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 * @(#)merge.c 8.2 (Berkeley) 2/14/94 37 * $DragonFly: src/lib/libc/stdlib/merge.c,v 1.4 2003/09/06 08:19:16 asmodai Exp $ 38 */ 39 40 /* 41 * Hybrid exponential search/linear search merge sort with hybrid 42 * natural/pairwise first pass. Requires about .3% more comparisons 43 * for random data than LSMS with pairwise first pass alone. 44 * It works for objects as small as two bytes. 45 */ 46 47 #define NATURAL 48 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */ 49 50 /* #define NATURAL to get hybrid natural merge. 51 * (The default is pairwise merging.) 52 */ 53 54 #include <sys/types.h> 55 56 #include <errno.h> 57 #include <stdlib.h> 58 #include <string.h> 59 60 static void setup (u_char *, u_char *, size_t, size_t, int (*)()); 61 static void insertionsort (u_char *, size_t, size_t, int (*)()); 62 63 #define ISIZE sizeof(int) 64 #define PSIZE sizeof(u_char *) 65 #define ICOPY_LIST(src, dst, last) \ 66 do \ 67 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ 68 while(src < last) 69 #define ICOPY_ELT(src, dst, i) \ 70 do \ 71 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ 72 while (i -= ISIZE) 73 74 #define CCOPY_LIST(src, dst, last) \ 75 do \ 76 *dst++ = *src++; \ 77 while (src < last) 78 #define CCOPY_ELT(src, dst, i) \ 79 do \ 80 *dst++ = *src++; \ 81 while (i -= 1) 82 83 /* 84 * Find the next possible pointer head. (Trickery for forcing an array 85 * to do double duty as a linked list when objects do not align with word 86 * boundaries. 87 */ 88 /* Assumption: PSIZE is a power of 2. */ 89 #define EVAL(p) (u_char **) \ 90 ((u_char *)0 + \ 91 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1))) 92 93 /* 94 * Arguments are as for qsort. 95 */ 96 int 97 mergesort(base, nmemb, size, cmp) 98 void *base; 99 size_t nmemb; 100 size_t size; 101 int (*cmp) (const void *, const void *); 102 { 103 int i, sense; 104 int big, iflag; 105 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; 106 u_char *list2, *list1, *p2, *p, *last, **p1; 107 108 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ 109 errno = EINVAL; 110 return (-1); 111 } 112 113 if (nmemb == 0) 114 return (0); 115 116 /* 117 * XXX 118 * Stupid subtraction for the Cray. 119 */ 120 iflag = 0; 121 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) 122 iflag = 1; 123 124 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) 125 return (-1); 126 127 list1 = base; 128 setup(list1, list2, nmemb, size, cmp); 129 last = list2 + nmemb * size; 130 i = big = 0; 131 while (*EVAL(list2) != last) { 132 l2 = list1; 133 p1 = EVAL(list1); 134 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { 135 p2 = *EVAL(p2); 136 f1 = l2; 137 f2 = l1 = list1 + (p2 - list2); 138 if (p2 != last) 139 p2 = *EVAL(p2); 140 l2 = list1 + (p2 - list2); 141 while (f1 < l1 && f2 < l2) { 142 if ((*cmp)(f1, f2) <= 0) { 143 q = f2; 144 b = f1, t = l1; 145 sense = -1; 146 } else { 147 q = f1; 148 b = f2, t = l2; 149 sense = 0; 150 } 151 if (!big) { /* here i = 0 */ 152 while ((b += size) < t && cmp(q, b) >sense) 153 if (++i == 6) { 154 big = 1; 155 goto EXPONENTIAL; 156 } 157 } else { 158 EXPONENTIAL: for (i = size; ; i <<= 1) 159 if ((p = (b + i)) >= t) { 160 if ((p = t - size) > b && 161 (*cmp)(q, p) <= sense) 162 t = p; 163 else 164 b = p; 165 break; 166 } else if ((*cmp)(q, p) <= sense) { 167 t = p; 168 if (i == size) 169 big = 0; 170 goto FASTCASE; 171 } else 172 b = p; 173 while (t > b+size) { 174 i = (((t - b) / size) >> 1) * size; 175 if ((*cmp)(q, p = b + i) <= sense) 176 t = p; 177 else 178 b = p; 179 } 180 goto COPY; 181 FASTCASE: while (i > size) 182 if ((*cmp)(q, 183 p = b + (i >>= 1)) <= sense) 184 t = p; 185 else 186 b = p; 187 COPY: b = t; 188 } 189 i = size; 190 if (q == f1) { 191 if (iflag) { 192 ICOPY_LIST(f2, tp2, b); 193 ICOPY_ELT(f1, tp2, i); 194 } else { 195 CCOPY_LIST(f2, tp2, b); 196 CCOPY_ELT(f1, tp2, i); 197 } 198 } else { 199 if (iflag) { 200 ICOPY_LIST(f1, tp2, b); 201 ICOPY_ELT(f2, tp2, i); 202 } else { 203 CCOPY_LIST(f1, tp2, b); 204 CCOPY_ELT(f2, tp2, i); 205 } 206 } 207 } 208 if (f2 < l2) { 209 if (iflag) 210 ICOPY_LIST(f2, tp2, l2); 211 else 212 CCOPY_LIST(f2, tp2, l2); 213 } else if (f1 < l1) { 214 if (iflag) 215 ICOPY_LIST(f1, tp2, l1); 216 else 217 CCOPY_LIST(f1, tp2, l1); 218 } 219 *p1 = l2; 220 } 221 tp2 = list1; /* swap list1, list2 */ 222 list1 = list2; 223 list2 = tp2; 224 last = list2 + nmemb*size; 225 } 226 if (base == list2) { 227 memmove(list2, list1, nmemb*size); 228 list2 = list1; 229 } 230 free(list2); 231 return (0); 232 } 233 234 #define swap(a, b) { \ 235 s = b; \ 236 i = size; \ 237 do { \ 238 tmp = *a; *a++ = *s; *s++ = tmp; \ 239 } while (--i); \ 240 a -= size; \ 241 } 242 #define reverse(bot, top) { \ 243 s = top; \ 244 do { \ 245 i = size; \ 246 do { \ 247 tmp = *bot; *bot++ = *s; *s++ = tmp; \ 248 } while (--i); \ 249 s -= size2; \ 250 } while(bot < s); \ 251 } 252 253 /* 254 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of 255 * increasing order, list2 in a corresponding linked list. Checks for runs 256 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL 257 * is defined. Otherwise simple pairwise merging is used.) 258 */ 259 void 260 setup(list1, list2, n, size, cmp) 261 size_t n, size; 262 int (*cmp) (const void *, const void *); 263 u_char *list1, *list2; 264 { 265 int i, length, size2, tmp, sense; 266 u_char *f1, *f2, *s, *l2, *last, *p2; 267 268 size2 = size*2; 269 if (n <= 5) { 270 insertionsort(list1, n, size, cmp); 271 *EVAL(list2) = (u_char*) list2 + n*size; 272 return; 273 } 274 /* 275 * Avoid running pointers out of bounds; limit n to evens 276 * for simplicity. 277 */ 278 i = 4 + (n & 1); 279 insertionsort(list1 + (n - i) * size, i, size, cmp); 280 last = list1 + size * (n - i); 281 *EVAL(list2 + (last - list1)) = list2 + n * size; 282 283 #ifdef NATURAL 284 p2 = list2; 285 f1 = list1; 286 sense = (cmp(f1, f1 + size) > 0); 287 for (; f1 < last; sense = !sense) { 288 length = 2; 289 /* Find pairs with same sense. */ 290 for (f2 = f1 + size2; f2 < last; f2 += size2) { 291 if ((cmp(f2, f2+ size) > 0) != sense) 292 break; 293 length += 2; 294 } 295 if (length < THRESHOLD) { /* Pairwise merge */ 296 do { 297 p2 = *EVAL(p2) = f1 + size2 - list1 + list2; 298 if (sense > 0) 299 swap (f1, f1 + size); 300 } while ((f1 += size2) < f2); 301 } else { /* Natural merge */ 302 l2 = f2; 303 for (f2 = f1 + size2; f2 < l2; f2 += size2) { 304 if ((cmp(f2-size, f2) > 0) != sense) { 305 p2 = *EVAL(p2) = f2 - list1 + list2; 306 if (sense > 0) 307 reverse(f1, f2-size); 308 f1 = f2; 309 } 310 } 311 if (sense > 0) 312 reverse (f1, f2-size); 313 f1 = f2; 314 if (f2 < last || cmp(f2 - size, f2) > 0) 315 p2 = *EVAL(p2) = f2 - list1 + list2; 316 else 317 p2 = *EVAL(p2) = list2 + n*size; 318 } 319 } 320 #else /* pairwise merge only. */ 321 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { 322 p2 = *EVAL(p2) = p2 + size2; 323 if (cmp (f1, f1 + size) > 0) 324 swap(f1, f1 + size); 325 } 326 #endif /* NATURAL */ 327 } 328 329 /* 330 * This is to avoid out-of-bounds addresses in sorting the 331 * last 4 elements. 332 */ 333 static void 334 insertionsort(a, n, size, cmp) 335 u_char *a; 336 size_t n, size; 337 int (*cmp) (const void *, const void *); 338 { 339 u_char *ai, *s, *t, *u, tmp; 340 int i; 341 342 for (ai = a+size; --n >= 1; ai += size) 343 for (t = ai; t > a; t -= size) { 344 u = t - size; 345 if (cmp(u, t) <= 0) 346 break; 347 swap(u, t); 348 } 349 } 350