1 /* @(#)compress.c 8.1 (Berkeley) 6/6/93 */
2 /* $NetBSD: compress.c,v 1.7 2009/04/14 08:50:06 lukem Exp $ */
5 * Copyright (c) 1989, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * Edward Wang at The University of California, Berkeley.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 /* tunable parameters */
53 int cc_token_max = 8; /* <= TOKEN_MAX */
54 int cc_token_min = 2; /* > tt.tt_put_token_cost */
58 int cc_bufsize = 1024 * 3; /* XXX, or 80 * 24 * 2 */
70 char string[TOKEN_MAX];
76 long time; /* time last seen */
77 short bcount; /* count in this buffer */
78 short ccount; /* count in compression */
79 short places; /* places in the buffer */
80 short code; /* token code */
81 struct cc *qforw, *qback;
82 struct cc *hforw, **hback;
85 short cc_thresholds[TOKEN_MAX + 1];
86 #define thresh(length) (cc_thresholds[length])
87 #define threshp(code, count, length) \
88 ((code) >= 0 || (short) (count) >= cc_thresholds[length])
91 short cc_wthresholds[TOKEN_MAX + 1];
92 #define wthresh(length) (cc_wthresholds[length])
93 #define wthreshp(weight, length) ((short) (weight) >= cc_wthresholds[length])
95 #define wthreshp(weight, length) (0)
99 short cc_wlimits[TOKEN_MAX + 1];
100 #define wlimit(length) (cc_wlimits[length])
103 #define put_token_score(length) ((length) - tt.tt_put_token_cost)
105 int cc_score_adjustments[TOKEN_MAX + 1][8]; /* XXX, 8 > max of cc_thresholds */
106 #define score_adjust(score, p) \
108 int _length = (p)->length; \
109 int _ccount = (p)->ccount; \
110 if (threshp((p)->code, _ccount, _length) || \
111 wthreshp((p)->weight, _length)) /* XXX */ \
112 (score) -= _length - tt.tt_put_token_cost; \
114 (score) += cc_score_adjustments[_length][_ccount]; \
117 int cc_initial_scores[TOKEN_MAX + 1][8]; /* XXX, 8 > max of cc_thresholds */
119 struct cc cc_q0a, cc_q0b, cc_q1a, cc_q1b;
121 #define qinsert(p1, p2) \
123 struct cc *forw = (p1)->qforw; \
124 struct cc *back = (p1)->qback; \
125 back->qforw = forw; \
126 forw->qback = back; \
127 forw = (p2)->qforw; \
128 (p1)->qforw = forw; \
129 forw->qback = (p1); \
130 (p2)->qforw = (p1); \
131 (p1)->qback = (p2); \
134 #define qinsertq(q, p) \
135 ((q)->qforw == (q) ? 0 : \
136 ((q)->qback->qforw = (p)->qforw, \
137 (p)->qforw->qback = (q)->qback, \
138 (q)->qforw->qback = (p), \
139 (p)->qforw = (q)->qforw, \
144 #define HSIZE (1 << H)
145 #define hash(h, c) (((((h) >> (H - 8)) | (h) << 8) ^ (c)) & (HSIZE - 1))
148 struct cc **cc_output; /* the output array */
149 short *cc_places[TOKEN_MAX + 1];
150 short *cc_hashcodes; /* for computing hashcodes */
151 struct cc **cc_htab; /* the hash table */
152 struct cc **cc_tokens; /* holds all the active tokens */
158 long cc_time, cc_time0;
160 char *cc_tt_ob, *cc_tt_obe;
162 int cc_compress(struct cc **, struct cc **, char);
163 void cc_compress_cleanup(struct cc **, int);
164 void cc_compress_phase(struct cc **, int, struct cc **, int);
165 void cc_compress_phase1(struct cc **, struct cc **, int, int);
166 void cc_output_phase(char *, struct cc **, int);
167 int cc_sweep(char *, int, struct cc **, int);
168 void cc_sweep0(char *, int, int);
169 int cc_sweep_phase(char *, int, struct cc **);
170 void cc_sweep_reverse(struct cc **, short *);
171 int cc_token_compare(const void *, const void *);
179 if (tt.tt_token_max > cc_token_max)
180 tt.tt_token_max = cc_token_max;
181 if (tt.tt_token_min < cc_token_min)
182 tt.tt_token_min = cc_token_min;
183 if (tt.tt_token_min > tt.tt_token_max) {
187 if (tt.tt_ntoken > cc_ntoken / 2) /* not likely */
188 tt.tt_ntoken = cc_ntoken / 2;
189 #define C(x) (sizeof (x) / sizeof *(x))
190 for (i = 0; i < (int)C(cc_thresholds); i++) {
191 int h = i - tt.tt_put_token_cost;
194 (tt.tt_set_token_cost + 1 + h - 1) / h + 1;
196 cc_thresholds[i] = 0;
198 for (i = 0; i < (int)C(cc_score_adjustments); i++) {
199 int t = cc_thresholds[i];
200 for (j = 0; j < (int)C(*cc_score_adjustments); j++) {
202 cc_score_adjustments[i][j] =
203 - (i - tt.tt_put_token_cost);
205 cc_score_adjustments[i][j] = 0;
209 * length * (ccount + 1) a
211 * set-token-cost + length +
212 * ccount * put-token-cost b
213 * the score adjustment is (b - a)
215 cc_score_adjustments[i][j] =
216 tt.tt_set_token_cost + i +
217 j * tt.tt_put_token_cost -
220 cc_initial_scores[i][j] = 0;
223 * - (set-token-cost +
224 * (length - put-token-cost) -
225 * (length - put-token-cost) * ccount)
227 cc_initial_scores[i][j] =
228 - (tt.tt_set_token_cost +
229 (i - tt.tt_put_token_cost) -
230 (i - tt.tt_put_token_cost) * j);
234 for (i = 1; i < C(cc_wthresholds); i++) {
236 ((tt.tt_set_token_cost + tt.tt_put_token_cost) / i +
239 cc_wlimits[i] = cc_wthresholds[i] + cc_weight;
243 if ((cc_output = (struct cc **)
244 malloc((unsigned) cc_bufsize * sizeof *cc_output)) == NULL)
246 if ((cc_hashcodes = (short *)
247 malloc((unsigned) cc_bufsize * sizeof *cc_hashcodes)) == NULL)
249 if ((cc_htab = (struct cc **) malloc(HSIZE * sizeof *cc_htab)) == NULL)
251 if ((cc_tokens = (struct cc **)
253 (cc_ntoken + tt.tt_token_max - tt.tt_token_min + 1) *
254 sizeof *cc_tokens)) == NULL)
256 if ((cc_undo = (struct cc_undo *)
257 malloc((unsigned) cc_bufsize * sizeof *cc_undo)) == NULL)
259 for (i = tt.tt_token_min; i <= tt.tt_token_max; i++)
260 if ((cc_places[i] = (short *)
261 malloc((unsigned) cc_bufsize * sizeof **cc_places)) == NULL)
263 cc_q0a.qforw = cc_q0a.qback = &cc_q0a;
264 cc_q0b.qforw = cc_q0b.qback = &cc_q0b;
265 cc_q1a.qforw = cc_q1a.qback = &cc_q1a;
266 cc_q1b.qforw = cc_q1b.qback = &cc_q1b;
267 if ((p = (struct cc *) malloc((unsigned) cc_ntoken * sizeof *p)) == NULL)
269 for (i = 0; i < tt.tt_ntoken; i++) {
272 p->qback = cc_q0a.qback;
278 for (; i < cc_ntoken; i++) {
281 p->qback = cc_q1a.qback;
289 if ((cc_buffer = malloc((unsigned) cc_bufsize)) == NULL)
301 tt_obp = tt_ob = cc_buffer;
302 tt_obe = tt_ob + cc_bufsize;
303 tt.tt_flush = ccflush;
305 cc_trace_fp = fopen("window-trace", "a");
306 (void) fcntl(fileno(cc_trace_fp), F_SETFD, 1);
316 memset((char *) cc_htab, 0, HSIZE * sizeof *cc_htab);
317 for (p = cc_q0a.qforw; p != &cc_q0a; p = p->qforw)
319 for (p = cc_q1a.qforw; p != &cc_q1a; p = p->qforw)
328 tt_obp = tt_ob = cc_tt_ob;
331 if (cc_trace_fp != NULL) {
332 (void) fclose(cc_trace_fp);
340 int bufsize = tt_obp - tt_ob;
343 if (tt_ob != cc_buffer)
345 if (cc_trace_fp != NULL) {
346 (void) fwrite(tt_ob, 1, bufsize, cc_trace_fp);
347 (void) putc(-1, cc_trace_fp);
350 (*tt.tt_compress)(1);
351 if (bufsize < tt.tt_token_min) {
355 tt_obp = tt_ob = cc_tt_ob;
359 n = cc_sweep_phase(cc_buffer, bufsize, cc_tokens);
360 cc_compress_phase(cc_output, bufsize, cc_tokens, n);
361 cc_output_phase(cc_buffer, cc_output, bufsize);
363 tt_obp = tt_ob = cc_buffer;
364 tt_obe = cc_buffer + cc_bufsize;
366 (*tt.tt_compress)(0);
367 tt.tt_flush = ccflush;
371 cc_sweep_phase(char *buffer, int bufsize, struct cc **tokens)
373 struct cc **pp = tokens;
385 cc_sweep0(buffer, bufsize, tt.tt_token_min - 1);
391 for (i = tt.tt_token_min; i <= tt.tt_token_max; i++) {
400 (void) fflush(stdout);
403 n = cc_sweep(buffer, bufsize, pp, i);
415 qinsertq(&cc_q1b, &cc_q1a);
418 printf("\n %d tokens, %d candidates\n",
427 cc_sweep0(char *buffer, int n, int length)
433 short pc = tt.tt_padc;
435 /* n and length are at least 1 */
440 if ((*hc++ = *p++) == pc)
446 for (i = n--; --i;) {
447 if ((c = *p++) == pc || *hc < 0)
457 cc_sweep(char *buffer, int bufsize, struct cc **tokens, int length)
463 short *places = cc_places[length];
464 struct cc **pp = tokens;
465 short threshold = thresh(length);
467 short wthreshold = wthresh(length);
468 short limit = wlimit(length);
471 short pc = tt.tt_padc;
478 for (i = 0; i < bufsize; i++, time0++) {
485 if ((hh = *hc1) < 0 || c == pc) {
490 h = cc_htab + (*hc1++ = hash(hh, c));
493 for (p = *h; p != NULL; p = p->hforw)
494 if (p->length == (char) length) {
495 char *p1 = p->string;
496 char *p2 = cp - length;
508 (p->time >= cc_time0 && p->length == (char) length))
510 if (p->hback != NULL)
511 if ((*p->hback = p->hforw) != NULL)
512 p->hforw->hback = p->hback;
514 char *p1 = p->string;
515 char *p2 = cp - length;
523 p->weight = cc_weight;
529 if ((p->hforw = *h) != NULL)
530 p->hforw->hback = &p->hforw;
539 } else if (p->time < cc_time0) {
541 if ((p->weight += p->time - time0) < 0)
542 p->weight = cc_weight;
543 else if ((p->weight += cc_weight) > limit)
554 if (p->weight >= wthreshold) {
569 } else if (p->time + length > time0) {
571 * overlapping token, don't count as two and
572 * don't update time0, but do adjust weight to offset
576 if (cc_weight != 0) { /* XXX */
577 p->weight += time0 - p->time;
578 if (!p->flag && p->weight >= wthreshold) {
585 places[i] = p->places;
589 if ((p->weight += p->time - time0) < 0)
590 p->weight = cc_weight;
591 else if ((p->weight += cc_weight) > limit)
597 /* code must be < 0 if flag false here */
598 (p->bcount >= threshold
600 || p->weight >= wthreshold
607 places[i] = p->places;
611 if ((i = pp - tokens) > 0) {
614 cc_sweep_reverse(tokens, places);
615 if (cc_sort && i > 1) {
616 qsort((char *) tokens, i, sizeof *tokens,
620 if ((i = i * cc_chop / 100) == 0)
630 cc_sweep_reverse(struct cc **pp, short int *places)
635 while ((p = *pp++) != NULL) {
638 /* the list is never empty */
643 } while ((t = frnt) >= 0);
649 cc_compress_phase(struct cc **output, int bufsize, struct cc **tokens, int ntoken)
653 memset((char *) output, 0, bufsize * sizeof *output);
654 for (i = 0; i < cc_npass0; i++)
655 cc_compress_phase1(output, tokens, ntoken, 0);
656 for (i = 0; i < cc_npass1; i++)
657 cc_compress_phase1(output, tokens, ntoken, 1);
658 cc_compress_cleanup(output, bufsize);
662 cc_compress_phase1(struct cc **output, struct cc **tokens, int ntoken, int flag)
667 int nt = 0, cc = 0, nc = 0;
677 while (pp < tokens + ntoken) {
688 printf(" (%d", (*pp)->length);
689 (void) fflush(stdout);
692 pp += cc_compress(output, pp, flag);
695 printf(" %dt %du %dc)", ntoken_stat, ccount_stat,
705 printf("\n total: (%dt %du %dc)\n", nt, cc, nc);
712 cc_compress_cleanup(struct cc **output, int bufsize)
716 /* the previous output phase may have been interrupted */
717 qinsertq(&cc_q0b, &cc_q0a);
718 for (end = output + bufsize; output < end;) {
721 if ((p = *output) == NULL) {
727 } else if (p->code >= 0) {
730 } else if (p->ccount == 0) {
732 } else if (p->ccount >= thresh(length)
734 || wthreshp(p->weight, length)
747 cc_compress(struct cc **output, struct cc **tokens, char flag)
749 struct cc **pp = tokens;
750 struct cc *p = *pp++;
751 int length = p->length;
752 int threshold = thresh(length);
754 short wthreshold = wthresh(length);
756 short *places = cc_places[length];
757 int *initial_scores = cc_initial_scores[length];
758 int initial_score0 = put_token_score(length);
762 struct cc_undo *undop;
770 if ((short) ccount >= p->bcount)
772 if (p->code >= 0 || ccount >= threshold)
775 else if (p->weight >= wthreshold)
776 /* allow one fewer match than normal */
777 /* XXX, should adjust for ccount */
778 score = - tt.tt_set_token_cost;
781 score = initial_scores[ccount];
786 for (i = p->places; i >= 0; i = places[i]) {
789 struct cc **ip = output + i;
790 int score0 = initial_score0;
791 struct cc **iip = ip + length;
792 struct cc_undo *undop1 = undop;
794 if ((x = *(jp = ip)) != NULL)
796 while (--jp >= output)
797 if ((x = *jp) != NULL) {
798 if (jp + x->length > ip)
804 if ((x = *jp) == NULL) {
819 score_adjust(score0, x);
820 if (score0 < 0 && flag)
832 while (--undop >= undop1)
833 if ((*undop->pos = x = undop->val))
839 ccount_stat += ccount - p->ccount;
841 ncover_stat += ncover;
845 struct cc_undo *u = cc_undo;
846 while (--undop >= u) {
848 if ((*undop->pos = x = undop->val))
852 } while ((p = *pp++) != NULL);
857 cc_output_phase(char *buffer, struct cc **output, int bufsize)
862 for (i = 0; i < bufsize;) {
863 if ((p = output[i]) == NULL) {
866 } else if (p->code >= 0) {
867 if (--p->ccount == 0)
869 (*tt.tt_put_token)(p->code, p->string, p->length);
871 wwntoksave += put_token_score(p->length);
873 } else if ((p1 = cc_q0a.qback) != &cc_q0a) {
876 qinsert(p1, &cc_q1a);
877 if (--p->ccount == 0)
881 (*tt.tt_set_token)(p->code, p->string, p->length);
883 wwntoksave -= tt.tt_set_token_cost;
887 ttwrite(p->string, p->length);
896 cc_token_compare(const void *p1, const void *p2)
898 const struct cc * const * vp1 = p1;
899 const struct cc * const * vp2 = p2;
900 return (*vp2)->bcount - (*vp1)->bcount;