2 * Copyright (C) 2009 Gabor Kovesdan <gabor@FreeBSD.org>
3 * Copyright (C) 2012 Oleg Moskalenko <mom040267@gmail.com>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
33 #include <sys/types.h>
34 #include <sys/queue.h>
38 #if defined(SORT_THREADS)
41 #include <semaphore.h>
51 #include "radixsort.h"
53 unsigned long long free_memory = 1000000;
54 unsigned long long available_free_memory = 1000000;
58 const char *tmpdir = "/var/tmp";
59 const char *compress_program;
61 size_t max_open_files = 16;
64 * How much space we read from file at once
66 #define READ_CHUNK (4096)
69 * File reader structure
73 struct reader_buffer rb;
76 unsigned char *buffer;
77 unsigned char *mmapaddr;
78 unsigned char *mmapptr;
88 * Structure to be used in file merge process.
92 struct file_reader *fr;
93 struct sort_list_item *si; /* current top line */
98 * List elements of "cleanable" files list.
100 struct CLEANABLE_FILE
103 LIST_ENTRY(CLEANABLE_FILE) files;
107 * List header of "cleanable" files list.
109 static LIST_HEAD(CLEANABLE_FILES,CLEANABLE_FILE) tmp_files;
112 * Semaphore to protect the tmp file list.
113 * We use semaphore here because it is signal-safe, according to POSIX.
114 * And semaphore does not require pthread library.
116 static sem_t tmp_files_sem;
118 static void mt_sort(struct sort_list *list,
119 int (*sort_func)(void *, size_t, size_t,
120 int (*)(const void *, const void *)), const char* fn);
123 * Init tmp files list
129 LIST_INIT(&tmp_files);
130 sem_init(&tmp_files_sem, 0, 1);
134 * Save name of a tmp file for signal cleanup
137 tmp_file_atexit(const char *tmp_file)
141 sem_wait(&tmp_files_sem);
142 struct CLEANABLE_FILE *item =
143 sort_malloc(sizeof(struct CLEANABLE_FILE));
144 item->fn = sort_strdup(tmp_file);
145 LIST_INSERT_HEAD(&tmp_files, item, files);
146 sem_post(&tmp_files_sem);
154 clear_tmp_files(void)
156 struct CLEANABLE_FILE *item;
158 sem_wait(&tmp_files_sem);
159 LIST_FOREACH(item,&tmp_files,files) {
160 if ((item) && (item->fn))
163 sem_post(&tmp_files_sem);
167 * Check whether a file is a temporary file
170 file_is_tmp(const char* fn)
172 struct CLEANABLE_FILE *item;
176 sem_wait(&tmp_files_sem);
177 LIST_FOREACH(item,&tmp_files,files) {
178 if ((item) && (item->fn))
179 if (strcmp(item->fn, fn) == 0) {
184 sem_post(&tmp_files_sem);
191 * Generate new temporary file name
194 new_tmp_file_name(void)
196 static size_t tfcounter = 0;
197 static const char *fn = ".bsdsort.";
201 sz = strlen(tmpdir) + 1 + strlen(fn) + 32 + 1;
202 ret = sort_malloc(sz);
204 sprintf(ret, "%s/%s%d.%lu", tmpdir, fn, (int) getpid(), (unsigned long)(tfcounter++));
205 tmp_file_atexit(ret);
210 * Initialize file list
213 file_list_init(struct file_list *fl, bool tmp)
225 * Add a file name to the list
228 file_list_add(struct file_list *fl, char *fn, bool allocate)
232 if (fl->count >= fl->sz || (fl->fns == NULL)) {
233 fl->sz = (fl->sz) * 2 + 1;
234 fl->fns = sort_realloc(fl->fns, fl->sz *
237 fl->fns[fl->count] = allocate ? sort_strdup(fn) : fn;
243 * Populate file list from array of file names
246 file_list_populate(struct file_list *fl, int argc, char **argv, bool allocate)
252 for (i = 0; i < argc; i++)
253 file_list_add(fl, argv[i], allocate);
258 * Clean file list data and delete the files,
259 * if this is a list of temporary files
262 file_list_clean(struct file_list *fl)
269 for (i = 0; i < fl->count; i++) {
273 sort_free(fl->fns[i]);
290 sort_list_init(struct sort_list *l)
296 l->memsize = sizeof(struct sort_list);
302 * Add string to sort list
305 sort_list_add(struct sort_list *l, struct bwstring *str)
309 size_t indx = l->count;
311 if ((l->list == NULL) || (indx >= l->size)) {
312 size_t newsize = (l->size + 1) + 1024;
314 l->list = sort_realloc(l->list,
315 sizeof(struct sort_list_item*) * newsize);
316 l->memsize += (newsize - l->size) *
317 sizeof(struct sort_list_item*);
320 l->list[indx] = sort_list_item_alloc();
321 sort_list_item_set(l->list[indx], str);
322 l->memsize += sort_list_item_size(l->list[indx]);
328 * Clean sort list data
331 sort_list_clean(struct sort_list *l)
338 for (i = 0; i < l->count; i++) {
339 struct sort_list_item *item;
344 sort_list_item_clean(item);
354 l->memsize = sizeof(struct sort_list);
359 * Write sort list to file
362 sort_list_dump(struct sort_list *l, const char *fn)
368 f = openfile(fn, "w");
374 if (!(sort_opts_vals.uflag)) {
375 for (i = 0; i < l->count; ++i)
376 bwsfwrite(l->list[i]->str, f,
377 sort_opts_vals.zflag);
379 struct sort_list_item *last_printed_item = NULL;
380 struct sort_list_item *item;
381 for (i = 0; i < l->count; ++i) {
383 if ((last_printed_item == NULL) ||
384 list_coll(&last_printed_item, &item)) {
385 bwsfwrite(item->str, f, sort_opts_vals.zflag);
386 last_printed_item = item;
397 * Checks if the given file is sorted. Stops at the first disorder,
398 * prints the disordered line and returns 1.
401 check(const char *fn)
403 struct bwstring *s1, *s2, *s1disorder, *s2disorder;
404 struct file_reader *fr;
405 struct keys_array *ka1, *ka2;
407 size_t pos, posdisorder;
409 s1 = s2 = s1disorder = s2disorder = NULL;
412 fr = file_reader_init(fn);
423 s1 = file_reader_readline(fr);
427 ka1 = keys_array_alloc();
430 s2 = file_reader_readline(fr);
434 ka2 = keys_array_alloc();
440 bwsprintf(stdout, s2, "s1=<", ">");
441 bwsprintf(stdout, s1, "s2=<", ">");
443 int cmp = key_coll(ka2, ka1, 0);
445 printf("; cmp1=%d", cmp);
447 if (!cmp && sort_opts_vals.complex_sort &&
448 !(sort_opts_vals.uflag) && !(sort_opts_vals.sflag)) {
449 cmp = top_level_str_coll(s2, s1);
451 printf("; cmp2=%d", cmp);
456 if ((sort_opts_vals.uflag && (cmp <= 0)) || (cmp < 0)) {
457 if (!(sort_opts_vals.csilentflag)) {
458 s2disorder = bwsdup(s2);
461 s1disorder = bwsdup(s1);
469 clean_keys_array(s1, ka1);
477 s2 = file_reader_readline(fr);
481 ka2 = keys_array_alloc();
487 clean_keys_array(s1, ka1);
495 clean_keys_array(s2, ka2);
502 if ((fn == NULL) || (*fn == 0) || (strcmp(fn, "-") == 0)) {
504 s2 = file_reader_readline(fr);
511 file_reader_free(fr);
514 bws_disorder_warnx(s2disorder, fn, posdisorder);
516 bws_disorder_warnx(s1disorder, fn, posdisorder);
517 if (s1disorder != s2disorder)
532 * Opens a file. If the given filename is "-", stdout will be
536 openfile(const char *fn, const char *mode)
540 if (strcmp(fn, "-") == 0) {
541 return ((mode && mode[0] == 'r') ? stdin : stdout);
543 mode_t orig_file_mask = 0;
544 int is_tmp = file_is_tmp(fn);
546 if (is_tmp && (mode[0] == 'w'))
547 orig_file_mask = umask(S_IWGRP | S_IWOTH |
550 if (is_tmp && (compress_program != NULL)) {
554 cmdsz = strlen(fn) + 128;
555 cmd = sort_malloc(cmdsz);
560 snprintf(cmd, cmdsz - 1, "cat %s | %s -d",
561 fn, compress_program);
562 else if (mode[0] == 'w')
563 snprintf(cmd, cmdsz - 1, "%s > %s",
564 compress_program, fn);
566 err(2, "%s", getstr(7));
568 if ((file = popen(cmd, mode)) == NULL)
574 if ((file = fopen(fn, mode)) == NULL)
577 if (is_tmp && (mode[0] == 'w'))
578 umask(orig_file_mask);
588 closefile(FILE *f, const char *fn)
592 } else if (f == stdin) {
594 } else if (f == stdout) {
597 if (file_is_tmp(fn) && compress_program != NULL) {
606 * Reads a file into the internal buffer.
609 file_reader_init(const char *fsrc)
611 struct file_reader *ret;
616 ret = sort_malloc(sizeof(struct file_reader));
617 memset(ret, 0, sizeof(struct file_reader));
620 if (sort_opts_vals.zflag)
623 ret->fname = sort_strdup(fsrc);
625 if (strcmp(fsrc, "-") && (compress_program == NULL) && use_mmap) {
628 struct stat stat_buf;
633 flags = MAP_NOCORE | MAP_NOSYNC;
635 fd = open(fsrc, O_RDONLY);
639 if (fstat(fd, &stat_buf) < 0) {
644 sz = stat_buf.st_size;
646 #if defined(MAP_PREFAULT_READ)
647 flags |= MAP_PREFAULT_READ;
650 addr = mmap(NULL, sz, PROT_READ, flags, fd, 0);
651 if (addr == MAP_FAILED) {
657 ret->mmapaddr = addr;
659 ret->mmapptr = ret->mmapaddr;
664 if (ret->mmapaddr == NULL) {
665 ret->file = openfile(fsrc, "r");
666 if (ret->file == NULL)
669 if (strcmp(fsrc, "-")) {
670 ret->cbsz = READ_CHUNK;
671 ret->buffer = sort_malloc(ret->cbsz);
675 ret->bsz = fread(ret->buffer, 1, ret->cbsz, ret->file);
677 if (ferror(ret->file))
687 file_reader_readline(struct file_reader *fr)
689 struct bwstring *ret = NULL;
692 unsigned char *mmapend;
694 mmapend = fr->mmapaddr + fr->mmapsize;
695 if (fr->mmapptr >= mmapend)
698 unsigned char *strend;
701 sz = mmapend - fr->mmapptr;
702 strend = memchr(fr->mmapptr, fr->elsymb, sz);
704 if (strend == NULL) {
705 ret = bwscsbdup(fr->mmapptr, sz);
706 fr->mmapptr = mmapend;
708 ret = bwscsbdup(fr->mmapptr, strend -
710 fr->mmapptr = strend + 1;
714 } else if (fr->file != stdin) {
715 unsigned char *strend;
716 size_t bsz1, remsz, search_start;
722 if (fr->bsz > fr->strbeg)
723 remsz = fr->bsz - fr->strbeg;
725 /* line read cycle */
727 if (remsz > search_start)
728 strend = memchr(fr->buffer + fr->strbeg +
729 search_start, fr->elsymb, remsz -
739 if (fr->bsz != fr->cbsz)
741 err(2, "File read software error 1");
743 if (remsz > (READ_CHUNK >> 1)) {
744 search_start = fr->cbsz - fr->strbeg;
745 fr->cbsz += READ_CHUNK;
746 fr->buffer = sort_realloc(fr->buffer,
748 bsz1 = fread(fr->buffer + fr->bsz, 1,
749 READ_CHUNK, fr->file);
751 if (ferror(fr->file))
758 if (remsz > 0 && fr->strbeg>0)
759 bcopy(fr->buffer + fr->strbeg,
763 search_start = remsz;
764 bsz1 = fread(fr->buffer + remsz, 1,
765 fr->cbsz - remsz, fr->file);
767 if (ferror(fr->file))
771 fr->bsz = remsz + bsz1;
777 strend = fr->buffer + fr->bsz;
779 if ((fr->buffer + fr->strbeg <= strend) &&
780 (fr->strbeg < fr->bsz) && (remsz>0))
781 ret = bwscsbdup(fr->buffer + fr->strbeg, strend -
782 fr->buffer - fr->strbeg);
784 fr->strbeg = (strend - fr->buffer) + 1;
789 ret = bwsfgetln(fr->file, &len, sort_opts_vals.zflag,
797 file_reader_clean(struct file_reader *fr)
802 munmap(fr->mmapaddr, fr->mmapsize);
808 sort_free(fr->buffer);
811 if (fr->file != stdin)
812 closefile(fr->file, fr->fname);
815 sort_free(fr->fname);
817 memset(fr, 0, sizeof(struct file_reader));
822 file_reader_free(struct file_reader *fr)
826 file_reader_clean(fr);
832 procfile(const char *fsrc, struct sort_list *list, struct file_list *fl)
834 struct file_reader *fr;
836 fr = file_reader_init(fsrc);
840 /* file browse cycle */
842 struct bwstring *bws;
844 bws = file_reader_readline(fr);
849 sort_list_add(list, bws);
851 if (list->memsize >= available_free_memory) {
854 fn = new_tmp_file_name();
855 sort_list_to_file(list, fn);
856 file_list_add(fl, fn, false);
857 sort_list_clean(list);
861 file_reader_free(fr);
867 * Compare file headers. Files with EOF always go to the end of the list.
870 file_header_cmp(struct file_header *f1, struct file_header *f2)
876 if (f1->fr == NULL) {
877 return ((f2->fr == NULL) ? 0 : +1);
878 } else if (f2->fr == NULL)
883 ret = list_coll(&(f1->si), &(f2->si));
885 return ((f1->file_pos < f2->file_pos) ? -1 : +1);
892 * Allocate and init file header structure
895 file_header_init(struct file_header **fh, const char *fn, size_t file_pos)
899 struct bwstring *line;
901 *fh = sort_malloc(sizeof(struct file_header));
902 (*fh)->file_pos = file_pos;
903 (*fh)->fr = file_reader_init(fn);
904 if ((*fh)->fr == NULL) {
906 err(2, "%s", getstr(8));
908 line = file_reader_readline((*fh)->fr);
910 file_reader_free((*fh)->fr);
914 (*fh)->si = sort_list_item_alloc();
915 sort_list_item_set((*fh)->si, line);
924 file_header_close(struct file_header **fh)
929 file_reader_free((*fh)->fr);
933 sort_list_item_clean((*fh)->si);
934 sort_free((*fh)->si);
943 * Swap two array elements
946 file_header_swap(struct file_header **fh, size_t i1, size_t i2)
948 struct file_header *tmp;
955 /* heap algorithm ==>> */
958 * See heap sort algorithm
959 * "Raises" last element to its right place
962 file_header_heap_swim(struct file_header **fh, size_t indx)
968 parent_index = (indx - 1) >> 1;
970 if (file_header_cmp(fh[indx], fh[parent_index]) < 0) {
971 /* swap child and parent and continue */
972 file_header_swap(fh, indx, parent_index);
973 file_header_heap_swim(fh, parent_index);
979 * Sink the top element to its correct position
982 file_header_heap_sink(struct file_header **fh, size_t indx, size_t size)
984 size_t left_child_index;
985 size_t right_child_index;
987 left_child_index = indx + indx + 1;
988 right_child_index = left_child_index + 1;
990 if (left_child_index < size) {
991 size_t min_child_index;
993 min_child_index = left_child_index;
995 if ((right_child_index < size) &&
996 (file_header_cmp(fh[left_child_index],
997 fh[right_child_index]) > 0))
998 min_child_index = right_child_index;
999 if (file_header_cmp(fh[indx], fh[min_child_index]) > 0) {
1000 file_header_swap(fh, indx, min_child_index);
1001 file_header_heap_sink(fh, min_child_index, size);
1006 /* <<== heap algorithm */
1009 * Adds element to the "left" end
1012 file_header_list_rearrange_from_header(struct file_header **fh, size_t size)
1015 file_header_heap_sink(fh, 0, size);
1019 * Adds element to the "right" end
1022 file_header_list_push(struct file_header *f, struct file_header **fh, size_t size)
1026 file_header_heap_swim(fh, size - 1);
1031 struct bwstring *str;
1035 * Prints the current line of the file
1038 file_header_print(struct file_header *fh, FILE *f_out, struct last_printed *lp)
1041 if (fh && fh->fr && f_out && fh->si && fh->si->str) {
1042 if (sort_opts_vals.uflag) {
1043 if ((lp->str == NULL) || (str_list_coll(lp->str, &(fh->si)))) {
1044 bwsfwrite(fh->si->str, f_out, sort_opts_vals.zflag);
1047 lp->str = bwsdup(fh->si->str);
1050 bwsfwrite(fh->si->str, f_out, sort_opts_vals.zflag);
1058 file_header_read_next(struct file_header *fh)
1062 struct bwstring *tmp;
1064 tmp = file_reader_readline(fh->fr);
1066 file_reader_free(fh->fr);
1069 sort_list_item_clean(fh->si);
1075 fh->si = sort_list_item_alloc();
1076 sort_list_item_set(fh->si, tmp);
1082 * Merge array of "files headers"
1085 file_headers_merge(size_t fnum, struct file_header **fh, FILE *f_out)
1087 struct last_printed lp;
1090 memset(&lp, 0, sizeof(lp));
1093 * construct the initial sort structure
1095 for (i = 0; i < fnum; i++)
1096 file_header_list_push(fh[i], fh, i);
1098 while (fh[0]->fr) { /* unfinished files are always in front */
1099 /* output the smallest line: */
1100 file_header_print(fh[0], f_out, &lp);
1101 /* read a new line, if possible: */
1102 file_header_read_next(fh[0]);
1103 /* re-arrange the list: */
1104 file_header_list_rearrange_from_header(fh, fnum);
1112 * Merges the given files into the output file, which can be
1116 merge_files_array(size_t argc, char **argv, const char *fn_out)
1119 if (argv && fn_out) {
1120 struct file_header **fh;
1124 f_out = openfile(fn_out, "w");
1129 fh = sort_malloc((argc + 1) * sizeof(struct file_header *));
1131 for (i = 0; i < argc; i++)
1132 file_header_init(fh + i, argv[i], (size_t) i);
1134 file_headers_merge(argc, fh, f_out);
1136 for (i = 0; i < argc; i++)
1137 file_header_close(fh + i);
1141 closefile(f_out, fn_out);
1146 * Shrinks the file list until its size smaller than max number of opened files
1149 shrink_file_list(struct file_list *fl)
1152 if ((fl == NULL) || (size_t) (fl->count) < max_open_files)
1155 struct file_list new_fl;
1158 file_list_init(&new_fl, true);
1159 while (indx < fl->count) {
1163 num = fl->count - indx;
1164 fnew = new_tmp_file_name();
1166 if ((size_t) num >= max_open_files)
1167 num = max_open_files - 1;
1168 merge_files_array(num, fl->fns + indx, fnew);
1172 for (i = 0; i < num; i++)
1173 unlink(fl->fns[indx + i]);
1175 file_list_add(&new_fl, fnew, false);
1178 fl->tmp = false; /* already taken care of */
1179 file_list_clean(fl);
1181 fl->count = new_fl.count;
1182 fl->fns = new_fl.fns;
1184 fl->tmp = new_fl.tmp;
1191 * Merge list of files
1194 merge_files(struct file_list *fl, const char *fn_out)
1198 while (shrink_file_list(fl));
1200 merge_files_array(fl->count, fl->fns, fn_out);
1205 get_sort_method_name(int sm)
1208 if (sm == SORT_MERGESORT)
1210 else if (sort_opts_vals.sort_method == SORT_RADIXSORT)
1212 else if (sort_opts_vals.sort_method == SORT_HEAPSORT)
1221 static int sort_qsort(void *list, size_t count, size_t elem_size,
1222 int (*cmp_func)(const void *, const void *))
1225 qsort(list, count, elem_size, cmp_func);
1230 * Sort list of lines and writes it to the file
1233 sort_list_to_file(struct sort_list *list, const char *outfile)
1235 struct sort_mods *sm = &(keys[0].sm);
1237 if (!(sm->Mflag) && !(sm->Rflag) && !(sm->Vflag) && !(sm->Vflag) &&
1238 !(sm->gflag) && !(sm->hflag) && !(sm->nflag)) {
1239 if ((sort_opts_vals.sort_method == SORT_DEFAULT) && byte_sort)
1240 sort_opts_vals.sort_method = SORT_RADIXSORT;
1242 } else if (sort_opts_vals.sort_method == SORT_RADIXSORT)
1243 err(2, "%s", getstr(9));
1246 * to handle stable sort and the unique cases in the
1247 * right order, we need stable basic algorithm
1249 if (sort_opts_vals.sflag) {
1250 switch (sort_opts_vals.sort_method){
1251 case SORT_MERGESORT:
1253 case SORT_RADIXSORT:
1256 sort_opts_vals.sort_method = SORT_MERGESORT;
1259 errx(2, "%s", getstr(10));
1263 if (sort_opts_vals.sort_method == SORT_DEFAULT)
1264 sort_opts_vals.sort_method = DEFAULT_SORT_ALGORITHM;
1267 printf("sort_method=%s\n",
1268 get_sort_method_name(sort_opts_vals.sort_method));
1270 switch (sort_opts_vals.sort_method){
1271 case SORT_RADIXSORT:
1272 rxsort(list->list, list->count);
1273 sort_list_dump(list, outfile);
1275 case SORT_MERGESORT:
1276 mt_sort(list, mergesort, outfile);
1279 mt_sort(list, heapsort, outfile);
1282 mt_sort(list, sort_qsort, outfile);
1285 mt_sort(list, DEFAULT_SORT_FUNC, outfile);
1290 /******************* MT SORT ************************/
1292 #if defined(SORT_THREADS)
1293 /* semaphore to count threads */
1296 /* current system sort function */
1297 static int (*g_sort_func)(void *, size_t, size_t,
1298 int(*)(const void *, const void *));
1301 * Sort cycle thread (in multi-threaded mode)
1304 mt_sort_thread(void* arg)
1306 struct sort_list *list = arg;
1308 g_sort_func(list->list, list->count, sizeof(struct sort_list_item *),
1309 (int(*)(const void *, const void *)) list_coll);
1317 * Compare sub-lists. Empty sub-lists always go to the end of the list.
1320 sub_list_cmp(struct sort_list *l1, struct sort_list *l2)
1326 if (l1->count == 0) {
1327 return ((l2->count == 0) ? 0 : +1);
1328 } else if (l2->count == 0) {
1333 ret = list_coll(&(l1->list[0]), &(l2->list[0]));
1335 return ((l1->sub_list_pos < l2->sub_list_pos) ?
1343 * Swap two array elements
1346 sub_list_swap(struct sort_list **sl, size_t i1, size_t i2)
1348 struct sort_list *tmp;
1355 /* heap algorithm ==>> */
1358 * See heap sort algorithm
1359 * "Raises" last element to its right place
1362 sub_list_swim(struct sort_list **sl, size_t indx)
1366 size_t parent_index;
1368 parent_index = (indx - 1) >> 1;
1370 if (sub_list_cmp(sl[indx], sl[parent_index]) < 0) {
1371 /* swap child and parent and continue */
1372 sub_list_swap(sl, indx, parent_index);
1373 sub_list_swim(sl, parent_index);
1379 * Sink the top element to its correct position
1382 sub_list_sink(struct sort_list **sl, size_t indx, size_t size)
1384 size_t left_child_index;
1385 size_t right_child_index;
1387 left_child_index = indx + indx + 1;
1388 right_child_index = left_child_index + 1;
1390 if (left_child_index < size) {
1391 size_t min_child_index;
1393 min_child_index = left_child_index;
1395 if ((right_child_index < size) &&
1396 (sub_list_cmp(sl[left_child_index],
1397 sl[right_child_index]) > 0))
1398 min_child_index = right_child_index;
1399 if (sub_list_cmp(sl[indx], sl[min_child_index]) > 0) {
1400 sub_list_swap(sl, indx, min_child_index);
1401 sub_list_sink(sl, min_child_index, size);
1406 /* <<== heap algorithm */
1409 * Adds element to the "right" end
1412 sub_list_push(struct sort_list *s, struct sort_list **sl, size_t size)
1416 sub_list_swim(sl, size - 1);
1419 struct last_printed_item
1421 struct sort_list_item *item;
1425 * Prints the current line of the file
1428 sub_list_header_print(struct sort_list *sl, FILE *f_out,
1429 struct last_printed_item *lp)
1432 if (sl && sl->count && f_out && sl->list[0]->str) {
1433 if (sort_opts_vals.uflag) {
1434 if ((lp->item == NULL) || (list_coll(&(lp->item),
1436 bwsfwrite(sl->list[0]->str, f_out,
1437 sort_opts_vals.zflag);
1438 lp->item = sl->list[0];
1441 bwsfwrite(sl->list[0]->str, f_out,
1442 sort_opts_vals.zflag);
1450 sub_list_next(struct sort_list *sl)
1453 if (sl && sl->count) {
1460 * Merge sub-lists to a file
1463 merge_sub_lists(struct sort_list **sl, size_t n, FILE* f_out)
1465 struct last_printed_item lp;
1468 memset(&lp,0,sizeof(lp));
1470 /* construct the initial list: */
1471 for (i = 0; i < n; i++)
1472 sub_list_push(sl[i], sl, i);
1474 while (sl[0]->count) { /* unfinished lists are always in front */
1475 /* output the smallest line: */
1476 sub_list_header_print(sl[0], f_out, &lp);
1477 /* move to a new line, if possible: */
1478 sub_list_next(sl[0]);
1479 /* re-arrange the list: */
1480 sub_list_sink(sl, 0, n);
1485 * Merge sub-lists to a file
1488 merge_list_parts(struct sort_list **parts, size_t n, const char *fn)
1492 f_out = openfile(fn,"w");
1494 merge_sub_lists(parts, n, f_out);
1496 closefile(f_out, fn);
1499 #endif /* defined(SORT_THREADS) */
1501 * Multi-threaded sort algorithm "driver"
1504 mt_sort(struct sort_list *list,
1505 int(*sort_func)(void *, size_t, size_t, int(*)(const void *, const void *)),
1508 #if defined(SORT_THREADS)
1509 if (nthreads < 2 || list->count < MT_SORT_THRESHOLD) {
1510 size_t nthreads_save = nthreads;
1513 /* if single thread or small data, do simple sort */
1514 sort_func(list->list, list->count,
1515 sizeof(struct sort_list_item *),
1516 (int(*)(const void *, const void *)) list_coll);
1517 sort_list_dump(list, fn);
1518 #if defined(SORT_THREADS)
1519 nthreads = nthreads_save;
1521 /* multi-threaded sort */
1522 struct sort_list **parts;
1523 size_t avgsize, cstart, i;
1525 /* array of sub-lists */
1526 parts = sort_malloc(sizeof(struct sort_list*) * nthreads);
1528 avgsize = list->count / nthreads;
1530 /* set global system sort function */
1531 g_sort_func = sort_func;
1534 for (i = 0; i < nthreads; ++i) {
1537 parts[i] = sort_malloc(sizeof(struct sort_list));
1538 parts[i]->list = list->list + cstart;
1539 parts[i]->memsize = 0;
1540 parts[i]->sub_list_pos = i;
1542 sz = (i == nthreads - 1) ? list->count - cstart :
1545 parts[i]->count = sz;
1547 parts[i]->size = parts[i]->count;
1552 /* init threads counting semaphore */
1553 sem_init(&mtsem, 0, 0);
1556 for (i = 0; i < nthreads; ++i) {
1558 pthread_attr_t attr;
1560 pthread_attr_init(&attr);
1561 pthread_attr_setdetachstate(&attr, PTHREAD_DETACHED);
1564 int res = pthread_create(&pth, &attr,
1565 mt_sort_thread, parts[i]);
1569 if (errno == EAGAIN) {
1576 pthread_attr_destroy(&attr);
1579 /* wait for threads completion */
1580 for (i = 0; i < nthreads; ++i) {
1583 /* destroy the semaphore - we do not need it anymore */
1584 sem_destroy(&mtsem);
1586 /* merge sorted sub-lists to the file */
1587 merge_list_parts(parts, nthreads, fn);
1589 /* free sub-lists data */
1590 for (i = 0; i < nthreads; ++i) {
1591 sort_free(parts[i]);
1595 #endif /* defined(SORT_THREADS) */