1 /* $NetBSD: merge.c,v 1.3 1995/12/28 08:52:28 thorpej 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 #if defined(LIBC_SCCS) && !defined(lint) 40 #if 0 41 static char sccsid[] = "from: @(#)merge.c 8.2 (Berkeley) 2/14/94"; 42 #else 43 static char *rcsid = "$NetBSD: merge.c,v 1.3 1995/12/28 08:52:28 thorpej Exp $"; 44 #endif 45 #endif /* LIBC_SCCS and not lint */ 46 47 /* 48 * Hybrid exponential search/linear search merge sort with hybrid 49 * natural/pairwise first pass. Requires about .3% more comparisons 50 * for random data than LSMS with pairwise first pass alone. 51 * It works for objects as small as two bytes. 52 */ 53 54 #define NATURAL 55 #define THRESHOLD 16 /* Best choice for natural merge cut-off. */ 56 57 /* #define NATURAL to get hybrid natural merge. 58 * (The default is pairwise merging.) 59 */ 60 61 #include <sys/types.h> 62 63 #include <errno.h> 64 #include <stdlib.h> 65 #include <string.h> 66 67 static void setup __P((u_char *, u_char *, size_t, size_t, int (*)())); 68 static void insertionsort __P((u_char *, size_t, size_t, int (*)())); 69 70 #define ISIZE sizeof(int) 71 #define PSIZE sizeof(u_char *) 72 #define ICOPY_LIST(src, dst, last) \ 73 do \ 74 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ 75 while(src < last) 76 #define ICOPY_ELT(src, dst, i) \ 77 do \ 78 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ 79 while (i -= ISIZE) 80 81 #define CCOPY_LIST(src, dst, last) \ 82 do \ 83 *dst++ = *src++; \ 84 while (src < last) 85 #define CCOPY_ELT(src, dst, i) \ 86 do \ 87 *dst++ = *src++; \ 88 while (i -= 1) 89 90 /* 91 * Find the next possible pointer head. (Trickery for forcing an array 92 * to do double duty as a linked list when objects do not align with word 93 * boundaries. 94 */ 95 /* Assumption: PSIZE is a power of 2. */ 96 #define EVAL(p) (u_char **) \ 97 ((u_char *)0 + \ 98 (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1))) 99 100 /* 101 * Arguments are as for qsort. 102 */ 103 int 104 mergesort(base, nmemb, size, cmp) 105 void *base; 106 size_t nmemb; 107 register size_t size; 108 int (*cmp) __P((const void *, const void *)); 109 { 110 register int i, sense; 111 int big, iflag; 112 register u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; 113 u_char *list2, *list1, *p2, *p, *last, **p1; 114 115 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ 116 errno = EINVAL; 117 return (-1); 118 } 119 120 /* 121 * XXX 122 * Stupid subtraction for the Cray. 123 */ 124 iflag = 0; 125 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) 126 iflag = 1; 127 128 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) 129 return (-1); 130 131 list1 = base; 132 setup(list1, list2, nmemb, size, cmp); 133 last = list2 + nmemb * size; 134 i = big = 0; 135 while (*EVAL(list2) != last) { 136 l2 = list1; 137 p1 = EVAL(list1); 138 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { 139 p2 = *EVAL(p2); 140 f1 = l2; 141 f2 = l1 = list1 + (p2 - list2); 142 if (p2 != last) 143 p2 = *EVAL(p2); 144 l2 = list1 + (p2 - list2); 145 while (f1 < l1 && f2 < l2) { 146 if ((*cmp)(f1, f2) <= 0) { 147 q = f2; 148 b = f1, t = l1; 149 sense = -1; 150 } else { 151 q = f1; 152 b = f2, t = l2; 153 sense = 0; 154 } 155 if (!big) { /* here i = 0 */ 156 LINEAR: while ((b += size) < t && cmp(q, b) >sense) 157 if (++i == 6) { 158 big = 1; 159 goto EXPONENTIAL; 160 } 161 } else { 162 EXPONENTIAL: for (i = size; ; i <<= 1) 163 if ((p = (b + i)) >= t) { 164 if ((p = t - size) > b && 165 (*cmp)(q, p) <= sense) 166 t = p; 167 else 168 b = p; 169 break; 170 } else if ((*cmp)(q, p) <= sense) { 171 t = p; 172 if (i == size) 173 big = 0; 174 goto FASTCASE; 175 } else 176 b = p; 177 SLOWCASE: while (t > b+size) { 178 i = (((t - b) / size) >> 1) * size; 179 if ((*cmp)(q, p = b + i) <= sense) 180 t = p; 181 else 182 b = p; 183 } 184 goto COPY; 185 FASTCASE: while (i > size) 186 if ((*cmp)(q, 187 p = b + (i >>= 1)) <= sense) 188 t = p; 189 else 190 b = p; 191 COPY: b = t; 192 } 193 i = size; 194 if (q == f1) { 195 if (iflag) { 196 ICOPY_LIST(f2, tp2, b); 197 ICOPY_ELT(f1, tp2, i); 198 } else { 199 CCOPY_LIST(f2, tp2, b); 200 CCOPY_ELT(f1, tp2, i); 201 } 202 } else { 203 if (iflag) { 204 ICOPY_LIST(f1, tp2, b); 205 ICOPY_ELT(f2, tp2, i); 206 } else { 207 CCOPY_LIST(f1, tp2, b); 208 CCOPY_ELT(f2, tp2, i); 209 } 210 } 211 } 212 if (f2 < l2) { 213 if (iflag) 214 ICOPY_LIST(f2, tp2, l2); 215 else 216 CCOPY_LIST(f2, tp2, l2); 217 } else if (f1 < l1) { 218 if (iflag) 219 ICOPY_LIST(f1, tp2, l1); 220 else 221 CCOPY_LIST(f1, tp2, l1); 222 } 223 *p1 = l2; 224 } 225 tp2 = list1; /* swap list1, list2 */ 226 list1 = list2; 227 list2 = tp2; 228 last = list2 + nmemb*size; 229 } 230 if (base == list2) { 231 memmove(list2, list1, nmemb*size); 232 list2 = list1; 233 } 234 free(list2); 235 return (0); 236 } 237 238 #define swap(a, b) { \ 239 s = b; \ 240 i = size; \ 241 do { \ 242 tmp = *a; *a++ = *s; *s++ = tmp; \ 243 } while (--i); \ 244 a -= size; \ 245 } 246 #define reverse(bot, top) { \ 247 s = top; \ 248 do { \ 249 i = size; \ 250 do { \ 251 tmp = *bot; *bot++ = *s; *s++ = tmp; \ 252 } while (--i); \ 253 s -= size2; \ 254 } while(bot < s); \ 255 } 256 257 /* 258 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of 259 * increasing order, list2 in a corresponding linked list. Checks for runs 260 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL 261 * is defined. Otherwise simple pairwise merging is used.) 262 */ 263 void 264 setup(list1, list2, n, size, cmp) 265 size_t n, size; 266 int (*cmp) __P((const void *, const void *)); 267 u_char *list1, *list2; 268 { 269 int i, length, size2, tmp, sense; 270 u_char *f1, *f2, *s, *l2, *last, *p2; 271 272 size2 = size*2; 273 if (n <= 5) { 274 insertionsort(list1, n, size, cmp); 275 *EVAL(list2) = (u_char*) list2 + n*size; 276 return; 277 } 278 /* 279 * Avoid running pointers out of bounds; limit n to evens 280 * for simplicity. 281 */ 282 i = 4 + (n & 1); 283 insertionsort(list1 + (n - i) * size, i, size, cmp); 284 last = list1 + size * (n - i); 285 *EVAL(list2 + (last - list1)) = list2 + n * size; 286 287 #ifdef NATURAL 288 p2 = list2; 289 f1 = list1; 290 sense = (cmp(f1, f1 + size) > 0); 291 for (; f1 < last; sense = !sense) { 292 length = 2; 293 /* Find pairs with same sense. */ 294 for (f2 = f1 + size2; f2 < last; f2 += size2) { 295 if ((cmp(f2, f2+ size) > 0) != sense) 296 break; 297 length += 2; 298 } 299 if (length < THRESHOLD) { /* Pairwise merge */ 300 do { 301 p2 = *EVAL(p2) = f1 + size2 - list1 + list2; 302 if (sense > 0) 303 swap (f1, f1 + size); 304 } while ((f1 += size2) < f2); 305 } else { /* Natural merge */ 306 l2 = f2; 307 for (f2 = f1 + size2; f2 < l2; f2 += size2) { 308 if ((cmp(f2-size, f2) > 0) != sense) { 309 p2 = *EVAL(p2) = f2 - list1 + list2; 310 if (sense > 0) 311 reverse(f1, f2-size); 312 f1 = f2; 313 } 314 } 315 if (sense > 0) 316 reverse (f1, f2-size); 317 f1 = f2; 318 if (f2 < last || cmp(f2 - size, f2) > 0) 319 p2 = *EVAL(p2) = f2 - list1 + list2; 320 else 321 p2 = *EVAL(p2) = list2 + n*size; 322 } 323 } 324 #else /* pairwise merge only. */ 325 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { 326 p2 = *EVAL(p2) = p2 + size2; 327 if (cmp (f1, f1 + size) > 0) 328 swap(f1, f1 + size); 329 } 330 #endif /* NATURAL */ 331 } 332 333 /* 334 * This is to avoid out-of-bounds addresses in sorting the 335 * last 4 elements. 336 */ 337 static void 338 insertionsort(a, n, size, cmp) 339 u_char *a; 340 size_t n, size; 341 int (*cmp) __P((const void *, const void *)); 342 { 343 u_char *ai, *s, *t, *u, tmp; 344 int i; 345 346 for (ai = a+size; --n >= 1; ai += size) 347 for (t = ai; t > a; t -= size) { 348 u = t - size; 349 if (cmp(u, t) <= 0) 350 break; 351 swap(u, t); 352 } 353 } 354