xref: /openbsd-src/sys/lib/libz/inffast.c (revision 3374c67d44f9b75b98444cbf63020f777792342e)
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(strm, start)
51 z_streamp strm;
52 unsigned start;         /* inflate()'s starting value for strm->avail_out */
53 {
54     struct inflate_state FAR *state;
55     z_const unsigned char FAR *in;      /* local strm->next_in */
56     z_const unsigned char FAR *last;    /* have enough input while in < last */
57     unsigned char FAR *out;     /* local strm->next_out */
58     unsigned char FAR *beg;     /* inflate()'s initial strm->next_out */
59     unsigned char FAR *end;     /* while out < end, enough space available */
60 #ifdef INFLATE_STRICT
61     unsigned dmax;              /* maximum distance from zlib header */
62 #endif
63     unsigned wsize;             /* window size or zero if not using window */
64     unsigned whave;             /* valid bytes in the window */
65     unsigned wnext;             /* window write index */
66     unsigned char FAR *window;  /* allocated sliding window, if wsize != 0 */
67     unsigned long hold;         /* local strm->hold */
68     unsigned bits;              /* local strm->bits */
69     code const FAR *lcode;      /* local strm->lencode */
70     code const FAR *dcode;      /* local strm->distcode */
71     unsigned lmask;             /* mask for first level of length codes */
72     unsigned dmask;             /* mask for first level of distance codes */
73     code const *here;           /* retrieved table entry */
74     unsigned op;                /* code bits, operation, extra bits, or */
75                                 /*  window position, window bytes to copy */
76     unsigned len;               /* match length, unused bytes */
77     unsigned dist;              /* match distance */
78     unsigned char FAR *from;    /* where to copy match from */
79 
80     /* copy state to local variables */
81     state = (struct inflate_state FAR *)strm->state;
82     in = strm->next_in;
83     last = in + (strm->avail_in - 5);
84     out = strm->next_out;
85     beg = out - (start - strm->avail_out);
86     end = out + (strm->avail_out - 257);
87 #ifdef INFLATE_STRICT
88     dmax = state->dmax;
89 #endif
90     wsize = state->wsize;
91     whave = state->whave;
92     wnext = state->wnext;
93     window = state->window;
94     hold = state->hold;
95     bits = state->bits;
96     lcode = state->lencode;
97     dcode = state->distcode;
98     lmask = (1U << state->lenbits) - 1;
99     dmask = (1U << state->distbits) - 1;
100 
101     /* decode literals and length/distances until end-of-block or not enough
102        input data or output space */
103     do {
104         if (bits < 15) {
105             hold += (unsigned long)(*in++) << bits;
106             bits += 8;
107             hold += (unsigned long)(*in++) << bits;
108             bits += 8;
109         }
110         here = lcode + (hold & lmask);
111       dolen:
112         op = (unsigned)(here->bits);
113         hold >>= op;
114         bits -= op;
115         op = (unsigned)(here->op);
116         if (op == 0) {                          /* literal */
117             Tracevv((stderr, here->val >= 0x20 && here->val < 0x7f ?
118                     "inflate:         literal '%c'\n" :
119                     "inflate:         literal 0x%02x\n", here->val));
120             *out++ = (unsigned char)(here->val);
121         }
122         else if (op & 16) {                     /* length base */
123             len = (unsigned)(here->val);
124             op &= 15;                           /* number of extra bits */
125             if (op) {
126                 if (bits < op) {
127                     hold += (unsigned long)(*in++) << bits;
128                     bits += 8;
129                 }
130                 len += (unsigned)hold & ((1U << op) - 1);
131                 hold >>= op;
132                 bits -= op;
133             }
134             Tracevv((stderr, "inflate:         length %u\n", len));
135             if (bits < 15) {
136                 hold += (unsigned long)(*in++) << bits;
137                 bits += 8;
138                 hold += (unsigned long)(*in++) << bits;
139                 bits += 8;
140             }
141             here = dcode + (hold & dmask);
142           dodist:
143             op = (unsigned)(here->bits);
144             hold >>= op;
145             bits -= op;
146             op = (unsigned)(here->op);
147             if (op & 16) {                      /* distance base */
148                 dist = (unsigned)(here->val);
149                 op &= 15;                       /* number of extra bits */
150                 if (bits < op) {
151                     hold += (unsigned long)(*in++) << bits;
152                     bits += 8;
153                     if (bits < op) {
154                         hold += (unsigned long)(*in++) << bits;
155                         bits += 8;
156                     }
157                 }
158                 dist += (unsigned)hold & ((1U << op) - 1);
159 #ifdef INFLATE_STRICT
160                 if (dist > dmax) {
161                     strm->msg = (char *)"invalid distance too far back";
162                     state->mode = BAD;
163                     break;
164                 }
165 #endif
166                 hold >>= op;
167                 bits -= op;
168                 Tracevv((stderr, "inflate:         distance %u\n", dist));
169                 op = (unsigned)(out - beg);     /* max distance in output */
170                 if (dist > op) {                /* see if copy from window */
171                     op = dist - op;             /* distance back in window */
172                     if (op > whave) {
173                         if (state->sane) {
174 #ifdef SMALL
175                             strm->msg = "error";
176 #else
177                             strm->msg =
178                                 (char *)"invalid distance too far back";
179 #endif
180                             state->mode = BAD;
181                             break;
182                         }
183 #ifdef INFLATE_ALLOW_INVALID_DISTANCE_TOOFAR_ARRR
184                         if (len <= op - whave) {
185                             do {
186                                 *out++ = 0;
187                             } while (--len);
188                             continue;
189                         }
190                         len -= op - whave;
191                         do {
192                             *out++ = 0;
193                         } while (--op > whave);
194                         if (op == 0) {
195                             from = out - dist;
196                             do {
197                                 *out++ = *from++;
198                             } while (--len);
199                             continue;
200                         }
201 #endif
202                     }
203                     from = window;
204                     if (wnext == 0) {           /* very common case */
205                         from += wsize - op;
206                         if (op < len) {         /* some from window */
207                             len -= op;
208                             do {
209                                 *out++ = *from++;
210                             } while (--op);
211                             from = out - dist;  /* rest from output */
212                         }
213                     }
214                     else if (wnext < op) {      /* wrap around window */
215                         from += wsize + wnext - op;
216                         op -= wnext;
217                         if (op < len) {         /* some from end of window */
218                             len -= op;
219                             do {
220                                 *out++ = *from++;
221                             } while (--op);
222                             from = window;
223                             if (wnext < len) {  /* some from start of window */
224                                 op = wnext;
225                                 len -= op;
226                                 do {
227                                     *out++ = *from++;
228                                 } while (--op);
229                                 from = out - dist;      /* rest from output */
230                             }
231                         }
232                     }
233                     else {                      /* contiguous in window */
234                         from += wnext - op;
235                         if (op < len) {         /* some from window */
236                             len -= op;
237                             do {
238                                 *out++ = *from++;
239                             } while (--op);
240                             from = out - dist;  /* rest from output */
241                         }
242                     }
243                     while (len > 2) {
244                         *out++ = *from++;
245                         *out++ = *from++;
246                         *out++ = *from++;
247                         len -= 3;
248                     }
249                     if (len) {
250                         *out++ = *from++;
251                         if (len > 1)
252                             *out++ = *from++;
253                     }
254                 }
255                 else {
256                     from = out - dist;          /* copy direct from output */
257                     do {                        /* minimum length is three */
258                         *out++ = *from++;
259                         *out++ = *from++;
260                         *out++ = *from++;
261                         len -= 3;
262                     } while (len > 2);
263                     if (len) {
264                         *out++ = *from++;
265                         if (len > 1)
266                             *out++ = *from++;
267                     }
268                 }
269             }
270             else if ((op & 64) == 0) {          /* 2nd level distance code */
271                 here = dcode + here->val + (hold & ((1U << op) - 1));
272                 goto dodist;
273             }
274             else {
275 #ifdef SMALL
276                 strm->msg = "error";
277 #else
278                 strm->msg = (char *)"invalid distance code";
279 #endif
280                 state->mode = BAD;
281                 break;
282             }
283         }
284         else if ((op & 64) == 0) {              /* 2nd level length code */
285             here = lcode + here->val + (hold & ((1U << op) - 1));
286             goto dolen;
287         }
288         else if (op & 32) {                     /* end-of-block */
289             Tracevv((stderr, "inflate:         end of block\n"));
290             state->mode = TYPE;
291             break;
292         }
293         else {
294 #ifdef SMALL
295             strm->msg = "error";
296 #else
297             strm->msg = (char *)"invalid literal/length code";
298 #endif
299             state->mode = BAD;
300             break;
301         }
302     } while (in < last && out < end);
303 
304     /* return unused bytes (on entry, bits < 8, so in won't go too far back) */
305     len = bits >> 3;
306     in -= len;
307     bits -= len << 3;
308     hold &= (1U << bits) - 1;
309 
310     /* update state and return */
311     strm->next_in = in;
312     strm->next_out = out;
313     strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last));
314     strm->avail_out = (unsigned)(out < end ?
315                                  257 + (end - out) : 257 - (out - end));
316     state->hold = hold;
317     state->bits = bits;
318     return;
319 }
320 
321 /*
322    inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe):
323    - Using bit fields for code structure
324    - Different op definition to avoid & for extra bits (do & for table bits)
325    - Three separate decoding do-loops for direct, window, and wnext == 0
326    - Special case for distance > 1 copies to do overlapped load and store copy
327    - Explicit branch predictions (based on measured branch probabilities)
328    - Deferring match copy and interspersed it with decoding subsequent codes
329    - Swapping literal/length else
330    - Swapping window/direct else
331    - Larger unrolled copy loops (three is about right)
332    - Moving len -= 3 statement into middle of loop
333  */
334 
335 #endif /* !ASMINF */
336