1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 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. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 #if defined(LIBC_SCCS) && !defined(lint) 36 static char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94"; 37 #endif /* LIBC_SCCS and not lint */ 38 #include <sys/cdefs.h> 39 __FBSDID("$FreeBSD$"); 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/param.h> 56 57 #include <errno.h> 58 #include <stdlib.h> 59 #include <string.h> 60 61 #ifdef I_AM_MERGESORT_B 62 #include "block_abi.h" 63 #define DECLARE_CMP DECLARE_BLOCK(int, cmp, const void *, const void *) 64 typedef DECLARE_BLOCK(int, cmp_t, const void *, const void *); 65 #define CMP(x, y) CALL_BLOCK(cmp, x, y) 66 #else 67 typedef int (*cmp_t)(const void *, const void *); 68 #define CMP(x, y) cmp(x, y) 69 #endif 70 71 static void setup(u_char *, u_char *, size_t, size_t, cmp_t); 72 static void insertionsort(u_char *, size_t, size_t, cmp_t); 73 74 #define ISIZE sizeof(int) 75 #define PSIZE sizeof(u_char *) 76 #define ICOPY_LIST(src, dst, last) \ 77 do \ 78 *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ 79 while(src < last) 80 #define ICOPY_ELT(src, dst, i) \ 81 do \ 82 *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ 83 while (i -= ISIZE) 84 85 #define CCOPY_LIST(src, dst, last) \ 86 do \ 87 *dst++ = *src++; \ 88 while (src < last) 89 #define CCOPY_ELT(src, dst, i) \ 90 do \ 91 *dst++ = *src++; \ 92 while (i -= 1) 93 94 /* 95 * Find the next possible pointer head. (Trickery for forcing an array 96 * to do double duty as a linked list when objects do not align with word 97 * boundaries. 98 */ 99 /* Assumption: PSIZE is a power of 2. */ 100 #define EVAL(p) (u_char **)roundup2((uintptr_t)p, PSIZE) 101 102 #ifdef I_AM_MERGESORT_B 103 int mergesort_b(void *, size_t, size_t, cmp_t); 104 #else 105 int mergesort(void *, size_t, size_t, cmp_t); 106 #endif 107 108 /* 109 * Arguments are as for qsort. 110 */ 111 int 112 #ifdef I_AM_MERGESORT_B 113 mergesort_b(void *base, size_t nmemb, size_t size, cmp_t cmp) 114 #else 115 mergesort(void *base, size_t nmemb, size_t size, cmp_t cmp) 116 #endif 117 { 118 size_t i; 119 int sense; 120 int big, iflag; 121 u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; 122 u_char *list2, *list1, *p2, *p, *last, **p1; 123 124 if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ 125 errno = EINVAL; 126 return (-1); 127 } 128 129 if (nmemb == 0) 130 return (0); 131 132 /* 133 * XXX 134 * Stupid subtraction for the Cray. 135 */ 136 iflag = 0; 137 if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) 138 iflag = 1; 139 140 if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) 141 return (-1); 142 143 list1 = base; 144 setup(list1, list2, nmemb, size, cmp); 145 last = list2 + nmemb * size; 146 i = big = 0; 147 while (*EVAL(list2) != last) { 148 l2 = list1; 149 p1 = EVAL(list1); 150 for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { 151 p2 = *EVAL(p2); 152 f1 = l2; 153 f2 = l1 = list1 + (p2 - list2); 154 if (p2 != last) 155 p2 = *EVAL(p2); 156 l2 = list1 + (p2 - list2); 157 while (f1 < l1 && f2 < l2) { 158 if (CMP(f1, f2) <= 0) { 159 q = f2; 160 b = f1, t = l1; 161 sense = -1; 162 } else { 163 q = f1; 164 b = f2, t = l2; 165 sense = 0; 166 } 167 if (!big) { /* here i = 0 */ 168 while ((b += size) < t && CMP(q, b) >sense) 169 if (++i == 6) { 170 big = 1; 171 goto EXPONENTIAL; 172 } 173 } else { 174 EXPONENTIAL: for (i = size; ; i <<= 1) 175 if ((p = (b + i)) >= t) { 176 if ((p = t - size) > b && 177 CMP(q, p) <= sense) 178 t = p; 179 else 180 b = p; 181 break; 182 } else if (CMP(q, p) <= sense) { 183 t = p; 184 if (i == size) 185 big = 0; 186 goto FASTCASE; 187 } else 188 b = p; 189 while (t > b+size) { 190 i = (((t - b) / size) >> 1) * size; 191 if (CMP(q, p = b + i) <= sense) 192 t = p; 193 else 194 b = p; 195 } 196 goto COPY; 197 FASTCASE: while (i > size) 198 if (CMP(q, 199 p = b + (i >>= 1)) <= sense) 200 t = p; 201 else 202 b = p; 203 COPY: b = t; 204 } 205 i = size; 206 if (q == f1) { 207 if (iflag) { 208 ICOPY_LIST(f2, tp2, b); 209 ICOPY_ELT(f1, tp2, i); 210 } else { 211 CCOPY_LIST(f2, tp2, b); 212 CCOPY_ELT(f1, tp2, i); 213 } 214 } else { 215 if (iflag) { 216 ICOPY_LIST(f1, tp2, b); 217 ICOPY_ELT(f2, tp2, i); 218 } else { 219 CCOPY_LIST(f1, tp2, b); 220 CCOPY_ELT(f2, tp2, i); 221 } 222 } 223 } 224 if (f2 < l2) { 225 if (iflag) 226 ICOPY_LIST(f2, tp2, l2); 227 else 228 CCOPY_LIST(f2, tp2, l2); 229 } else if (f1 < l1) { 230 if (iflag) 231 ICOPY_LIST(f1, tp2, l1); 232 else 233 CCOPY_LIST(f1, tp2, l1); 234 } 235 *p1 = l2; 236 } 237 tp2 = list1; /* swap list1, list2 */ 238 list1 = list2; 239 list2 = tp2; 240 last = list2 + nmemb*size; 241 } 242 if (base == list2) { 243 memmove(list2, list1, nmemb*size); 244 list2 = list1; 245 } 246 free(list2); 247 return (0); 248 } 249 250 #define swap(a, b) { \ 251 s = b; \ 252 i = size; \ 253 do { \ 254 tmp = *a; *a++ = *s; *s++ = tmp; \ 255 } while (--i); \ 256 a -= size; \ 257 } 258 #define reverse(bot, top) { \ 259 s = top; \ 260 do { \ 261 i = size; \ 262 do { \ 263 tmp = *bot; *bot++ = *s; *s++ = tmp; \ 264 } while (--i); \ 265 s -= size2; \ 266 } while(bot < s); \ 267 } 268 269 /* 270 * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of 271 * increasing order, list2 in a corresponding linked list. Checks for runs 272 * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL 273 * is defined. Otherwise simple pairwise merging is used.) 274 */ 275 void 276 setup(u_char *list1, u_char *list2, size_t n, size_t size, cmp_t cmp) 277 { 278 int i, length, size2, tmp, sense; 279 u_char *f1, *f2, *s, *l2, *last, *p2; 280 281 size2 = size*2; 282 if (n <= 5) { 283 insertionsort(list1, n, size, cmp); 284 *EVAL(list2) = (u_char*) list2 + n*size; 285 return; 286 } 287 /* 288 * Avoid running pointers out of bounds; limit n to evens 289 * for simplicity. 290 */ 291 i = 4 + (n & 1); 292 insertionsort(list1 + (n - i) * size, i, size, cmp); 293 last = list1 + size * (n - i); 294 *EVAL(list2 + (last - list1)) = list2 + n * size; 295 296 #ifdef NATURAL 297 p2 = list2; 298 f1 = list1; 299 sense = (CMP(f1, f1 + size) > 0); 300 for (; f1 < last; sense = !sense) { 301 length = 2; 302 /* Find pairs with same sense. */ 303 for (f2 = f1 + size2; f2 < last; f2 += size2) { 304 if ((CMP(f2, f2+ size) > 0) != sense) 305 break; 306 length += 2; 307 } 308 if (length < THRESHOLD) { /* Pairwise merge */ 309 do { 310 p2 = *EVAL(p2) = f1 + size2 - list1 + list2; 311 if (sense > 0) 312 swap (f1, f1 + size); 313 } while ((f1 += size2) < f2); 314 } else { /* Natural merge */ 315 l2 = f2; 316 for (f2 = f1 + size2; f2 < l2; f2 += size2) { 317 if ((CMP(f2-size, f2) > 0) != sense) { 318 p2 = *EVAL(p2) = f2 - list1 + list2; 319 if (sense > 0) 320 reverse(f1, f2-size); 321 f1 = f2; 322 } 323 } 324 if (sense > 0) 325 reverse (f1, f2-size); 326 f1 = f2; 327 if (f2 < last || CMP(f2 - size, f2) > 0) 328 p2 = *EVAL(p2) = f2 - list1 + list2; 329 else 330 p2 = *EVAL(p2) = list2 + n*size; 331 } 332 } 333 #else /* pairwise merge only. */ 334 for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { 335 p2 = *EVAL(p2) = p2 + size2; 336 if (CMP (f1, f1 + size) > 0) 337 swap(f1, f1 + size); 338 } 339 #endif /* NATURAL */ 340 } 341 342 /* 343 * This is to avoid out-of-bounds addresses in sorting the 344 * last 4 elements. 345 */ 346 static void 347 insertionsort(u_char *a, size_t n, size_t size, cmp_t cmp) 348 { 349 u_char *ai, *s, *t, *u, tmp; 350 int i; 351 352 for (ai = a+size; --n >= 1; ai += size) 353 for (t = ai; t > a; t -= size) { 354 u = t - size; 355 if (CMP(u, t) <= 0) 356 break; 357 swap(u, t); 358 } 359 } 360