From: Peter Johnson Date: Fri, 26 Oct 2001 04:32:18 +0000 (-0000) Subject: Add mergesort() replacement from BSD sources. X-Git-Tag: v0.1.0~238 X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=0a98a7416389e0bfaef6d83833aec2fb288039f5;p=yasm Add mergesort() replacement from BSD sources. svn path=/trunk/yasm/; revision=292 --- diff --git a/configure.ac b/configure.ac index f8a1803b..d2d19677 100644 --- a/configure.ac +++ b/configure.ac @@ -117,7 +117,7 @@ AC_TYPE_SIZE_T AC_FUNC_VPRINTF AC_CHECK_FUNCS(memcpy toascii abort) AC_CHECK_FUNCS(strcasecmp stricmp strcmpi, break) -AC_REPLACE_FUNCS(strsep) +AC_REPLACE_FUNCS(strsep mergesort) AC_CHECK_HEADERS(limits.h sys/queue.h sys/cdefs.h) diff --git a/configure.in b/configure.in index f8a1803b..d2d19677 100644 --- a/configure.in +++ b/configure.in @@ -117,7 +117,7 @@ AC_TYPE_SIZE_T AC_FUNC_VPRINTF AC_CHECK_FUNCS(memcpy toascii abort) AC_CHECK_FUNCS(strcasecmp stricmp strcmpi, break) -AC_REPLACE_FUNCS(strsep) +AC_REPLACE_FUNCS(strsep mergesort) AC_CHECK_HEADERS(limits.h sys/queue.h sys/cdefs.h) diff --git a/libyasm/mergesort.c b/libyasm/mergesort.c new file mode 100644 index 00000000..89d71235 --- /dev/null +++ b/libyasm/mergesort.c @@ -0,0 +1,351 @@ +/* $IdPath$ + * mergesort() implementation for systems that don't have it. + * + * Copyright (c) 1992, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Peter McIlroy. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "util.h" + +RCSID("$IdPath$"); + +#if defined(LIBC_SCCS) && !defined(lint) +static char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94"; +#endif /* LIBC_SCCS and not lint */ + +/* + * Hybrid exponential search/linear search merge sort with hybrid + * natural/pairwise first pass. Requires about .3% more comparisons + * for random data than LSMS with pairwise first pass alone. + * It works for objects as small as two bytes. + */ + +#define NATURAL +#define THRESHOLD 16 /* Best choice for natural merge cut-off. */ + +/* #define NATURAL to get hybrid natural merge. + * (The default is pairwise merging.) + */ + +#ifdef STDC_HEADERS +# include +# include +# include +#endif + +static void setup(unsigned char *, unsigned char *, size_t, size_t, + int (*)(const void *, const void *)); +static void insertionsort(unsigned char *, size_t, size_t, + int (*)(const void *, const void *)); + +#define ISIZE sizeof(int) +#define PSIZE sizeof(unsigned char *) +#define ICOPY_LIST(src, dst, last) \ + do \ + *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ + while(src < last) +#define ICOPY_ELT(src, dst, i) \ + do \ + *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ + while (i -= ISIZE) + +#define CCOPY_LIST(src, dst, last) \ + do \ + *dst++ = *src++; \ + while (src < last) +#define CCOPY_ELT(src, dst, i) \ + do \ + *dst++ = *src++; \ + while (i -= 1) + +/* + * Find the next possible pointer head. (Trickery for forcing an array + * to do double duty as a linked list when objects do not align with word + * boundaries. + */ +/* Assumption: PSIZE is a power of 2. */ +#define EVAL(p) (unsigned char **) \ + ((unsigned char *)0 + \ + (((unsigned char *)p + PSIZE - 1 - (unsigned char *) 0) & ~(PSIZE - 1))) + +/* + * Arguments are as for qsort. + */ +int +mergesort(void *base, size_t nmemb, size_t size, + int (*cmp)(const void *, const void *)) +{ + int i, sense; + int big, iflag; + unsigned char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; + unsigned char *list2, *list1, *p2, *p, *last, **p1; + + if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ + errno = EINVAL; + return (-1); + } + + if (nmemb == 0) + return (0); + + /* + * XXX + * Stupid subtraction for the Cray. + */ + iflag = 0; + if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) + iflag = 1; + + if ((list2 = xmalloc(nmemb * size + PSIZE)) == NULL) + return (-1); + + list1 = base; + setup(list1, list2, nmemb, size, cmp); + last = list2 + nmemb * size; + i = big = 0; + while (*EVAL(list2) != last) { + l2 = list1; + p1 = EVAL(list1); + for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { + p2 = *EVAL(p2); + f1 = l2; + f2 = l1 = list1 + (p2 - list2); + if (p2 != last) + p2 = *EVAL(p2); + l2 = list1 + (p2 - list2); + while (f1 < l1 && f2 < l2) { + if ((*cmp)(f1, f2) <= 0) { + q = f2; + b = f1, t = l1; + sense = -1; + } else { + q = f1; + b = f2, t = l2; + sense = 0; + } + if (!big) { /* here i = 0 */ + while ((b += size) < t && cmp(q, b) >sense) + if (++i == 6) { + big = 1; + goto EXPONENTIAL; + } + } else { +EXPONENTIAL: for (i = size; ; i <<= 1) + if ((p = (b + i)) >= t) { + if ((p = t - size) > b && + (*cmp)(q, p) <= sense) + t = p; + else + b = p; + break; + } else if ((*cmp)(q, p) <= sense) { + t = p; + if (i == size) + big = 0; + goto FASTCASE; + } else + b = p; + while (t > b+size) { + i = (((t - b) / size) >> 1) * size; + if ((*cmp)(q, p = b + i) <= sense) + t = p; + else + b = p; + } + goto COPY; +FASTCASE: while (i > size) + if ((*cmp)(q, + p = b + (i >>= 1)) <= sense) + t = p; + else + b = p; +COPY: b = t; + } + i = size; + if (q == f1) { + if (iflag) { + ICOPY_LIST(f2, tp2, b); + ICOPY_ELT(f1, tp2, i); + } else { + CCOPY_LIST(f2, tp2, b); + CCOPY_ELT(f1, tp2, i); + } + } else { + if (iflag) { + ICOPY_LIST(f1, tp2, b); + ICOPY_ELT(f2, tp2, i); + } else { + CCOPY_LIST(f1, tp2, b); + CCOPY_ELT(f2, tp2, i); + } + } + } + if (f2 < l2) { + if (iflag) + ICOPY_LIST(f2, tp2, l2); + else + CCOPY_LIST(f2, tp2, l2); + } else if (f1 < l1) { + if (iflag) + ICOPY_LIST(f1, tp2, l1); + else + CCOPY_LIST(f1, tp2, l1); + } + *p1 = l2; + } + tp2 = list1; /* swap list1, list2 */ + list1 = list2; + list2 = tp2; + last = list2 + nmemb*size; + } + if (base == list2) { + memmove(list2, list1, nmemb*size); + list2 = list1; + } + free(list2); + return (0); +} + +#define swap(a, b) { \ + s = b; \ + i = size; \ + do { \ + tmp = *a; *a++ = *s; *s++ = tmp; \ + } while (--i); \ + a -= size; \ + } +#define reverse(bot, top) { \ + s = top; \ + do { \ + i = size; \ + do { \ + tmp = *bot; *bot++ = *s; *s++ = tmp; \ + } while (--i); \ + s -= size2; \ + } while(bot < s); \ +} + +/* + * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of + * increasing order, list2 in a corresponding linked list. Checks for runs + * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL + * is defined. Otherwise simple pairwise merging is used.) + */ +void +setup(unsigned char *list1, unsigned char *list2, size_t n, size_t size, + int (*cmp)(const void *, const void *)) +{ + int i, length, size2, tmp, sense; + unsigned char *f1, *f2, *s, *l2, *last, *p2; + + size2 = size*2; + if (n <= 5) { + insertionsort(list1, n, size, cmp); + *EVAL(list2) = (unsigned char*) list2 + n*size; + return; + } + /* + * Avoid running pointers out of bounds; limit n to evens + * for simplicity. + */ + i = 4 + (n & 1); + insertionsort(list1 + (n - i) * size, i, size, cmp); + last = list1 + size * (n - i); + *EVAL(list2 + (last - list1)) = list2 + n * size; + +#ifdef NATURAL + p2 = list2; + f1 = list1; + sense = (cmp(f1, f1 + size) > 0); + for (; f1 < last; sense = !sense) { + length = 2; + /* Find pairs with same sense. */ + for (f2 = f1 + size2; f2 < last; f2 += size2) { + if ((cmp(f2, f2+ size) > 0) != sense) + break; + length += 2; + } + if (length < THRESHOLD) { /* Pairwise merge */ + do { + p2 = *EVAL(p2) = f1 + size2 - list1 + list2; + if (sense > 0) + swap (f1, f1 + size); + } while ((f1 += size2) < f2); + } else { /* Natural merge */ + l2 = f2; + for (f2 = f1 + size2; f2 < l2; f2 += size2) { + if ((cmp(f2-size, f2) > 0) != sense) { + p2 = *EVAL(p2) = f2 - list1 + list2; + if (sense > 0) + reverse(f1, f2-size); + f1 = f2; + } + } + if (sense > 0) + reverse (f1, f2-size); + f1 = f2; + if (f2 < last || cmp(f2 - size, f2) > 0) + p2 = *EVAL(p2) = f2 - list1 + list2; + else + p2 = *EVAL(p2) = list2 + n*size; + } + } +#else /* pairwise merge only. */ + for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { + p2 = *EVAL(p2) = p2 + size2; + if (cmp (f1, f1 + size) > 0) + swap(f1, f1 + size); + } +#endif /* NATURAL */ +} + +/* + * This is to avoid out-of-bounds addresses in sorting the + * last 4 elements. + */ +static void +insertionsort(unsigned char *a, size_t n, size_t size, + int (*cmp)(const void *, const void *)) +{ + unsigned char *ai, *s, *t, *u, tmp; + int i; + + for (ai = a+size; --n >= 1; ai += size) + for (t = ai; t > a; t -= size) { + u = t - size; + if (cmp(u, t) <= 0) + break; + swap(u, t); + } +} diff --git a/libyasm/util.h b/libyasm/util.h index 116a1c83..0fa3f546 100644 --- a/libyasm/util.h +++ b/libyasm/util.h @@ -29,6 +29,11 @@ /* strdup() implementation with error checking (using xmalloc). */ char *xstrdup(const char *str); +#if !defined(HAVE_MERGESORT) +int mergesort(void *base, size_t nmemb, size_t size, + int (*compar)(const void *, const void *)); +#endif + #if !defined(HAVE_STRSEP) || defined(HAVE_GNU_C_LIBRARY) char *strsep(char **stringp, const char *delim); #endif diff --git a/mergesort.c b/mergesort.c new file mode 100644 index 00000000..89d71235 --- /dev/null +++ b/mergesort.c @@ -0,0 +1,351 @@ +/* $IdPath$ + * mergesort() implementation for systems that don't have it. + * + * Copyright (c) 1992, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Peter McIlroy. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "util.h" + +RCSID("$IdPath$"); + +#if defined(LIBC_SCCS) && !defined(lint) +static char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94"; +#endif /* LIBC_SCCS and not lint */ + +/* + * Hybrid exponential search/linear search merge sort with hybrid + * natural/pairwise first pass. Requires about .3% more comparisons + * for random data than LSMS with pairwise first pass alone. + * It works for objects as small as two bytes. + */ + +#define NATURAL +#define THRESHOLD 16 /* Best choice for natural merge cut-off. */ + +/* #define NATURAL to get hybrid natural merge. + * (The default is pairwise merging.) + */ + +#ifdef STDC_HEADERS +# include +# include +# include +#endif + +static void setup(unsigned char *, unsigned char *, size_t, size_t, + int (*)(const void *, const void *)); +static void insertionsort(unsigned char *, size_t, size_t, + int (*)(const void *, const void *)); + +#define ISIZE sizeof(int) +#define PSIZE sizeof(unsigned char *) +#define ICOPY_LIST(src, dst, last) \ + do \ + *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ + while(src < last) +#define ICOPY_ELT(src, dst, i) \ + do \ + *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ + while (i -= ISIZE) + +#define CCOPY_LIST(src, dst, last) \ + do \ + *dst++ = *src++; \ + while (src < last) +#define CCOPY_ELT(src, dst, i) \ + do \ + *dst++ = *src++; \ + while (i -= 1) + +/* + * Find the next possible pointer head. (Trickery for forcing an array + * to do double duty as a linked list when objects do not align with word + * boundaries. + */ +/* Assumption: PSIZE is a power of 2. */ +#define EVAL(p) (unsigned char **) \ + ((unsigned char *)0 + \ + (((unsigned char *)p + PSIZE - 1 - (unsigned char *) 0) & ~(PSIZE - 1))) + +/* + * Arguments are as for qsort. + */ +int +mergesort(void *base, size_t nmemb, size_t size, + int (*cmp)(const void *, const void *)) +{ + int i, sense; + int big, iflag; + unsigned char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; + unsigned char *list2, *list1, *p2, *p, *last, **p1; + + if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ + errno = EINVAL; + return (-1); + } + + if (nmemb == 0) + return (0); + + /* + * XXX + * Stupid subtraction for the Cray. + */ + iflag = 0; + if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) + iflag = 1; + + if ((list2 = xmalloc(nmemb * size + PSIZE)) == NULL) + return (-1); + + list1 = base; + setup(list1, list2, nmemb, size, cmp); + last = list2 + nmemb * size; + i = big = 0; + while (*EVAL(list2) != last) { + l2 = list1; + p1 = EVAL(list1); + for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { + p2 = *EVAL(p2); + f1 = l2; + f2 = l1 = list1 + (p2 - list2); + if (p2 != last) + p2 = *EVAL(p2); + l2 = list1 + (p2 - list2); + while (f1 < l1 && f2 < l2) { + if ((*cmp)(f1, f2) <= 0) { + q = f2; + b = f1, t = l1; + sense = -1; + } else { + q = f1; + b = f2, t = l2; + sense = 0; + } + if (!big) { /* here i = 0 */ + while ((b += size) < t && cmp(q, b) >sense) + if (++i == 6) { + big = 1; + goto EXPONENTIAL; + } + } else { +EXPONENTIAL: for (i = size; ; i <<= 1) + if ((p = (b + i)) >= t) { + if ((p = t - size) > b && + (*cmp)(q, p) <= sense) + t = p; + else + b = p; + break; + } else if ((*cmp)(q, p) <= sense) { + t = p; + if (i == size) + big = 0; + goto FASTCASE; + } else + b = p; + while (t > b+size) { + i = (((t - b) / size) >> 1) * size; + if ((*cmp)(q, p = b + i) <= sense) + t = p; + else + b = p; + } + goto COPY; +FASTCASE: while (i > size) + if ((*cmp)(q, + p = b + (i >>= 1)) <= sense) + t = p; + else + b = p; +COPY: b = t; + } + i = size; + if (q == f1) { + if (iflag) { + ICOPY_LIST(f2, tp2, b); + ICOPY_ELT(f1, tp2, i); + } else { + CCOPY_LIST(f2, tp2, b); + CCOPY_ELT(f1, tp2, i); + } + } else { + if (iflag) { + ICOPY_LIST(f1, tp2, b); + ICOPY_ELT(f2, tp2, i); + } else { + CCOPY_LIST(f1, tp2, b); + CCOPY_ELT(f2, tp2, i); + } + } + } + if (f2 < l2) { + if (iflag) + ICOPY_LIST(f2, tp2, l2); + else + CCOPY_LIST(f2, tp2, l2); + } else if (f1 < l1) { + if (iflag) + ICOPY_LIST(f1, tp2, l1); + else + CCOPY_LIST(f1, tp2, l1); + } + *p1 = l2; + } + tp2 = list1; /* swap list1, list2 */ + list1 = list2; + list2 = tp2; + last = list2 + nmemb*size; + } + if (base == list2) { + memmove(list2, list1, nmemb*size); + list2 = list1; + } + free(list2); + return (0); +} + +#define swap(a, b) { \ + s = b; \ + i = size; \ + do { \ + tmp = *a; *a++ = *s; *s++ = tmp; \ + } while (--i); \ + a -= size; \ + } +#define reverse(bot, top) { \ + s = top; \ + do { \ + i = size; \ + do { \ + tmp = *bot; *bot++ = *s; *s++ = tmp; \ + } while (--i); \ + s -= size2; \ + } while(bot < s); \ +} + +/* + * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of + * increasing order, list2 in a corresponding linked list. Checks for runs + * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL + * is defined. Otherwise simple pairwise merging is used.) + */ +void +setup(unsigned char *list1, unsigned char *list2, size_t n, size_t size, + int (*cmp)(const void *, const void *)) +{ + int i, length, size2, tmp, sense; + unsigned char *f1, *f2, *s, *l2, *last, *p2; + + size2 = size*2; + if (n <= 5) { + insertionsort(list1, n, size, cmp); + *EVAL(list2) = (unsigned char*) list2 + n*size; + return; + } + /* + * Avoid running pointers out of bounds; limit n to evens + * for simplicity. + */ + i = 4 + (n & 1); + insertionsort(list1 + (n - i) * size, i, size, cmp); + last = list1 + size * (n - i); + *EVAL(list2 + (last - list1)) = list2 + n * size; + +#ifdef NATURAL + p2 = list2; + f1 = list1; + sense = (cmp(f1, f1 + size) > 0); + for (; f1 < last; sense = !sense) { + length = 2; + /* Find pairs with same sense. */ + for (f2 = f1 + size2; f2 < last; f2 += size2) { + if ((cmp(f2, f2+ size) > 0) != sense) + break; + length += 2; + } + if (length < THRESHOLD) { /* Pairwise merge */ + do { + p2 = *EVAL(p2) = f1 + size2 - list1 + list2; + if (sense > 0) + swap (f1, f1 + size); + } while ((f1 += size2) < f2); + } else { /* Natural merge */ + l2 = f2; + for (f2 = f1 + size2; f2 < l2; f2 += size2) { + if ((cmp(f2-size, f2) > 0) != sense) { + p2 = *EVAL(p2) = f2 - list1 + list2; + if (sense > 0) + reverse(f1, f2-size); + f1 = f2; + } + } + if (sense > 0) + reverse (f1, f2-size); + f1 = f2; + if (f2 < last || cmp(f2 - size, f2) > 0) + p2 = *EVAL(p2) = f2 - list1 + list2; + else + p2 = *EVAL(p2) = list2 + n*size; + } + } +#else /* pairwise merge only. */ + for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { + p2 = *EVAL(p2) = p2 + size2; + if (cmp (f1, f1 + size) > 0) + swap(f1, f1 + size); + } +#endif /* NATURAL */ +} + +/* + * This is to avoid out-of-bounds addresses in sorting the + * last 4 elements. + */ +static void +insertionsort(unsigned char *a, size_t n, size_t size, + int (*cmp)(const void *, const void *)) +{ + unsigned char *ai, *s, *t, *u, tmp; + int i; + + for (ai = a+size; --n >= 1; ai += size) + for (t = ai; t > a; t -= size) { + u = t - size; + if (cmp(u, t) <= 0) + break; + swap(u, t); + } +} diff --git a/src/mergesort.c b/src/mergesort.c new file mode 100644 index 00000000..89d71235 --- /dev/null +++ b/src/mergesort.c @@ -0,0 +1,351 @@ +/* $IdPath$ + * mergesort() implementation for systems that don't have it. + * + * Copyright (c) 1992, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Peter McIlroy. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "util.h" + +RCSID("$IdPath$"); + +#if defined(LIBC_SCCS) && !defined(lint) +static char sccsid[] = "@(#)merge.c 8.2 (Berkeley) 2/14/94"; +#endif /* LIBC_SCCS and not lint */ + +/* + * Hybrid exponential search/linear search merge sort with hybrid + * natural/pairwise first pass. Requires about .3% more comparisons + * for random data than LSMS with pairwise first pass alone. + * It works for objects as small as two bytes. + */ + +#define NATURAL +#define THRESHOLD 16 /* Best choice for natural merge cut-off. */ + +/* #define NATURAL to get hybrid natural merge. + * (The default is pairwise merging.) + */ + +#ifdef STDC_HEADERS +# include +# include +# include +#endif + +static void setup(unsigned char *, unsigned char *, size_t, size_t, + int (*)(const void *, const void *)); +static void insertionsort(unsigned char *, size_t, size_t, + int (*)(const void *, const void *)); + +#define ISIZE sizeof(int) +#define PSIZE sizeof(unsigned char *) +#define ICOPY_LIST(src, dst, last) \ + do \ + *(int*)dst = *(int*)src, src += ISIZE, dst += ISIZE; \ + while(src < last) +#define ICOPY_ELT(src, dst, i) \ + do \ + *(int*) dst = *(int*) src, src += ISIZE, dst += ISIZE; \ + while (i -= ISIZE) + +#define CCOPY_LIST(src, dst, last) \ + do \ + *dst++ = *src++; \ + while (src < last) +#define CCOPY_ELT(src, dst, i) \ + do \ + *dst++ = *src++; \ + while (i -= 1) + +/* + * Find the next possible pointer head. (Trickery for forcing an array + * to do double duty as a linked list when objects do not align with word + * boundaries. + */ +/* Assumption: PSIZE is a power of 2. */ +#define EVAL(p) (unsigned char **) \ + ((unsigned char *)0 + \ + (((unsigned char *)p + PSIZE - 1 - (unsigned char *) 0) & ~(PSIZE - 1))) + +/* + * Arguments are as for qsort. + */ +int +mergesort(void *base, size_t nmemb, size_t size, + int (*cmp)(const void *, const void *)) +{ + int i, sense; + int big, iflag; + unsigned char *f1, *f2, *t, *b, *tp2, *q, *l1, *l2; + unsigned char *list2, *list1, *p2, *p, *last, **p1; + + if (size < PSIZE / 2) { /* Pointers must fit into 2 * size. */ + errno = EINVAL; + return (-1); + } + + if (nmemb == 0) + return (0); + + /* + * XXX + * Stupid subtraction for the Cray. + */ + iflag = 0; + if (!(size % ISIZE) && !(((char *)base - (char *)0) % ISIZE)) + iflag = 1; + + if ((list2 = xmalloc(nmemb * size + PSIZE)) == NULL) + return (-1); + + list1 = base; + setup(list1, list2, nmemb, size, cmp); + last = list2 + nmemb * size; + i = big = 0; + while (*EVAL(list2) != last) { + l2 = list1; + p1 = EVAL(list1); + for (tp2 = p2 = list2; p2 != last; p1 = EVAL(l2)) { + p2 = *EVAL(p2); + f1 = l2; + f2 = l1 = list1 + (p2 - list2); + if (p2 != last) + p2 = *EVAL(p2); + l2 = list1 + (p2 - list2); + while (f1 < l1 && f2 < l2) { + if ((*cmp)(f1, f2) <= 0) { + q = f2; + b = f1, t = l1; + sense = -1; + } else { + q = f1; + b = f2, t = l2; + sense = 0; + } + if (!big) { /* here i = 0 */ + while ((b += size) < t && cmp(q, b) >sense) + if (++i == 6) { + big = 1; + goto EXPONENTIAL; + } + } else { +EXPONENTIAL: for (i = size; ; i <<= 1) + if ((p = (b + i)) >= t) { + if ((p = t - size) > b && + (*cmp)(q, p) <= sense) + t = p; + else + b = p; + break; + } else if ((*cmp)(q, p) <= sense) { + t = p; + if (i == size) + big = 0; + goto FASTCASE; + } else + b = p; + while (t > b+size) { + i = (((t - b) / size) >> 1) * size; + if ((*cmp)(q, p = b + i) <= sense) + t = p; + else + b = p; + } + goto COPY; +FASTCASE: while (i > size) + if ((*cmp)(q, + p = b + (i >>= 1)) <= sense) + t = p; + else + b = p; +COPY: b = t; + } + i = size; + if (q == f1) { + if (iflag) { + ICOPY_LIST(f2, tp2, b); + ICOPY_ELT(f1, tp2, i); + } else { + CCOPY_LIST(f2, tp2, b); + CCOPY_ELT(f1, tp2, i); + } + } else { + if (iflag) { + ICOPY_LIST(f1, tp2, b); + ICOPY_ELT(f2, tp2, i); + } else { + CCOPY_LIST(f1, tp2, b); + CCOPY_ELT(f2, tp2, i); + } + } + } + if (f2 < l2) { + if (iflag) + ICOPY_LIST(f2, tp2, l2); + else + CCOPY_LIST(f2, tp2, l2); + } else if (f1 < l1) { + if (iflag) + ICOPY_LIST(f1, tp2, l1); + else + CCOPY_LIST(f1, tp2, l1); + } + *p1 = l2; + } + tp2 = list1; /* swap list1, list2 */ + list1 = list2; + list2 = tp2; + last = list2 + nmemb*size; + } + if (base == list2) { + memmove(list2, list1, nmemb*size); + list2 = list1; + } + free(list2); + return (0); +} + +#define swap(a, b) { \ + s = b; \ + i = size; \ + do { \ + tmp = *a; *a++ = *s; *s++ = tmp; \ + } while (--i); \ + a -= size; \ + } +#define reverse(bot, top) { \ + s = top; \ + do { \ + i = size; \ + do { \ + tmp = *bot; *bot++ = *s; *s++ = tmp; \ + } while (--i); \ + s -= size2; \ + } while(bot < s); \ +} + +/* + * Optional hybrid natural/pairwise first pass. Eats up list1 in runs of + * increasing order, list2 in a corresponding linked list. Checks for runs + * when THRESHOLD/2 pairs compare with same sense. (Only used when NATURAL + * is defined. Otherwise simple pairwise merging is used.) + */ +void +setup(unsigned char *list1, unsigned char *list2, size_t n, size_t size, + int (*cmp)(const void *, const void *)) +{ + int i, length, size2, tmp, sense; + unsigned char *f1, *f2, *s, *l2, *last, *p2; + + size2 = size*2; + if (n <= 5) { + insertionsort(list1, n, size, cmp); + *EVAL(list2) = (unsigned char*) list2 + n*size; + return; + } + /* + * Avoid running pointers out of bounds; limit n to evens + * for simplicity. + */ + i = 4 + (n & 1); + insertionsort(list1 + (n - i) * size, i, size, cmp); + last = list1 + size * (n - i); + *EVAL(list2 + (last - list1)) = list2 + n * size; + +#ifdef NATURAL + p2 = list2; + f1 = list1; + sense = (cmp(f1, f1 + size) > 0); + for (; f1 < last; sense = !sense) { + length = 2; + /* Find pairs with same sense. */ + for (f2 = f1 + size2; f2 < last; f2 += size2) { + if ((cmp(f2, f2+ size) > 0) != sense) + break; + length += 2; + } + if (length < THRESHOLD) { /* Pairwise merge */ + do { + p2 = *EVAL(p2) = f1 + size2 - list1 + list2; + if (sense > 0) + swap (f1, f1 + size); + } while ((f1 += size2) < f2); + } else { /* Natural merge */ + l2 = f2; + for (f2 = f1 + size2; f2 < l2; f2 += size2) { + if ((cmp(f2-size, f2) > 0) != sense) { + p2 = *EVAL(p2) = f2 - list1 + list2; + if (sense > 0) + reverse(f1, f2-size); + f1 = f2; + } + } + if (sense > 0) + reverse (f1, f2-size); + f1 = f2; + if (f2 < last || cmp(f2 - size, f2) > 0) + p2 = *EVAL(p2) = f2 - list1 + list2; + else + p2 = *EVAL(p2) = list2 + n*size; + } + } +#else /* pairwise merge only. */ + for (f1 = list1, p2 = list2; f1 < last; f1 += size2) { + p2 = *EVAL(p2) = p2 + size2; + if (cmp (f1, f1 + size) > 0) + swap(f1, f1 + size); + } +#endif /* NATURAL */ +} + +/* + * This is to avoid out-of-bounds addresses in sorting the + * last 4 elements. + */ +static void +insertionsort(unsigned char *a, size_t n, size_t size, + int (*cmp)(const void *, const void *)) +{ + unsigned char *ai, *s, *t, *u, tmp; + int i; + + for (ai = a+size; --n >= 1; ai += size) + for (t = ai; t > a; t -= size) { + u = t - size; + if (cmp(u, t) <= 0) + break; + swap(u, t); + } +} diff --git a/src/util.h b/src/util.h index 116a1c83..0fa3f546 100644 --- a/src/util.h +++ b/src/util.h @@ -29,6 +29,11 @@ /* strdup() implementation with error checking (using xmalloc). */ char *xstrdup(const char *str); +#if !defined(HAVE_MERGESORT) +int mergesort(void *base, size_t nmemb, size_t size, + int (*compar)(const void *, const void *)); +#endif + #if !defined(HAVE_STRSEP) || defined(HAVE_GNU_C_LIBRARY) char *strsep(char **stringp, const char *delim); #endif diff --git a/util.h b/util.h index 116a1c83..0fa3f546 100644 --- a/util.h +++ b/util.h @@ -29,6 +29,11 @@ /* strdup() implementation with error checking (using xmalloc). */ char *xstrdup(const char *str); +#if !defined(HAVE_MERGESORT) +int mergesort(void *base, size_t nmemb, size_t size, + int (*compar)(const void *, const void *)); +#endif + #if !defined(HAVE_STRSEP) || defined(HAVE_GNU_C_LIBRARY) char *strsep(char **stringp, const char *delim); #endif