11573Srgrimes/*- 21573Srgrimes * Copyright (c) 1992, 1993 31573Srgrimes * The Regents of the University of California. All rights reserved. 41573Srgrimes * 51573Srgrimes * This code is derived from software contributed to Berkeley by 61573Srgrimes * Peter McIlroy. 71573Srgrimes * 81573Srgrimes * Redistribution and use in source and binary forms, with or without 91573Srgrimes * modification, are permitted provided that the following conditions 101573Srgrimes * are met: 111573Srgrimes * 1. Redistributions of source code must retain the above copyright 121573Srgrimes * notice, this list of conditions and the following disclaimer. 131573Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 141573Srgrimes * notice, this list of conditions and the following disclaimer in the 151573Srgrimes * documentation and/or other materials provided with the distribution. 16251672Semaste * 3. Neither the name of the University nor the names of its contributors 171573Srgrimes * may be used to endorse or promote products derived from this software 181573Srgrimes * without specific prior written permission. 191573Srgrimes * 201573Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 211573Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 221573Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 231573Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 241573Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 251573Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 261573Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 271573Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 281573Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 291573Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 301573Srgrimes * SUCH DAMAGE. 311573Srgrimes */ 321573Srgrimes 331573Srgrimes#if defined(LIBC_SCCS) && !defined(lint) 341573Srgrimesstatic char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94"; 351573Srgrimes#endif /* LIBC_SCCS and not lint */ 3692889Sobrien#include <sys/cdefs.h> 3792889Sobrien__FBSDID("$FreeBSD$"); 381573Srgrimes 391573Srgrimes/* 401573Srgrimes * Hybrid exponential search/linear search merge sort with hybrid 411573Srgrimes * natural/pairwise first pass. Requires about .3% more comparisons 421573Srgrimes * for random data than LSMS with pairwise first pass alone. 431573Srgrimes * It works for objects as small as two bytes. 441573Srgrimes */ 451573Srgrimes 461573Srgrimes#define NATURAL 471573Srgrimes#define THRESHOLD 16 /* Best choice for natural merge cut-off. */ 481573Srgrimes 491573Srgrimes/* #define NATURAL to get hybrid natural merge. 501573Srgrimes * (The default is pairwise merging.) 511573Srgrimes */ 521573Srgrimes 531573Srgrimes#include <sys/types.h> 541573Srgrimes 551573Srgrimes#include <errno.h> 561573Srgrimes#include <stdlib.h> 571573Srgrimes#include <string.h> 581573Srgrimes 59140098Sbrianstatic void setup(u_char *, u_char *, size_t, size_t, 60140098Sbrian int (*)(const void *, const void *)); 61140098Sbrianstatic void insertionsort(u_char *, size_t, size_t, 62140098Sbrian int (*)(const void *, const void *)); 631573Srgrimes 641573Srgrimes#define ISIZE sizeof(int) 651573Srgrimes#define PSIZE sizeof(u_char *) 661573Srgrimes#define ICOPY_LIST(src, dst, last) \ 671573Srgrimes do \ 681573Srgrimes *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ 691573Srgrimes while(src < last) 701573Srgrimes#define ICOPY_ELT(src, dst, i) \ 711573Srgrimes do \ 721573Srgrimes *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ 731573Srgrimes while (i -= ISIZE) 741573Srgrimes 751573Srgrimes#define CCOPY_LIST(src, dst, last) \ 761573Srgrimes do \ 771573Srgrimes *dst++ = *src++; \ 781573Srgrimes while (src < last) 791573Srgrimes#define CCOPY_ELT(src, dst, i) \ 801573Srgrimes do \ 811573Srgrimes *dst++ = *src++; \ 821573Srgrimes while (i -= 1) 838870Srgrimes 841573Srgrimes/* 851573Srgrimes * Find the next possible pointer head. (Trickery for forcing an array 861573Srgrimes * to do double duty as a linked list when objects do not align with word 871573Srgrimes * boundaries. 881573Srgrimes */ 891573Srgrimes/* Assumption: PSIZE is a power of 2. */ 901573Srgrimes#define EVAL(p) (u_char **) \ 911573Srgrimes ((u_char *)0 + \ 921573Srgrimes (((u_char *)p + PSIZE - 1 - (u_char *) 0) & ~(PSIZE - 1))) 931573Srgrimes 941573Srgrimes/* 951573Srgrimes * Arguments are as for qsort. 961573Srgrimes */ 971573Srgrimesint 981573Srgrimesmergesort(base, nmemb, size, cmp) 991573Srgrimes void *base; 1001573Srgrimes size_t nmemb; 10192889Sobrien size_t size; 10292905Sobrien int (*cmp)(const void *, const void *); 1031573Srgrimes{ 104140098Sbrian size_t i; 105140098Sbrian int sense; 1061573Srgrimes int big, iflag; 10792889Sobrien u_char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; 1081573Srgrimes u_char *list2, *list1, *p2, *p, *last, **p1; 1091573Srgrimes 1101573Srgrimes if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ 1111573Srgrimes errno = EINVAL; 1121573Srgrimes return (-1); 1131573Srgrimes } 1141573Srgrimes 11540896Srnordier if (nmemb == 0) 11640896Srnordier return (0); 11740896Srnordier 1181573Srgrimes /* 1191573Srgrimes * XXX 1201573Srgrimes * Stupid subtraction for the Cray. 1211573Srgrimes */ 1221573Srgrimes iflag = 0; 1231573Srgrimes if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) 1241573Srgrimes iflag = 1; 1251573Srgrimes 1261573Srgrimes if ((list2 = malloc(nmemb * size + PSIZE)) == NULL) 1271573Srgrimes return (-1); 1281573Srgrimes 1291573Srgrimes list1 = base; 1301573Srgrimes setup(list1, list2, nmemb, size, cmp); 1311573Srgrimes last = list2 + nmemb * size; 1321573Srgrimes i = big = 0; 1331573Srgrimes while (*EVAL(list2) != last) { 1341573Srgrimes l2 = list1; 1351573Srgrimes p1 = EVAL(list1); 1361573Srgrimes for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { 1371573Srgrimes p2 = *EVAL(p2); 1381573Srgrimes f1 = l2; 1391573Srgrimes f2 = l1 = list1 + (p2 - list2); 1401573Srgrimes if (p2 != last) 1411573Srgrimes p2 = *EVAL(p2); 1421573Srgrimes l2 = list1 + (p2 - list2); 1431573Srgrimes while (f1 < l1 && f2 < l2) { 1441573Srgrimes if ((*cmp)(f1, f2) <= 0) { 1451573Srgrimes q = f2; 1461573Srgrimes b = f1, t = l1; 1471573Srgrimes sense = -1; 1481573Srgrimes } else { 1491573Srgrimes q = f1; 1501573Srgrimes b = f2, t = l2; 1511573Srgrimes sense = 0; 1521573Srgrimes } 1531573Srgrimes if (!big) { /* here i = 0 */ 15440900Srnordier while ((b += size) < t && cmp(q, b) >sense) 1551573Srgrimes if (++i == 6) { 1561573Srgrimes big = 1; 1571573Srgrimes goto EXPONENTIAL; 1581573Srgrimes } 1591573Srgrimes } else { 1601573SrgrimesEXPONENTIAL: for (i = size; ; i <<= 1) 1611573Srgrimes if ((p = (b + i)) >= t) { 1621573Srgrimes if ((p = t - size) > b && 1631573Srgrimes (*cmp)(q, p) <= sense) 1641573Srgrimes t = p; 1651573Srgrimes else 1661573Srgrimes b = p; 1671573Srgrimes break; 1681573Srgrimes } else if ((*cmp)(q, p) <= sense) { 1691573Srgrimes t = p; 1701573Srgrimes if (i == size) 1718870Srgrimes big = 0; 1721573Srgrimes goto FASTCASE; 1731573Srgrimes } else 1741573Srgrimes b = p; 17540900Srnordier while (t > b+size) { 1761573Srgrimes i = (((t - b) / size) >> 1) * size; 1771573Srgrimes if ((*cmp)(q, p = b + i) <= sense) 1781573Srgrimes t = p; 1791573Srgrimes else 1801573Srgrimes b = p; 1811573Srgrimes } 1821573Srgrimes goto COPY; 1831573SrgrimesFASTCASE: while (i > size) 1841573Srgrimes if ((*cmp)(q, 1851573Srgrimes p = b + (i >>= 1)) <= sense) 1861573Srgrimes t = p; 1871573Srgrimes else 1881573Srgrimes b = p; 1891573SrgrimesCOPY: b = t; 1901573Srgrimes } 1911573Srgrimes i = size; 1921573Srgrimes if (q == f1) { 1931573Srgrimes if (iflag) { 1941573Srgrimes ICOPY_LIST(f2, tp2, b); 1951573Srgrimes ICOPY_ELT(f1, tp2, i); 1961573Srgrimes } else { 1971573Srgrimes CCOPY_LIST(f2, tp2, b); 1981573Srgrimes CCOPY_ELT(f1, tp2, i); 1991573Srgrimes } 2001573Srgrimes } else { 2011573Srgrimes if (iflag) { 2021573Srgrimes ICOPY_LIST(f1, tp2, b); 2031573Srgrimes ICOPY_ELT(f2, tp2, i); 2041573Srgrimes } else { 2051573Srgrimes CCOPY_LIST(f1, tp2, b); 2061573Srgrimes CCOPY_ELT(f2, tp2, i); 2071573Srgrimes } 2081573Srgrimes } 2091573Srgrimes } 2101573Srgrimes if (f2 < l2) { 2111573Srgrimes if (iflag) 2121573Srgrimes ICOPY_LIST(f2, tp2, l2); 2131573Srgrimes else 2141573Srgrimes CCOPY_LIST(f2, tp2, l2); 2151573Srgrimes } else if (f1 < l1) { 2161573Srgrimes if (iflag) 2171573Srgrimes ICOPY_LIST(f1, tp2, l1); 2181573Srgrimes else 2191573Srgrimes CCOPY_LIST(f1, tp2, l1); 2201573Srgrimes } 2211573Srgrimes *p1 = l2; 2221573Srgrimes } 2231573Srgrimes tp2 = list1; /* swap list1, list2 */ 2241573Srgrimes list1 = list2; 2251573Srgrimes list2 = tp2; 2261573Srgrimes last = list2 + nmemb*size; 2271573Srgrimes } 2281573Srgrimes if (base == list2) { 2291573Srgrimes memmove(list2, list1, nmemb*size); 2301573Srgrimes list2 = list1; 2311573Srgrimes } 2321573Srgrimes free(list2); 2331573Srgrimes return (0); 2341573Srgrimes} 2351573Srgrimes 2361573Srgrimes#define swap(a, b) { \ 2371573Srgrimes s = b; \ 2381573Srgrimes i = size; \ 2391573Srgrimes do { \ 2401573Srgrimes tmp = *a; *a++ = *s; *s++ = tmp; \ 2411573Srgrimes } while (--i); \ 2421573Srgrimes a -= size; \ 2431573Srgrimes } 2441573Srgrimes#define reverse(bot, top) { \ 2451573Srgrimes s = top; \ 2461573Srgrimes do { \ 2471573Srgrimes i = size; \ 2481573Srgrimes do { \ 2491573Srgrimes tmp = *bot; *bot++ = *s; *s++ = tmp; \ 2501573Srgrimes } while (--i); \ 2511573Srgrimes s -= size2; \ 2521573Srgrimes } while(bot < s); \ 2531573Srgrimes} 2541573Srgrimes 2551573Srgrimes/* 2561573Srgrimes * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of 2571573Srgrimes * increasing order, list2 in a corresponding linked list. Checks for runs 2581573Srgrimes * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL 2591573Srgrimes * is defined. Otherwise simple pairwise merging is used.) 2601573Srgrimes */ 2611573Srgrimesvoid 2621573Srgrimessetup(list1, list2, n, size, cmp) 2631573Srgrimes size_t n, size; 26492905Sobrien int (*cmp)(const void *, const void *); 2651573Srgrimes u_char *list1, *list2; 2661573Srgrimes{ 2671573Srgrimes int i, length, size2, tmp, sense; 2681573Srgrimes u_char *f1, *f2, *s, *l2, *last, *p2; 2691573Srgrimes 2701573Srgrimes size2 = size*2; 2711573Srgrimes if (n <= 5) { 2721573Srgrimes insertionsort(list1, n, size, cmp); 2731573Srgrimes *EVAL(list2) = (u_char*) list2 + n*size; 2741573Srgrimes return; 2751573Srgrimes } 2761573Srgrimes /* 2771573Srgrimes * Avoid running pointers out of bounds; limit n to evens 2781573Srgrimes * for simplicity. 2791573Srgrimes */ 2801573Srgrimes i = 4 + (n & 1); 2811573Srgrimes insertionsort(list1 + (n - i) * size, i, size, cmp); 2821573Srgrimes last = list1 + size * (n - i); 2831573Srgrimes *EVAL(list2 + (last - list1)) = list2 + n * size; 2841573Srgrimes 2851573Srgrimes#ifdef NATURAL 2861573Srgrimes p2 = list2; 2871573Srgrimes f1 = list1; 2881573Srgrimes sense = (cmp(f1, f1 + size) > 0); 2891573Srgrimes for (; f1 < last; sense = !sense) { 2901573Srgrimes length = 2; 2911573Srgrimes /* Find pairs with same sense. */ 2921573Srgrimes for (f2 = f1 + size2; f2 < last; f2 += size2) { 2931573Srgrimes if ((cmp(f2, f2+ size) > 0) != sense) 2941573Srgrimes break; 2951573Srgrimes length += 2; 2961573Srgrimes } 2971573Srgrimes if (length < THRESHOLD) { /* Pairwise merge */ 2981573Srgrimes do { 2991573Srgrimes p2 = *EVAL(p2) = f1 + size2 - list1 + list2; 3001573Srgrimes if (sense > 0) 3011573Srgrimes swap (f1, f1 + size); 3021573Srgrimes } while ((f1 += size2) < f2); 3031573Srgrimes } else { /* Natural merge */ 3041573Srgrimes l2 = f2; 3051573Srgrimes for (f2 = f1 + size2; f2 < l2; f2 += size2) { 3061573Srgrimes if ((cmp(f2-size, f2) > 0) != sense) { 3071573Srgrimes p2 = *EVAL(p2) = f2 - list1 + list2; 3081573Srgrimes if (sense > 0) 3091573Srgrimes reverse(f1, f2-size); 3101573Srgrimes f1 = f2; 3111573Srgrimes } 3121573Srgrimes } 3131573Srgrimes if (sense > 0) 3141573Srgrimes reverse (f1, f2-size); 3151573Srgrimes f1 = f2; 3161573Srgrimes if (f2 < last || cmp(f2 - size, f2) > 0) 3171573Srgrimes p2 = *EVAL(p2) = f2 - list1 + list2; 3181573Srgrimes else 3191573Srgrimes p2 = *EVAL(p2) = list2 + n*size; 3201573Srgrimes } 3211573Srgrimes } 3221573Srgrimes#else /* pairwise merge only. */ 3231573Srgrimes for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { 3241573Srgrimes p2 = *EVAL(p2) = p2 + size2; 3251573Srgrimes if (cmp (f1, f1 + size) > 0) 3261573Srgrimes swap(f1, f1 + size); 3271573Srgrimes } 3281573Srgrimes#endif /* NATURAL */ 3291573Srgrimes} 3301573Srgrimes 3311573Srgrimes/* 3321573Srgrimes * This is to avoid out-of-bounds addresses in sorting the 3331573Srgrimes * last 4 elements. 3341573Srgrimes */ 3351573Srgrimesstatic void 3361573Srgrimesinsertionsort(a, n, size, cmp) 3371573Srgrimes u_char *a; 3381573Srgrimes size_t n, size; 33992905Sobrien int (*cmp)(const void *, const void *); 3401573Srgrimes{ 3411573Srgrimes u_char *ai, *s, *t, *u, tmp; 3421573Srgrimes int i; 3431573Srgrimes 3441573Srgrimes for (ai = a+size; --n >= 1; ai += size) 3451573Srgrimes for (t = ai; t > a; t -= size) { 3461573Srgrimes u = t - size; 3471573Srgrimes if (cmp(u, t) <= 0) 3481573Srgrimes break; 3491573Srgrimes swap(u, t); 3501573Srgrimes } 3511573Srgrimes} 352