xref: /openbsd/sys/lib/libz/inffast.c (revision 4fb9ab68)
1 /* inffast.c -- fast decoding
2  * Copyright (C) 1995-2017 Mark Adler
3  * For conditions of distribution and use, see copyright notice in zlib.h
4  */
5 
6 #include "zutil.h"
7 #include "inftrees.h"
8 #include "inflate.h"
9 #include "inffast.h"
10 
11 #ifdef ASMINF
12 #  pragma message("Assembler code may have bugs -- use at your own risk")
13 #else
14 
15 /*
16    Decode literal, length, and distance codes and write out the resulting
17    literal and match bytes until either not enough input or output is
18    available, an end-of-block is encountered, or a data error is encountered.
19    When large enough input and output buffers are supplied to inflate(), for
20    example, a 16K input buffer and a 64K output buffer, more than 95% of the
21    inflate execution time is spent in this routine.
22 
23    Entry assumptions:
24 
25         state->mode == LEN
26         strm->avail_in >= 6
27         strm->avail_out >= 258
28         start >= strm->avail_out
29         state->bits < 8
30 
31    On return, state->mode is one of:
32 
33         LEN -- ran out of enough output space or enough available input
34         TYPE -- reached end of block code, inflate() to interpret next block
35         BAD -- error in block data
36 
37    Notes:
38 
39     - The maximum input bits used by a length/distance pair is 15 bits for the
40       length code, 5 bits for the length extra, 15 bits for the distance code,
41       and 13 bits for the distance extra.  This totals 48 bits, or six bytes.
42       Therefore if strm->avail_in >= 6, then there is enough input to avoid
43       checking for available input while decoding.
44 
45     - The maximum bytes that a single length/distance pair can output is 258
46       bytes, which is the maximum length that can be coded.  inflate_fast()
47       requires strm->avail_out >= 258 for each loop to avoid checking for
48       output space.
49  */
50 void ZLIB_INTERNAL inflate_fast(z_streamp strm, unsigned start) {
51     struct inflate_state FAR *state;
52     z_const unsigned char FAR *in;      /* local strm->next_in */
53     z_const unsigned char FAR *last;    /* have enough input while in < last */
54     unsigned char FAR *out;     /* local strm->next_out */
55     unsigned char FAR *beg;     /* inflate()'s initial strm->next_out */
56     unsigned char FAR *end;     /* while out < end, enough space available */
57 #ifdef INFLATE_STRICT
58     unsigned dmax;              /* maximum distance from zlib header */
59 #endif
60     unsigned wsize;             /* window size or zero if not using window */
61     unsigned whave;             /* valid bytes in the window */
62     unsigned wnext;             /* window write index */
63     unsigned char FAR *window;  /* allocated sliding window, if wsize != 0 */
64     unsigned long hold;         /* local strm->hold */
65     unsigned bits;              /* local strm->bits */
66     code const FAR *lcode;      /* local strm->lencode */
67     code const FAR *dcode;      /* local strm->distcode */
68     unsigned lmask;             /* mask for first level of length codes */
69     unsigned dmask;             /* mask for first level of distance codes */
70     code const *here;           /* retrieved table entry */
71     unsigned op;                /* code bits, operation, extra bits, or */
72                                 /*  window position, window bytes to copy */
73     unsigned len;               /* match length, unused bytes */
74     unsigned dist;              /* match distance */
75     unsigned char FAR *from;    /* where to copy match from */
76 
77     /* copy state to local variables */
78     state = (struct inflate_state FAR *)strm->state;
79     in = strm->next_in;
80     last = in + (strm->avail_in - 5);
81     out = strm->next_out;
82     beg = out - (start - strm->avail_out);
83     end = out + (strm->avail_out - 257);
84 #ifdef INFLATE_STRICT
85     dmax = state->dmax;
86 #endif
87     wsize = state->wsize;
88     whave = state->whave;
89     wnext = state->wnext;
90     window = state->window;
91     hold = state->hold;
92     bits = state->bits;
93     lcode = state->lencode;
94     dcode = state->distcode;
95     lmask = (1U << state->lenbits) - 1;
96     dmask = (1U << state->distbits) - 1;
97 
98     /* decode literals and length/distances until end-of-block or not enough
99        input data or output space */
100     do {
101         if (bits < 15) {
102             hold += (unsigned long)(*in++) << bits;
103             bits += 8;
104             hold += (unsigned long)(*in++) << bits;
105             bits += 8;
106         }
107         here = lcode + (hold & lmask);
108       dolen:
109         op = (unsigned)(here->bits);
110         hold >>= op;
111         bits -= op;
112         op = (unsigned)(here->op);
113         if (op == 0) {                          /* literal */
114             Tracevv((stderr, here->val >= 0x20 && here->val < 0x7f ?
115                     "inflate:         literal '%c'\n" :
116                     "inflate:         literal 0x%02x\n", here->val));
117             *out++ = (unsigned char)(here->val);
118         }
119         else if (op & 16) {                     /* length base */
120             len = (unsigned)(here->val);
121             op &= 15;                           /* number of extra bits */
122             if (op) {
123                 if (bits < op) {
124                     hold += (unsigned long)(*in++) << bits;
125                     bits += 8;
126                 }
127                 len += (unsigned)hold & ((1U << op) - 1);
128                 hold >>= op;
129                 bits -= op;
130             }
131             Tracevv((stderr, "inflate:         length %u\n", len));
132             if (bits < 15) {
133                 hold += (unsigned long)(*in++) << bits;
134                 bits += 8;
135                 hold += (unsigned long)(*in++) << bits;
136                 bits += 8;
137             }
138             here = dcode + (hold & dmask);
139           dodist:
140             op = (unsigned)(here->bits);
141             hold >>= op;
142             bits -= op;
143             op = (unsigned)(here->op);
144             if (op & 16) {                      /* distance base */
145                 dist = (unsigned)(here->val);
146                 op &= 15;                       /* number of extra bits */
147                 if (bits < op) {
148                     hold += (unsigned long)(*in++) << bits;
149                     bits += 8;
150                     if (bits < op) {
151                         hold += (unsigned long)(*in++) << bits;
152                         bits += 8;
153                     }
154                 }
155                 dist += (unsigned)hold & ((1U << op) - 1);
156 #ifdef INFLATE_STRICT
157                 if (dist > dmax) {
158                     strm->msg = (z_const char *)"invalid distance too far back";
159                     state->mode = BAD;
160                     break;
161                 }
162 #endif
163                 hold >>= op;
164                 bits -= op;
165                 Tracevv((stderr, "inflate:         distance %u\n", dist));
166                 op = (unsigned)(out - beg);     /* max distance in output */
167                 if (dist > op) {                /* see if copy from window */
168                     op = dist - op;             /* distance back in window */
169                     if (op > whave) {
170                         if (state->sane) {
171 #ifdef SMALL
172                             strm->msg = (z_const char *)"error";
173 #else
174                             strm->msg =
175                                 (z_const char *)"invalid distance too far back";
176 #endif
177                             state->mode = BAD;
178                             break;
179                         }
180 #ifdef INFLATE_ALLOW_INVALID_DISTANCE_TOOFAR_ARRR
181                         if (len <= op - whave) {
182                             do {
183                                 *out++ = 0;
184                             } while (--len);
185                             continue;
186                         }
187                         len -= op - whave;
188                         do {
189                             *out++ = 0;
190                         } while (--op > whave);
191                         if (op == 0) {
192                             from = out - dist;
193                             do {
194                                 *out++ = *from++;
195                             } while (--len);
196                             continue;
197                         }
198 #endif
199                     }
200                     from = window;
201                     if (wnext == 0) {           /* very common case */
202                         from += wsize - op;
203                         if (op < len) {         /* some from window */
204                             len -= op;
205                             do {
206                                 *out++ = *from++;
207                             } while (--op);
208                             from = out - dist;  /* rest from output */
209                         }
210                     }
211                     else if (wnext < op) {      /* wrap around window */
212                         from += wsize + wnext - op;
213                         op -= wnext;
214                         if (op < len) {         /* some from end of window */
215                             len -= op;
216                             do {
217                                 *out++ = *from++;
218                             } while (--op);
219                             from = window;
220                             if (wnext < len) {  /* some from start of window */
221                                 op = wnext;
222                                 len -= op;
223                                 do {
224                                     *out++ = *from++;
225                                 } while (--op);
226                                 from = out - dist;      /* rest from output */
227                             }
228                         }
229                     }
230                     else {                      /* contiguous in window */
231                         from += wnext - op;
232                         if (op < len) {         /* some from window */
233                             len -= op;
234                             do {
235                                 *out++ = *from++;
236                             } while (--op);
237                             from = out - dist;  /* rest from output */
238                         }
239                     }
240                     while (len > 2) {
241                         *out++ = *from++;
242                         *out++ = *from++;
243                         *out++ = *from++;
244                         len -= 3;
245                     }
246                     if (len) {
247                         *out++ = *from++;
248                         if (len > 1)
249                             *out++ = *from++;
250                     }
251                 }
252                 else {
253                     from = out - dist;          /* copy direct from output */
254                     do {                        /* minimum length is three */
255                         *out++ = *from++;
256                         *out++ = *from++;
257                         *out++ = *from++;
258                         len -= 3;
259                     } while (len > 2);
260                     if (len) {
261                         *out++ = *from++;
262                         if (len > 1)
263                             *out++ = *from++;
264                     }
265                 }
266             }
267             else if ((op & 64) == 0) {          /* 2nd level distance code */
268                 here = dcode + here->val + (hold & ((1U << op) - 1));
269                 goto dodist;
270             }
271             else {
272 #ifdef SMALL
273                 strm->msg = (z_const char *)"error";
274 #else
275                 strm->msg = (z_const char *)"invalid distance code";
276 #endif
277                 state->mode = BAD;
278                 break;
279             }
280         }
281         else if ((op & 64) == 0) {              /* 2nd level length code */
282             here = lcode + here->val + (hold & ((1U << op) - 1));
283             goto dolen;
284         }
285         else if (op & 32) {                     /* end-of-block */
286             Tracevv((stderr, "inflate:         end of block\n"));
287             state->mode = TYPE;
288             break;
289         }
290         else {
291 #ifdef SMALL
292             strm->msg = (z_const char *)"error";
293 #else
294             strm->msg = (z_const char *)"invalid literal/length code";
295 #endif
296             state->mode = BAD;
297             break;
298         }
299     } while (in < last && out < end);
300 
301     /* return unused bytes (on entry, bits < 8, so in won't go too far back) */
302     len = bits >> 3;
303     in -= len;
304     bits -= len << 3;
305     hold &= (1U << bits) - 1;
306 
307     /* update state and return */
308     strm->next_in = in;
309     strm->next_out = out;
310     strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last));
311     strm->avail_out = (unsigned)(out < end ?
312                                  257 + (end - out) : 257 - (out - end));
313     state->hold = hold;
314     state->bits = bits;
315     return;
316 }
317 
318 /*
319    inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe):
320    - Using bit fields for code structure
321    - Different op definition to avoid & for extra bits (do & for table bits)
322    - Three separate decoding do-loops for direct, window, and wnext == 0
323    - Special case for distance > 1 copies to do overlapped load and store copy
324    - Explicit branch predictions (based on measured branch probabilities)
325    - Deferring match copy and interspersed it with decoding subsequent codes
326    - Swapping literal/length else
327    - Swapping window/direct else
328    - Larger unrolled copy loops (three is about right)
329    - Moving len -= 3 statement into middle of loop
330  */
331 
332 #endif /* !ASMINF */
333