2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Edward Wang at The University of California, Berkeley.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 static char sccsid[] = "@(#)compress.c 8.1 (Berkeley) 6/6/93";
39 static char rcsid[] = "@(#)$FreeBSD$";
52 /* tunable parameters */
58 int cc_token_max = 8; /* <= TOKEN_MAX */
59 int cc_token_min = 2; /* > tt.tt_put_token_cost */
63 int cc_bufsize = 1024 * 3; /* XXX, or 80 * 24 * 2 */
75 char string[TOKEN_MAX];
81 long time; /* time last seen */
82 short bcount; /* count in this buffer */
83 short ccount; /* count in compression */
84 short places; /* places in the buffer */
85 short code; /* token code */
86 struct cc *qforw, *qback;
87 struct cc *hforw, **hback;
90 short cc_thresholds[TOKEN_MAX + 1];
91 #define thresh(length) (cc_thresholds[length])
92 #define threshp(code, count, length) \
93 ((code) >= 0 || (short) (count) >= cc_thresholds[length])
96 short cc_wthresholds[TOKEN_MAX + 1];
97 #define wthresh(length) (cc_wthresholds[length])
98 #define wthreshp(weight, length) ((short) (weight) >= cc_wthresholds[length])
100 #define wthreshp(weight, length) (0)
104 short cc_wlimits[TOKEN_MAX + 1];
105 #define wlimit(length) (cc_wlimits[length])
108 #define put_token_score(length) ((length) - tt.tt_put_token_cost)
110 int cc_score_adjustments[TOKEN_MAX + 1][8]; /* XXX, 8 > max of cc_thresholds */
111 #define score_adjust(score, p) \
113 int length = (p)->length; \
114 int ccount = (p)->ccount; \
115 if (threshp((p)->code, ccount, length) || \
116 wthreshp((p)->weight, length)) /* XXX */ \
117 (score) -= length - tt.tt_put_token_cost; \
119 (score) += cc_score_adjustments[length][ccount]; \
122 int cc_initial_scores[TOKEN_MAX + 1][8]; /* XXX, 8 > max of cc_thresholds */
124 struct cc cc_q0a, cc_q0b, cc_q1a, cc_q1b;
126 #define qinsert(p1, p2) \
128 register struct cc *forw = (p1)->qforw; \
129 register struct cc *back = (p1)->qback; \
130 back->qforw = forw; \
131 forw->qback = back; \
132 forw = (p2)->qforw; \
133 (p1)->qforw = forw; \
134 forw->qback = (p1); \
135 (p2)->qforw = (p1); \
136 (p1)->qback = (p2); \
139 #define qinsertq(q, p) \
140 ((q)->qforw == (q) ? 0 : \
141 ((q)->qback->qforw = (p)->qforw, \
142 (p)->qforw->qback = (q)->qback, \
143 (q)->qforw->qback = (p), \
144 (p)->qforw = (q)->qforw, \
149 #define HSIZE (1 << H)
150 #define hash(h, c) ((((h) >> H - 8 | (h) << 8) ^ (c)) & HSIZE - 1)
153 struct cc **cc_output; /* the output array */
154 short *cc_places[TOKEN_MAX + 1];
155 short *cc_hashcodes; /* for computing hashcodes */
156 struct cc **cc_htab; /* the hash table */
157 struct cc **cc_tokens; /* holds all the active tokens */
163 long cc_time, cc_time0;
165 char *cc_tt_ob, *cc_tt_obe;
170 register struct cc *p;
172 if (tt.tt_token_max > cc_token_max)
173 tt.tt_token_max = cc_token_max;
174 if (tt.tt_token_min < cc_token_min)
175 tt.tt_token_min = cc_token_min;
176 if (tt.tt_token_min > tt.tt_token_max) {
180 if (tt.tt_ntoken > cc_ntoken / 2) /* not likely */
181 tt.tt_ntoken = cc_ntoken / 2;
182 #define C(x) (sizeof (x) / sizeof *(x))
183 for (i = 0; i < C(cc_thresholds); i++) {
184 int h = i - tt.tt_put_token_cost;
187 (tt.tt_set_token_cost + 1 + h - 1) / h + 1;
189 cc_thresholds[i] = 0;
191 for (i = 0; i < C(cc_score_adjustments); i++) {
192 int t = cc_thresholds[i];
193 for (j = 0; j < C(*cc_score_adjustments); j++) {
195 cc_score_adjustments[i][j] =
196 - (i - tt.tt_put_token_cost);
198 cc_score_adjustments[i][j] = 0;
202 * length * (ccount + 1) a
204 * set-token-cost + length +
205 * ccount * put-token-cost b
206 * the score adjustment is (b - a)
208 cc_score_adjustments[i][j] =
209 tt.tt_set_token_cost + i +
210 j * tt.tt_put_token_cost -
213 cc_initial_scores[i][j] = 0;
216 * - (set-token-cost +
217 * (length - put-token-cost) -
218 * (length - put-token-cost) * ccount)
220 cc_initial_scores[i][j] =
221 - (tt.tt_set_token_cost +
222 (i - tt.tt_put_token_cost) -
223 (i - tt.tt_put_token_cost) * j);
227 for (i = 1; i < C(cc_wthresholds); i++) {
229 ((tt.tt_set_token_cost + tt.tt_put_token_cost) / i +
232 cc_wlimits[i] = cc_wthresholds[i] + cc_weight;
236 if ((cc_output = (struct cc **)
237 malloc((unsigned) cc_bufsize * sizeof *cc_output)) == 0)
239 if ((cc_hashcodes = (short *)
240 malloc((unsigned) cc_bufsize * sizeof *cc_hashcodes)) == 0)
242 if ((cc_htab = (struct cc **) malloc(HSIZE * sizeof *cc_htab)) == 0)
244 if ((cc_tokens = (struct cc **)
246 (cc_ntoken + tt.tt_token_max - tt.tt_token_min + 1) *
247 sizeof *cc_tokens)) == 0)
249 if ((cc_undo = (struct cc_undo *)
250 malloc((unsigned) cc_bufsize * sizeof *cc_undo)) == 0)
252 for (i = tt.tt_token_min; i <= tt.tt_token_max; i++)
253 if ((cc_places[i] = (short *)
254 malloc((unsigned) cc_bufsize * sizeof **cc_places)) == 0)
256 cc_q0a.qforw = cc_q0a.qback = &cc_q0a;
257 cc_q0b.qforw = cc_q0b.qback = &cc_q0b;
258 cc_q1a.qforw = cc_q1a.qback = &cc_q1a;
259 cc_q1b.qforw = cc_q1b.qback = &cc_q1b;
260 if ((p = (struct cc *) malloc((unsigned) cc_ntoken * sizeof *p)) == 0)
262 for (i = 0; i < tt.tt_ntoken; i++) {
265 p->qback = cc_q0a.qback;
271 for (; i < cc_ntoken; i++) {
274 p->qback = cc_q1a.qback;
282 if ((cc_buffer = malloc((unsigned) cc_bufsize)) == 0)
295 tt_obp = tt_ob = cc_buffer;
296 tt_obe = tt_ob + cc_bufsize;
297 tt.tt_flush = ccflush;
299 cc_trace_fp = fopen("window-trace", "a");
300 (void) fcntl(fileno(cc_trace_fp), F_SETFD, 1);
307 register struct cc *p;
309 bzero((char *) cc_htab, HSIZE * sizeof *cc_htab);
310 for (p = cc_q0a.qforw; p != &cc_q0a; p = p->qforw)
312 for (p = cc_q1a.qforw; p != &cc_q1a; p = p->qforw)
320 tt_obp = tt_ob = cc_tt_ob;
323 if (cc_trace_fp != NULL) {
324 (void) fclose(cc_trace_fp);
331 int bufsize = tt_obp - tt_ob;
334 if (tt_ob != cc_buffer)
336 if (cc_trace_fp != NULL) {
337 (void) fwrite(tt_ob, 1, bufsize, cc_trace_fp);
338 (void) putc(-1, cc_trace_fp);
341 (*tt.tt_compress)(1);
342 if (bufsize < tt.tt_token_min) {
346 tt_obp = tt_ob = cc_tt_ob;
350 n = cc_sweep_phase(cc_buffer, bufsize, cc_tokens);
351 cc_compress_phase(cc_output, bufsize, cc_tokens, n);
352 cc_output_phase(cc_buffer, cc_output, bufsize);
354 tt_obp = tt_ob = cc_buffer;
355 tt_obe = cc_buffer + cc_bufsize;
357 (*tt.tt_compress)(0);
358 tt.tt_flush = ccflush;
361 cc_sweep_phase(buffer, bufsize, tokens)
365 register struct cc **pp = tokens;
377 cc_sweep0(buffer, bufsize, tt.tt_token_min - 1);
383 for (i = tt.tt_token_min; i <= tt.tt_token_max; i++) {
392 (void) fflush(stdout);
395 n = cc_sweep(buffer, bufsize, pp, i);
407 qinsertq(&cc_q1b, &cc_q1a);
410 printf("\n %d tokens, %d candidates\n",
418 cc_sweep0(buffer, n, length)
425 register short pc = tt.tt_padc;
427 /* n and length are at least 1 */
432 if ((*hc++ = *p++) == pc)
438 for (i = n--; --i;) {
439 if ((c = *p++) == pc || *hc < 0)
448 cc_sweep(buffer, bufsize, tokens, length)
453 register struct cc *p;
457 short *places = cc_places[length];
458 struct cc **pp = tokens;
459 short threshold = thresh(length);
461 short wthreshold = wthresh(length);
462 short limit = wlimit(length);
465 short pc = tt.tt_padc;
472 for (i = 0; i < bufsize; i++, time++) {
476 register short *hc1 = hc;
477 register short c = *cp++;
479 if ((hh = *hc1) < 0 || c == pc) {
484 h = cc_htab + (*hc1++ = hash(hh, c));
487 for (p = *h; p != 0; p = p->hforw)
488 if (p->length == (char) length) {
489 register char *p1 = p->string;
490 register char *p2 = cp - length;
502 p->time >= cc_time0 && p->length == (char) length)
505 if ((*p->hback = p->hforw) != 0)
506 p->hforw->hback = p->hback;
508 register char *p1 = p->string;
509 register char *p2 = cp - length;
517 p->weight = cc_weight;
523 if ((p->hforw = *h) != 0)
524 p->hforw->hback = &p->hforw;
533 } else if (p->time < cc_time0) {
535 if ((p->weight += p->time - time) < 0)
536 p->weight = cc_weight;
537 else if ((p->weight += cc_weight) > limit)
548 if (p->weight >= wthreshold) {
563 } else if (p->time + length > time) {
565 * overlapping token, don't count as two and
566 * don't update time, but do adjust weight to offset
570 if (cc_weight != 0) { /* XXX */
571 p->weight += time - p->time;
572 if (!p->flag && p->weight >= wthreshold) {
579 places[i] = p->places;
583 if ((p->weight += p->time - time) < 0)
584 p->weight = cc_weight;
585 else if ((p->weight += cc_weight) > limit)
591 /* code must be < 0 if flag false here */
592 (p->bcount >= threshold
594 || p->weight >= wthreshold
601 places[i] = p->places;
605 if ((i = pp - tokens) > 0) {
608 cc_sweep_reverse(tokens, places);
609 if (cc_sort && i > 1) {
610 int cc_token_compare();
611 qsort((char *) tokens, i, sizeof *tokens,
615 if ((i = i * cc_chop / 100) == 0)
624 cc_sweep_reverse(pp, places)
625 register struct cc **pp;
626 register short *places;
628 register struct cc *p;
629 register short front, back, t;
631 while ((p = *pp++) != 0) {
634 /* the list is never empty */
639 } while ((t = front) >= 0);
644 cc_compress_phase(output, bufsize, tokens, ntoken)
650 bzero((char *) output, bufsize * sizeof *output);
651 for (i = 0; i < cc_npass0; i++)
652 cc_compress_phase1(output, tokens, ntoken, 0);
653 for (i = 0; i < cc_npass1; i++)
654 cc_compress_phase1(output, tokens, ntoken, 1);
655 cc_compress_cleanup(output, bufsize);
658 cc_compress_phase1(output, tokens, ntoken, flag)
659 register struct cc **output;
662 register struct cc **pp;
665 int nt = 0, cc = 0, nc = 0;
675 while (pp < tokens + ntoken) {
686 printf(" (%d", (*pp)->length);
687 (void) fflush(stdout);
690 pp += cc_compress(output, pp, flag);
693 printf(" %dt %du %dc)", ntoken_stat, ccount_stat,
703 printf("\n total: (%dt %du %dc)\n", nt, cc, nc);
709 cc_compress_cleanup(output, bufsize)
710 register struct cc **output;
712 register struct cc **end;
714 /* the previous output phase may have been interrupted */
715 qinsertq(&cc_q0b, &cc_q0a);
716 for (end = output + bufsize; output < end;) {
717 register struct cc *p;
719 if ((p = *output) == 0) {
725 } else if (p->code >= 0) {
728 } else if (p->ccount == 0) {
730 } else if (p->ccount >= thresh(length)
732 || wthreshp(p->weight, length)
744 cc_compress(output, tokens, flag)
749 struct cc **pp = tokens;
750 register 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 register 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]) {
787 register struct cc **jp;
788 register struct cc *x;
789 register struct cc **ip = output + i;
790 register score0 = initial_score0;
791 struct cc **iip = ip + length;
792 struct cc_undo *undop1 = undop;
794 if ((x = *(jp = ip)) != 0)
796 while (--jp >= output)
797 if ((x = *jp) != 0) {
798 if (jp + x->length > ip)
804 if ((x = *jp) == 0) {
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 register struct cc_undo *u = cc_undo;
846 while (--undop >= u) {
847 register struct cc *x;
848 if (*undop->pos = x = undop->val)
852 } while ((p = *pp++) != 0);
856 cc_output_phase(buffer, output, bufsize)
857 register char *buffer;
858 register struct cc **output;
862 register struct cc *p, *p1;
864 for (i = 0; i < bufsize;) {
865 if ((p = output[i]) == 0) {
868 } else if (p->code >= 0) {
869 if (--p->ccount == 0)
871 (*tt.tt_put_token)(p->code, p->string, p->length);
873 wwntoksave += put_token_score(p->length);
875 } else if ((p1 = cc_q0a.qback) != &cc_q0a) {
878 qinsert(p1, &cc_q1a);
879 if (--p->ccount == 0)
883 (*tt.tt_set_token)(p->code, p->string, p->length);
885 wwntoksave -= tt.tt_set_token_cost;
889 ttwrite(p->string, p->length);
897 cc_token_compare(p1, p2)
898 struct cc **p1, **p2;
900 return (*p2)->bcount - (*p1)->bcount;