 += size2) { 322 if ((cmp(f2, f2+ size) > 0) != sense) 323 break; 324 length += 2; 325 } 326 if (length < THRESHOLD) { /* Pairwise merge */ 327 do { 328 p2 = *EVAL(p2) = f1 + size2 - list1 + list2; 329 if (sense > 0) 330 swap (f1, f1 + size); 331 } while ((f1 += size2) < f2); 332 } else { /* Natural merge */ 333 l2 = f2; 334 for (f2 = f1 + size2; f2 < l2; f2 += size2) { 335 if ((cmp(f2-size, f2) > 0) != sense) { 336 p2 = *EVAL(p2) = f2 - list1 + list2; 337 if (sense > 0) 338 reverse(f1, f2-size); 339 f1 = f2; 340 } 341 } 342 if (sense > 0) 343 reverse (f1, f2-size); 344 f1 = f2; 345 if (f2 < last || cmp(f2 - size, f2) > 0) 346 p2 = *EVAL(p2) = f2 - list1 + list2; 347 else 348 p2 = *EVAL(p2) = list2 + n*size; 349 } 350 } 351 #else /* pairwise merge only. */ 352 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { 353 p2 = *EVAL(p2) = p2 + size2; 354 if (cmp (f1, f1 + size) > 0) 355 swap(f1, f1 + size); 356 } 357 #endif /* NATURAL */ 358 } 359 360 /* 361 * This is to avoid out-of-bounds addresses in sorting the 362 * last 4 elements. 363 */ 364 static void 365 insertionsort(a, n, size, cmp) 366 u_char *a; 367 size_t n, size; 368 int (*cmp) __P((const void *, const void *)); 369 { 370 u_char *ai, *s, *t, *u, tmp; 371 int i; 372 373 _DIAGASSERT(a != NULL); 374 _DIAGASSERT(cmp != NULL); 375 376 for (ai = a+size; --n >= 1; ai += size) 377 for (t = ai; t > a; t -= size) { 378 u = t - size; 379 if (cmp(u, t) <= 0) 380 break; 381 swap(u, t); 382 } 383 } 384