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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)merge.c 8.2 (Berkeley) 2/14/94 33 * $FreeBSD: src/lib/libc/stdlib/merge.c,v 1.8 2007/01/09 00:28:10 imp Exp $ 34 * $DragonFly: src/lib/libc/stdlib/merge.c,v 1.6 2005/11/20 12:37:48 swildner Exp $ 35 */ 36 37 /* 38 * Hybrid exponential search/linear search merge sort with hybrid 39 * natural/pairwise first pass. Requires about .3% more comparisons 40 * for random data than LSMS with pairwise first pass alone. 41 * It works for objects as small as two bytes. 42 */ 43 44 #define NATURAL 45 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */ 46 47 /* #define NATURAL to get hybrid natural merge. 48 * (The default is pairwise merging.) 49 */ 50 51 #include <sys/types.h> 52 53 #include <errno.h> 54 #include <stdlib.h> 55 #include <string.h> 56 57 static void setup(u_char *, u_char *, size_t, size_t, 58 int (*)(const void *, const void *)); 59 static void insertionsort(u_char *, size_t, size_t, 60 int (*)(const void *, const void *)); 61 62 #define ISIZE sizeof(int) 63 #define PSIZE sizeof(u_char *) 64 #define ICOPY_LIST(src, dst, last) \ 65 do \ 66 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ 67 while(src < last) 68 #define ICOPY_ELT(src, dst, i) \ 69 do \ 70 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ 71 while (i -= ISIZE) 72 73 #define CCOPY_LIST(src, dst, last) \ 74 do \ 75 *dst++ = *src++; \ 76 while (src < last) 77 #define CCOPY_ELT(src, dst, i) \ 78 do \ 79 *dst++ = *src++; \ 80 while (i -= 1) 81 82 /* 83 * Find the next possible pointer head. (Trickery for forcing an array 84 * to do double duty as a linked list when objects do not align with word 85 * boundaries. 86 */ 87 /* Assumption: PSIZE is a power of 2. */ 88 #define EVAL(p) (u_char **) \ 89 ((u_char *)0 + \ 90 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1))) 91 92 /* 93 * Arguments are as for qsort. 94 */ 95 int 96 mergesort(void *base, size_t nmemb, size_t size, 97 int (*cmp)(const void *, const void *)) 98 { 99 size_t i; 100 int sense; 101 int big, iflag; 102 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; 103 u_char *list2, *list1, *p2, *p, *last, **p1; 104 105 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ 106 errno = EINVAL; 107 return (-1); 108 } 109 110 if (nmemb == 0) 111 return (0); 112 113 /* 114 * XXX 115 * Stupid subtraction for the Cray. 116 */ 117 iflag = 0; 118 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) 119 iflag = 1; 120 121 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) 122 return (-1); 123 124 list1 = base; 125 setup(list1, list2, nmemb, size, cmp); 126 last = list2 + nmemb * size; 127 i = big = 0; 128 while (*EVAL(list2) != last) { 129 l2 = list1; 130 p1 = EVAL(list1); 131 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { 132 p2 = *EVAL(p2); 133 f1 = l2; 134 f2 = l1 = list1 + (p2 - list2); 135 if (p2 != last) 136 p2 = *EVAL(p2); 137 l2 = list1 + (p2 - list2); 138 while (f1 < l1 && f2 < l2) { 139 if ((*cmp)(f1, f2) <= 0) { 140 q = f2; 141 b = f1, t = l1; 142 sense = -1; 143 } else { 144 q = f1; 145 b = f2, t = l2; 146 sense = 0; 147 } 148 if (!big) { /* here i = 0 */ 149 while ((b += size) < t && cmp(q, b) >sense) 150 if (++i == 6) { 151 big = 1; 152 goto EXPONENTIAL; 153 } 154 } else { 155 EXPONENTIAL: for (i = size; ; i <<= 1) 156 if ((p = (b + i)) >= t) { 157 if ((p = t - size) > b && 158 (*cmp)(q, p) <= sense) 159 t = p; 160 else 161 b = p; 162 break; 163 } else if ((*cmp)(q, p) <= sense) { 164 t = p; 165 if (i == size) 166 big = 0; 167 goto FASTCASE; 168 } else 169 b = p; 170 while (t > b+size) { 171 i = (((t - b) / size) >> 1) * size; 172 if ((*cmp)(q, p = b + i) <= sense) 173 t = p; 174 else 175 b = p; 176 } 177 goto COPY; 178 FASTCASE: while (i > size) 179 if ((*cmp)(q, 180 p = b + (i >>= 1)) <= sense) 181 t = p; 182 else 183 b = p; 184 COPY: b = t; 185 } 186 i = size; 187 if (q == f1) { 188 if (iflag) { 189 ICOPY_LIST(f2, tp2, b); 190 ICOPY_ELT(f1, tp2, i); 191 } else { 192 CCOPY_LIST(f2, tp2, b); 193 CCOPY_ELT(f1, tp2, i); 194 } 195 } else { 196 if (iflag) { 197 ICOPY_LIST(f1, tp2, b); 198 ICOPY_ELT(f2, tp2, i); 199 } else { 200 CCOPY_LIST(f1, tp2, b); 201 CCOPY_ELT(f2, tp2, i); 202 } 203 } 204 } 205 if (f2 < l2) { 206 if (iflag) 207 ICOPY_LIST(f2, tp2, l2); 208 else 209 CCOPY_LIST(f2, tp2, l2); 210 } else if (f1 < l1) { 211 if (iflag) 212 ICOPY_LIST(f1, tp2, l1); 213 else 214 CCOPY_LIST(f1, tp2, l1); 215 } 216 *p1 = l2; 217 } 218 tp2 = list1; /* swap list1, list2 */ 219 list1 = list2; 220 list2 = tp2; 221 last = list2 + nmemb*size; 222 } 223 if (base == list2) { 224 memmove(list2, list1, nmemb*size); 225 list2 = list1; 226 } 227 free(list2); 228 return (0); 229 } 230 231 #define swap(a, b) { \ 232 s = b; \ 233 i = size; \ 234 do { \ 235 tmp = *a; *a++ = *s; *s++ = tmp; \ 236 } while (--i); \ 237 a -= size; \ 238 } 239 #define reverse(bot, top) { \ 240 s = top; \ 241 do { \ 242 i = size; \ 243 do { \ 244 tmp = *bot; *bot++ = *s; *s++ = tmp; \ 245 } while (--i); \ 246 s -= size2; \ 247 } while(bot < s); \ 248 } 249 250 /* 251 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of 252 * increasing order, list2 in a corresponding linked list. Checks for runs 253 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL 254 * is defined. Otherwise simple pairwise merging is used.) 255 */ 256 static void 257 setup(u_char *list1, u_char *list2, size_t n, size_t size, 258 int (*cmp)(const void *, const void *)) 259 { 260 int i, length, size2, tmp, sense; 261 u_char *f1, *f2, *s, *l2, *last, *p2; 262 263 size2 = size*2; 264 if (n <= 5) { 265 insertionsort(list1, n, size, cmp); 266 *EVAL(list2) = (u_char*) list2 + n*size; 267 return; 268 } 269 /* 270 * Avoid running pointers out of bounds; limit n to evens 271 * for simplicity. 272 */ 273 i = 4 + (n & 1); 274 insertionsort(list1 + (n - i) * size, i, size, cmp); 275 last = list1 + size * (n - i); 276 *EVAL(list2 + (last - list1)) = list2 + n * size; 277 278 #ifdef NATURAL 279 p2 = list2; 280 f1 = list1; 281 sense = (cmp(f1, f1 + size) > 0); 282 for (; f1 < last; sense = !sense) { 283 length = 2; 284 /* Find pairs with same sense. */ 285 for (f2 = f1 + size2; f2 < last; f2 += size2) { 286 if ((cmp(f2, f2+ size) > 0) != sense) 287 break; 288 length += 2; 289 } 290 if (length < THRESHOLD) { /* Pairwise merge */ 291 do { 292 p2 = *EVAL(p2) = f1 + size2 - list1 + list2; 293 if (sense > 0) 294 swap (f1, f1 + size); 295 } while ((f1 += size2) < f2); 296 } else { /* Natural merge */ 297 l2 = f2; 298 for (f2 = f1 + size2; f2 < l2; f2 += size2) { 299 if ((cmp(f2-size, f2) > 0) != sense) { 300 p2 = *EVAL(p2) = f2 - list1 + list2; 301 if (sense > 0) 302 reverse(f1, f2-size); 303 f1 = f2; 304 } 305 } 306 if (sense > 0) 307 reverse (f1, f2-size); 308 f1 = f2; 309 if (f2 < last || cmp(f2 - size, f2) > 0) 310 p2 = *EVAL(p2) = f2 - list1 + list2; 311 else 312 p2 = *EVAL(p2) = list2 + n*size; 313 } 314 } 315 #else /* pairwise merge only. */ 316 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { 317 p2 = *EVAL(p2) = p2 + size2; 318 if (cmp (f1, f1 + size) > 0) 319 swap(f1, f1 + size); 320 } 321 #endif /* NATURAL */ 322 } 323 324 /* 325 * This is to avoid out-of-bounds addresses in sorting the 326 * last 4 elements. 327 */ 328 static void 329 insertionsort(u_char *a, size_t n, size_t size, 330 int (*cmp)(const void *, const void *)) 331 { 332 u_char *ai, *s, *t, *u, tmp; 333 int i; 334 335 for (ai = a+size; --n >= 1; ai += size) 336 for (t = ai; t > a; t -= size) { 337 u = t - size; 338 if (cmp(u, t) <= 0) 339 break; 340 swap(u, t); 341 } 342 } 343