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