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";
50 /* tunable parameters */
56 int cc_token_max = 8; /* <= TOKEN_MAX */
57 int cc_token_min = 2; /* > tt.tt_put_token_cost */
61 int cc_bufsize = 1024 * 3; /* XXX, or 80 * 24 * 2 */
73 char string[TOKEN_MAX];
79 long time; /* time last seen */
80 short bcount; /* count in this buffer */
81 short ccount; /* count in compression */
82 short places; /* places in the buffer */
83 short code; /* token code */
84 struct cc *qforw, *qback;
85 struct cc *hforw, **hback;
88 short cc_thresholds[TOKEN_MAX + 1];
89 #define thresh(length) (cc_thresholds[length])
90 #define threshp(code, count, length) \
91 ((code) >= 0 || (short) (count) >= cc_thresholds[length])
94 short cc_wthresholds[TOKEN_MAX + 1];
95 #define wthresh(length) (cc_wthresholds[length])
96 #define wthreshp(weight, length) ((short) (weight) >= cc_wthresholds[length])
98 #define wthreshp(weight, length) (0)
102 short cc_wlimits[TOKEN_MAX + 1];
103 #define wlimit(length) (cc_wlimits[length])
106 #define put_token_score(length) ((length) - tt.tt_put_token_cost)
108 int cc_score_adjustments[TOKEN_MAX + 1][8]; /* XXX, 8 > max of cc_thresholds */
109 #define score_adjust(score, p) \
111 int length = (p)->length; \
112 int ccount = (p)->ccount; \
113 if (threshp((p)->code, ccount, length) || \
114 wthreshp((p)->weight, length)) /* XXX */ \
115 (score) -= length - tt.tt_put_token_cost; \
117 (score) += cc_score_adjustments[length][ccount]; \
120 int cc_initial_scores[TOKEN_MAX + 1][8]; /* XXX, 8 > max of cc_thresholds */
122 struct cc cc_q0a, cc_q0b, cc_q1a, cc_q1b;
124 #define qinsert(p1, p2) \
126 register struct cc *forw = (p1)->qforw; \
127 register struct cc *back = (p1)->qback; \
128 back->qforw = forw; \
129 forw->qback = back; \
130 forw = (p2)->qforw; \
131 (p1)->qforw = forw; \
132 forw->qback = (p1); \
133 (p2)->qforw = (p1); \
134 (p1)->qback = (p2); \
137 #define qinsertq(q, p) \
138 ((q)->qforw == (q) ? 0 : \
139 ((q)->qback->qforw = (p)->qforw, \
140 (p)->qforw->qback = (q)->qback, \
141 (q)->qforw->qback = (p), \
142 (p)->qforw = (q)->qforw, \
147 #define HSIZE (1 << H)
148 #define hash(h, c) ((((h) >> H - 8 | (h) << 8) ^ (c)) & HSIZE - 1)
151 struct cc **cc_output; /* the output array */
152 short *cc_places[TOKEN_MAX + 1];
153 short *cc_hashcodes; /* for computing hashcodes */
154 struct cc **cc_htab; /* the hash table */
155 struct cc **cc_tokens; /* holds all the active tokens */
161 long cc_time, cc_time0;
163 char *cc_tt_ob, *cc_tt_obe;
168 register struct cc *p;
170 if (tt.tt_token_max > cc_token_max)
171 tt.tt_token_max = cc_token_max;
172 if (tt.tt_token_min < cc_token_min)
173 tt.tt_token_min = cc_token_min;
174 if (tt.tt_token_min > tt.tt_token_max) {
178 if (tt.tt_ntoken > cc_ntoken / 2) /* not likely */
179 tt.tt_ntoken = cc_ntoken / 2;
180 #define C(x) (sizeof (x) / sizeof *(x))
181 for (i = 0; i < C(cc_thresholds); i++) {
182 int h = i - tt.tt_put_token_cost;
185 (tt.tt_set_token_cost + 1 + h - 1) / h + 1;
187 cc_thresholds[i] = 0;
189 for (i = 0; i < C(cc_score_adjustments); i++) {
190 int t = cc_thresholds[i];
191 for (j = 0; j < C(*cc_score_adjustments); j++) {
193 cc_score_adjustments[i][j] =
194 - (i - tt.tt_put_token_cost);
196 cc_score_adjustments[i][j] = 0;
200 * length * (ccount + 1) a
202 * set-token-cost + length +
203 * ccount * put-token-cost b
204 * the score adjustment is (b - a)
206 cc_score_adjustments[i][j] =
207 tt.tt_set_token_cost + i +
208 j * tt.tt_put_token_cost -
211 cc_initial_scores[i][j] = 0;
214 * - (set-token-cost +
215 * (length - put-token-cost) -
216 * (length - put-token-cost) * ccount)
218 cc_initial_scores[i][j] =
219 - (tt.tt_set_token_cost +
220 (i - tt.tt_put_token_cost) -
221 (i - tt.tt_put_token_cost) * j);
225 for (i = 1; i < C(cc_wthresholds); i++) {
227 ((tt.tt_set_token_cost + tt.tt_put_token_cost) / i +
230 cc_wlimits[i] = cc_wthresholds[i] + cc_weight;
234 if ((cc_output = (struct cc **)
235 malloc((unsigned) cc_bufsize * sizeof *cc_output)) == 0)
237 if ((cc_hashcodes = (short *)
238 malloc((unsigned) cc_bufsize * sizeof *cc_hashcodes)) == 0)
240 if ((cc_htab = (struct cc **) malloc(HSIZE * sizeof *cc_htab)) == 0)
242 if ((cc_tokens = (struct cc **)
244 (cc_ntoken + tt.tt_token_max - tt.tt_token_min + 1) *
245 sizeof *cc_tokens)) == 0)
247 if ((cc_undo = (struct cc_undo *)
248 malloc((unsigned) cc_bufsize * sizeof *cc_undo)) == 0)
250 for (i = tt.tt_token_min; i <= tt.tt_token_max; i++)
251 if ((cc_places[i] = (short *)
252 malloc((unsigned) cc_bufsize * sizeof **cc_places)) == 0)
254 cc_q0a.qforw = cc_q0a.qback = &cc_q0a;
255 cc_q0b.qforw = cc_q0b.qback = &cc_q0b;
256 cc_q1a.qforw = cc_q1a.qback = &cc_q1a;
257 cc_q1b.qforw = cc_q1b.qback = &cc_q1b;
258 if ((p = (struct cc *) malloc((unsigned) cc_ntoken * sizeof *p)) == 0)
260 for (i = 0; i < tt.tt_ntoken; i++) {
263 p->qback = cc_q0a.qback;
269 for (; i < cc_ntoken; i++) {
272 p->qback = cc_q1a.qback;
280 if ((cc_buffer = malloc((unsigned) cc_bufsize)) == 0)
293 tt_obp = tt_ob = cc_buffer;
294 tt_obe = tt_ob + cc_bufsize;
295 tt.tt_flush = ccflush;
297 cc_trace_fp = fopen("window-trace", "a");
298 (void) fcntl(fileno(cc_trace_fp), F_SETFD, 1);
305 register struct cc *p;
307 bzero((char *) cc_htab, HSIZE * sizeof *cc_htab);
308 for (p = cc_q0a.qforw; p != &cc_q0a; p = p->qforw)
310 for (p = cc_q1a.qforw; p != &cc_q1a; p = p->qforw)
318 tt_obp = tt_ob = cc_tt_ob;
321 if (cc_trace_fp != NULL) {
322 (void) fclose(cc_trace_fp);
329 int bufsize = tt_obp - tt_ob;
332 if (tt_ob != cc_buffer)
334 if (cc_trace_fp != NULL) {
335 (void) fwrite(tt_ob, 1, bufsize, cc_trace_fp);
336 (void) putc(-1, cc_trace_fp);
339 (*tt.tt_compress)(1);
340 if (bufsize < tt.tt_token_min) {
344 tt_obp = tt_ob = cc_tt_ob;
348 n = cc_sweep_phase(cc_buffer, bufsize, cc_tokens);
349 cc_compress_phase(cc_output, bufsize, cc_tokens, n);
350 cc_output_phase(cc_buffer, cc_output, bufsize);
352 tt_obp = tt_ob = cc_buffer;
353 tt_obe = cc_buffer + cc_bufsize;
355 (*tt.tt_compress)(0);
356 tt.tt_flush = ccflush;
359 cc_sweep_phase(buffer, bufsize, tokens)
363 register struct cc **pp = tokens;
375 cc_sweep0(buffer, bufsize, tt.tt_token_min - 1);
381 for (i = tt.tt_token_min; i <= tt.tt_token_max; i++) {
390 (void) fflush(stdout);
393 n = cc_sweep(buffer, bufsize, pp, i);
405 qinsertq(&cc_q1b, &cc_q1a);
408 printf("\n %d tokens, %d candidates\n",
416 cc_sweep0(buffer, n, length)
423 register short pc = tt.tt_padc;
425 /* n and length are at least 1 */
430 if ((*hc++ = *p++) == pc)
436 for (i = n--; --i;) {
437 if ((c = *p++) == pc || *hc < 0)
446 cc_sweep(buffer, bufsize, tokens, length)
451 register struct cc *p;
455 short *places = cc_places[length];
456 struct cc **pp = tokens;
457 short threshold = thresh(length);
459 short wthreshold = wthresh(length);
460 short limit = wlimit(length);
463 short pc = tt.tt_padc;
470 for (i = 0; i < bufsize; i++, time++) {
474 register short *hc1 = hc;
475 register short c = *cp++;
477 if ((hh = *hc1) < 0 || c == pc) {
482 h = cc_htab + (*hc1++ = hash(hh, c));
485 for (p = *h; p != 0; p = p->hforw)
486 if (p->length == (char) length) {
487 register char *p1 = p->string;
488 register char *p2 = cp - length;
500 p->time >= cc_time0 && p->length == (char) length)
503 if ((*p->hback = p->hforw) != 0)
504 p->hforw->hback = p->hback;
506 register char *p1 = p->string;
507 register char *p2 = cp - length;
515 p->weight = cc_weight;
521 if ((p->hforw = *h) != 0)
522 p->hforw->hback = &p->hforw;
531 } else if (p->time < cc_time0) {
533 if ((p->weight += p->time - time) < 0)
534 p->weight = cc_weight;
535 else if ((p->weight += cc_weight) > limit)
546 if (p->weight >= wthreshold) {
561 } else if (p->time + length > time) {
563 * overlapping token, don't count as two and
564 * don't update time, but do adjust weight to offset
568 if (cc_weight != 0) { /* XXX */
569 p->weight += time - p->time;
570 if (!p->flag && p->weight >= wthreshold) {
577 places[i] = p->places;
581 if ((p->weight += p->time - time) < 0)
582 p->weight = cc_weight;
583 else if ((p->weight += cc_weight) > limit)
589 /* code must be < 0 if flag false here */
590 (p->bcount >= threshold
592 || p->weight >= wthreshold
599 places[i] = p->places;
603 if ((i = pp - tokens) > 0) {
606 cc_sweep_reverse(tokens, places);
607 if (cc_sort && i > 1) {
608 int cc_token_compare();
609 qsort((char *) tokens, i, sizeof *tokens,
613 if ((i = i * cc_chop / 100) == 0)
622 cc_sweep_reverse(pp, places)
623 register struct cc **pp;
624 register short *places;
626 register struct cc *p;
627 register short front, back, t;
629 while ((p = *pp++) != 0) {
632 /* the list is never empty */
637 } while ((t = front) >= 0);
642 cc_compress_phase(output, bufsize, tokens, ntoken)
648 bzero((char *) output, bufsize * sizeof *output);
649 for (i = 0; i < cc_npass0; i++)
650 cc_compress_phase1(output, tokens, ntoken, 0);
651 for (i = 0; i < cc_npass1; i++)
652 cc_compress_phase1(output, tokens, ntoken, 1);
653 cc_compress_cleanup(output, bufsize);
656 cc_compress_phase1(output, tokens, ntoken, flag)
657 register struct cc **output;
660 register struct cc **pp;
663 int nt = 0, cc = 0, nc = 0;
673 while (pp < tokens + ntoken) {
684 printf(" (%d", (*pp)->length);
685 (void) fflush(stdout);
688 pp += cc_compress(output, pp, flag);
691 printf(" %dt %du %dc)", ntoken_stat, ccount_stat,
701 printf("\n total: (%dt %du %dc)\n", nt, cc, nc);
707 cc_compress_cleanup(output, bufsize)
708 register struct cc **output;
710 register struct cc **end;
712 /* the previous output phase may have been interrupted */
713 qinsertq(&cc_q0b, &cc_q0a);
714 for (end = output + bufsize; output < end;) {
715 register struct cc *p;
717 if ((p = *output) == 0) {
723 } else if (p->code >= 0) {
726 } else if (p->ccount == 0) {
728 } else if (p->ccount >= thresh(length)
730 || wthreshp(p->weight, length)
742 cc_compress(output, tokens, flag)
747 struct cc **pp = tokens;
748 register struct cc *p = *pp++;
749 int length = p->length;
750 int threshold = thresh(length);
752 short wthreshold = wthresh(length);
754 short *places = cc_places[length];
755 int *initial_scores = cc_initial_scores[length];
756 int initial_score0 = put_token_score(length);
760 register struct cc_undo *undop;
768 if ((short) ccount >= p->bcount)
770 if (p->code >= 0 || ccount >= threshold)
773 else if (p->weight >= wthreshold)
774 /* allow one fewer match than normal */
775 /* XXX, should adjust for ccount */
776 score = - tt.tt_set_token_cost;
779 score = initial_scores[ccount];
784 for (i = p->places; i >= 0; i = places[i]) {
785 register struct cc **jp;
786 register struct cc *x;
787 register struct cc **ip = output + i;
788 register score0 = initial_score0;
789 struct cc **iip = ip + length;
790 struct cc_undo *undop1 = undop;
792 if ((x = *(jp = ip)) != 0)
794 while (--jp >= output)
795 if ((x = *jp) != 0) {
796 if (jp + x->length > ip)
802 if ((x = *jp) == 0) {
817 score_adjust(score0, x);
818 if (score0 < 0 && flag)
830 while (--undop >= undop1)
831 if (*undop->pos = x = undop->val)
837 ccount_stat += ccount - p->ccount;
839 ncover_stat += ncover;
843 register struct cc_undo *u = cc_undo;
844 while (--undop >= u) {
845 register struct cc *x;
846 if (*undop->pos = x = undop->val)
850 } while ((p = *pp++) != 0);
854 cc_output_phase(buffer, output, bufsize)
855 register char *buffer;
856 register struct cc **output;
860 register struct cc *p, *p1;
862 for (i = 0; i < bufsize;) {
863 if ((p = output[i]) == 0) {
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);
895 cc_token_compare(p1, p2)
896 struct cc **p1, **p2;
898 return (*p2)->bcount - (*p1)->bcount;