1 /*- 2 * Copyright (c) 1992, 1993 3 * The Regents of the University of California. All rights reserved. 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 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37 #if defined(LIBC_SCCS) && !defined(lint) 38 /*static char sccsid[] = "from: @(#)merge.c 8.2 (Berkeley) 2/14/94";*/ 39 static char *rcsid = "$Id: merge.c,v 1.1.1.1 1995/10/18 08:42:18 deraadt Exp $"; 40 #endif /* LIBC_SCCS and not lint */ 41 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 49 #define NATURAL 50 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */ 51 52 /* #define NATURAL to get hybrid natural merge. 53 * (The default is pairwise merging.) 54 */ 55 56 #include <sys/types.h> 57 58 #include <errno.h> 59 #include <stdlib.h> 60 #include <string.h> 61 62 static void setup __P((u_char *, u_char *, size_t, size_t, int (*)())); 63 static void insertionsort __P((u_char *, size_t, size_t, int (*)())); 64 65 #define ISIZE sizeof(int) 66 #define PSIZE sizeof(u_char *) 67 #define ICOPY_LIST(src, dst, last) \ 68 do \ 69 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ 70 while(src < last) 71 #define ICOPY_ELT(src, dst, i) \ 72 do \ 73 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ 74 while (i -= ISIZE) 75 76 #define CCOPY_LIST(src, dst, last) \ 77 do \ 78 *dst++ = *src++; \ 79 while (src < last) 80 #define CCOPY_ELT(src, dst, i) \ 81 do \ 82 *dst++ = *src++; \ 83 while (i -= 1) 84 85 /* 86 * Find the next possible pointer head. (Trickery for forcing an array 87 * to do double duty as a linked list when objects do not align with word 88 * boundaries. 89 */ 90 /* Assumption: PSIZE is a power of 2. */ 91 #define EVAL(p) (u_char **) \ 92 ((u_char *)0 + \ 93 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1))) 94 95 /* 96 * Arguments are as for qsort. 97 */ 98 int 99 mergesort(base, nmemb, size, cmp) 100 void *base; 101 size_t nmemb; 102 register size_t size; 103 int (*cmp) __P((const void *, const void *)); 104 { 105 register int i, sense; 106 int big, iflag; 107 register u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; 108 u_char *list2, *list1, *p2, *p, *last, **p1; 109 110 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ 111 errno = EINVAL; 112 return (-1); 113 } 114 115 /* 116 * XXX 117 * Stupid subtraction for the Cray. 118 */ 119 iflag = 0; 120 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) 121 iflag = 1; 122 123 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) 124 return (-1); 125 126 list1 = base; 127 setup(list1, list2, nmemb, size, cmp); 128 last = list2 + nmemb * size; 129 i = big = 0; 130 while (*EVAL(list2) != last) { 131 l2 = list1; 132 p1 = EVAL(list1); 133 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { 134 p2 = *EVAL(p2); 135 f1 = l2; 136 f2 = l1 = list1 + (p2 - list2); 137 if (p2 != last) 138 p2 = *EVAL(p2); 139 l2 = list1 + (p2 - list2); 140 while (f1 < l1 && f2 < l2) { 141 if ((*cmp)(f1, f2) <= 0) { 142 q = f2; 143 b = f1, t = l1; 144 sense = -1; 145 } else { 146 q = f1; 147 b = f2, t = l2; 148 sense = 0; 149 } 150 if (!big) { /* here i = 0 */ 151 LINEAR: while ((b += size) < t && cmp(q, b) >sense) 152 if (++i == 6) { 153 big = 1; 154 goto EXPONENTIAL; 155 } 156 } else { 157 EXPONENTIAL: for (i = size; ; i <<= 1) 158 if ((p = (b + i)) >= t) { 159 if ((p = t - size) > b && 160 (*cmp)(q, p) <= sense) 161 t = p; 162 else 163 b = p; 164 break; 165 } else if ((*cmp)(q, p) <= sense) { 166 t = p; 167 if (i == size) 168 big = 0; 169 goto FASTCASE; 170 } else 171 b = p; 172 SLOWCASE: while (t > b+size) { 173 i = (((t - b) / size) >> 1) * size; 174 if ((*cmp)(q, p = b + i) <= sense) 175 t = p; 176 else 177 b = p; 178 } 179 goto COPY; 180 FASTCASE: while (i > size) 181 if ((*cmp)(q, 182 p = b + (i >>= 1)) <= sense) 183 t = p; 184 else 185 b = p; 186 COPY: b = t; 187 } 188 i = size; 189 if (q == f1) { 190 if (iflag) { 191 ICOPY_LIST(f2, tp2, b); 192 ICOPY_ELT(f1, tp2, i); 193 } else { 194 CCOPY_LIST(f2, tp2, b); 195 CCOPY_ELT(f1, tp2, i); 196 } 197 } else { 198 if (iflag) { 199 ICOPY_LIST(f1, tp2, b); 200 ICOPY_ELT(f2, tp2, i); 201 } else { 202 CCOPY_LIST(f1, tp2, b); 203 CCOPY_ELT(f2, tp2, i); 204 } 205 } 206 } 207 if (f2 < l2) { 208 if (iflag) 209 ICOPY_LIST(f2, tp2, l2); 210 else 211 CCOPY_LIST(f2, tp2, l2); 212 } else if (f1 < l1) { 213 if (iflag) 214 ICOPY_LIST(f1, tp2, l1); 215 else 216 CCOPY_LIST(f1, tp2, l1); 217 } 218 *p1 = l2; 219 } 220 tp2 = list1; /* swap list1, list2 */ 221 list1 = list2; 222 list2 = tp2; 223 last = list2 + nmemb*size; 224 } 225 if (base == list2) { 226 memmove(list2, list1, nmemb*size); 227 list2 = list1; 228 } 229 free(list2); 230 return (0); 231 } 232 233 #define swap(a, b) { \ 234 s = b; \ 235 i = size; \ 236 do { \ 237 tmp = *a; *a++ = *s; *s++ = tmp; \ 238 } while (--i); \ 239 a -= size; \ 240 } 241 #define reverse(bot, top) { \ 242 s = top; \ 243 do { \ 244 i = size; \ 245 do { \ 246 tmp = *bot; *bot++ = *s; *s++ = tmp; \ 247 } while (--i); \ 248 s -= size2; \ 249 } while(bot < s); \ 250 } 251 252 /* 253 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of 254 * increasing order, list2 in a corresponding linked list. Checks for runs 255 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL 256 * is defined. Otherwise simple pairwise merging is used.) 257 */ 258 void 259 setup(list1, list2, n, size, cmp) 260 size_t n, size; 261 int (*cmp) __P((const void *, const void *)); 262 u_char *list1, *list2; 263 { 264 int i, length, size2, tmp, sense; 265 u_char *f1, *f2, *s, *l2, *last, *p2; 266 267 size2 = size*2; 268 if (n <= 5) { 269 insertionsort(list1, n, size, cmp); 270 *EVAL(list2) = (u_char*) list2 + n*size; 271 return; 272 } 273 /* 274 * Avoid running pointers out of bounds; limit n to evens 275 * for simplicity. 276 */ 277 i = 4 + (n & 1); 278 insertionsort(list1 + (n - i) * size, i, size, cmp); 279 last = list1 + size * (n - i); 280 *EVAL(list2 + (last - list1)) = list2 + n * size; 281 282 #ifdef NATURAL 283 p2 = list2; 284 f1 = list1; 285 sense = (cmp(f1, f1 + size) > 0); 286 for (; f1 < last; sense = !sense) { 287 length = 2; 288 /* Find pairs with same sense. */ 289 for (f2 = f1 + size2; f2 < last; f2 += size2) { 290 if ((cmp(f2, f2+ size) > 0) != sense) 291 break; 292 length += 2; 293 } 294 if (length < THRESHOLD) { /* Pairwise merge */ 295 do { 296 p2 = *EVAL(p2) = f1 + size2 - list1 + list2; 297 if (sense > 0) 298 swap (f1, f1 + size); 299 } while ((f1 += size2) < f2); 300 } else { /* Natural merge */ 301 l2 = f2; 302 for (f2 = f1 + size2; f2 < l2; f2 += size2) { 303 if ((cmp(f2-size, f2) > 0) != sense) { 304 p2 = *EVAL(p2) = f2 - list1 + list2; 305 if (sense > 0) 306 reverse(f1, f2-size); 307 f1 = f2; 308 } 309 } 310 if (sense > 0) 311 reverse (f1, f2-size); 312 f1 = f2; 313 if (f2 < last || cmp(f2 - size, f2) > 0) 314 p2 = *EVAL(p2) = f2 - list1 + list2; 315 else 316 p2 = *EVAL(p2) = list2 + n*size; 317 } 318 } 319 #else /* pairwise merge only. */ 320 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { 321 p2 = *EVAL(p2) = p2 + size2; 322 if (cmp (f1, f1 + size) > 0) 323 swap(f1, f1 + size); 324 } 325 #endif /* NATURAL */ 326 } 327 328 /* 329 * This is to avoid out-of-bounds addresses in sorting the 330 * last 4 elements. 331 */ 332 static void 333 insertionsort(a, n, size, cmp) 334 u_char *a; 335 size_t n, size; 336 int (*cmp) __P((const void *, const void *)); 337 { 338 u_char *ai, *s, *t, *u, tmp; 339 int i; 340 341 for (ai = a+size; --n >= 1; ai += size) 342 for (t = ai; t > a; t -= size) { 343 u = t - size; 344 if (cmp(u, t) <= 0) 345 break; 346 swap(u, t); 347 } 348 } 349