]>
Commit | Line | Data |
---|---|---|
87a2b23a | 1 | /* inffast.c -- fast decoding |
239eb531 | 2 | * Copyright (C) 1995-2017 Mark Adler |
87a2b23a | 3 | * For conditions of distribution and use, see copyright notice in zlib.h |
d8b6dda4 | 4 | */ |
5 | ||
6 | #include "zutil.h" | |
7 | #include "inftrees.h" | |
87a2b23a | 8 | #include "inflate.h" |
d8b6dda4 | 9 | #include "inffast.h" |
10 | ||
1e5dce21 | 11 | #ifdef ASMINF |
12 | # pragma message("Assembler code may have bugs -- use at your own risk") | |
87a2b23a | 13 | #else |
87a2b23a | 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 | |
d8b6dda4 | 36 | |
87a2b23a | 37 | Notes: |
d8b6dda4 | 38 | |
87a2b23a | 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. | |
d8b6dda4 | 44 | |
87a2b23a | 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 | */ | |
1f54c5b6 | 50 | void ZLIB_INTERNAL inflate_fast(strm, start) |
87a2b23a | 51 | z_streamp strm; |
52 | unsigned start; /* inflate()'s starting value for strm->avail_out */ | |
d8b6dda4 | 53 | { |
87a2b23a | 54 | struct inflate_state FAR *state; |
4b57a6ef | 55 | z_const unsigned char FAR *in; /* local strm->next_in */ |
56 | z_const unsigned char FAR *last; /* have enough input while in < last */ | |
87a2b23a | 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 */ | |
d919a5db | 60 | #ifdef INFLATE_STRICT |
61 | unsigned dmax; /* maximum distance from zlib header */ | |
62 | #endif | |
87a2b23a | 63 | unsigned wsize; /* window size or zero if not using window */ |
64 | unsigned whave; /* valid bytes in the window */ | |
1f54c5b6 | 65 | unsigned wnext; /* window write index */ |
87a2b23a | 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 */ | |
1f54c5b6 | 73 | code here; /* retrieved table entry */ |
87a2b23a | 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; | |
1e5dce21 | 82 | in = strm->next_in; |
87a2b23a | 83 | last = in + (strm->avail_in - 5); |
1e5dce21 | 84 | out = strm->next_out; |
87a2b23a | 85 | beg = out - (start - strm->avail_out); |
86 | end = out + (strm->avail_out - 257); | |
d919a5db | 87 | #ifdef INFLATE_STRICT |
88 | dmax = state->dmax; | |
89 | #endif | |
87a2b23a | 90 | wsize = state->wsize; |
91 | whave = state->whave; | |
1f54c5b6 | 92 | wnext = state->wnext; |
87a2b23a | 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 */ | |
d8b6dda4 | 103 | do { |
87a2b23a | 104 | if (bits < 15) { |
1e5dce21 | 105 | hold += (unsigned long)(*in++) << bits; |
87a2b23a | 106 | bits += 8; |
1e5dce21 | 107 | hold += (unsigned long)(*in++) << bits; |
87a2b23a | 108 | bits += 8; |
109 | } | |
1f54c5b6 | 110 | here = lcode[hold & lmask]; |
87a2b23a | 111 | dolen: |
1f54c5b6 | 112 | op = (unsigned)(here.bits); |
87a2b23a | 113 | hold >>= op; |
114 | bits -= op; | |
1f54c5b6 | 115 | op = (unsigned)(here.op); |
87a2b23a | 116 | if (op == 0) { /* literal */ |
1f54c5b6 | 117 | Tracevv((stderr, here.val >= 0x20 && here.val < 0x7f ? |
87a2b23a | 118 | "inflate: literal '%c'\n" : |
1f54c5b6 | 119 | "inflate: literal 0x%02x\n", here.val)); |
1e5dce21 | 120 | *out++ = (unsigned char)(here.val); |
87a2b23a | 121 | } |
122 | else if (op & 16) { /* length base */ | |
1f54c5b6 | 123 | len = (unsigned)(here.val); |
87a2b23a | 124 | op &= 15; /* number of extra bits */ |
125 | if (op) { | |
126 | if (bits < op) { | |
1e5dce21 | 127 | hold += (unsigned long)(*in++) << bits; |
87a2b23a | 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) { | |
1e5dce21 | 136 | hold += (unsigned long)(*in++) << bits; |
87a2b23a | 137 | bits += 8; |
1e5dce21 | 138 | hold += (unsigned long)(*in++) << bits; |
87a2b23a | 139 | bits += 8; |
8c16c8c7 | 140 | } |
1f54c5b6 | 141 | here = dcode[hold & dmask]; |
87a2b23a | 142 | dodist: |
1f54c5b6 | 143 | op = (unsigned)(here.bits); |
87a2b23a | 144 | hold >>= op; |
145 | bits -= op; | |
1f54c5b6 | 146 | op = (unsigned)(here.op); |
87a2b23a | 147 | if (op & 16) { /* distance base */ |
1f54c5b6 | 148 | dist = (unsigned)(here.val); |
87a2b23a | 149 | op &= 15; /* number of extra bits */ |
150 | if (bits < op) { | |
1e5dce21 | 151 | hold += (unsigned long)(*in++) << bits; |
87a2b23a | 152 | bits += 8; |
153 | if (bits < op) { | |
1e5dce21 | 154 | hold += (unsigned long)(*in++) << bits; |
87a2b23a | 155 | bits += 8; |
156 | } | |
157 | } | |
158 | dist += (unsigned)hold & ((1U << op) - 1); | |
d919a5db | 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 | |
87a2b23a | 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) { | |
1f54c5b6 | 173 | if (state->sane) { |
174 | strm->msg = | |
175 | (char *)"invalid distance too far back"; | |
176 | state->mode = BAD; | |
177 | break; | |
178 | } | |
179 | #ifdef INFLATE_ALLOW_INVALID_DISTANCE_TOOFAR_ARRR | |
180 | if (len <= op - whave) { | |
181 | do { | |
1e5dce21 | 182 | *out++ = 0; |
1f54c5b6 | 183 | } while (--len); |
184 | continue; | |
185 | } | |
186 | len -= op - whave; | |
187 | do { | |
1e5dce21 | 188 | *out++ = 0; |
1f54c5b6 | 189 | } while (--op > whave); |
190 | if (op == 0) { | |
191 | from = out - dist; | |
192 | do { | |
1e5dce21 | 193 | *out++ = *from++; |
1f54c5b6 | 194 | } while (--len); |
195 | continue; | |
196 | } | |
197 | #endif | |
87a2b23a | 198 | } |
1e5dce21 | 199 | from = window; |
1f54c5b6 | 200 | if (wnext == 0) { /* very common case */ |
87a2b23a | 201 | from += wsize - op; |
202 | if (op < len) { /* some from window */ | |
203 | len -= op; | |
204 | do { | |
1e5dce21 | 205 | *out++ = *from++; |
87a2b23a | 206 | } while (--op); |
207 | from = out - dist; /* rest from output */ | |
208 | } | |
209 | } | |
1f54c5b6 | 210 | else if (wnext < op) { /* wrap around window */ |
211 | from += wsize + wnext - op; | |
212 | op -= wnext; | |
87a2b23a | 213 | if (op < len) { /* some from end of window */ |
214 | len -= op; | |
215 | do { | |
1e5dce21 | 216 | *out++ = *from++; |
87a2b23a | 217 | } while (--op); |
1e5dce21 | 218 | from = window; |
1f54c5b6 | 219 | if (wnext < len) { /* some from start of window */ |
220 | op = wnext; | |
87a2b23a | 221 | len -= op; |
222 | do { | |
1e5dce21 | 223 | *out++ = *from++; |
87a2b23a | 224 | } while (--op); |
225 | from = out - dist; /* rest from output */ | |
226 | } | |
227 | } | |
228 | } | |
229 | else { /* contiguous in window */ | |
1f54c5b6 | 230 | from += wnext - op; |
87a2b23a | 231 | if (op < len) { /* some from window */ |
232 | len -= op; | |
233 | do { | |
1e5dce21 | 234 | *out++ = *from++; |
87a2b23a | 235 | } while (--op); |
236 | from = out - dist; /* rest from output */ | |
237 | } | |
238 | } | |
239 | while (len > 2) { | |
1e5dce21 | 240 | *out++ = *from++; |
241 | *out++ = *from++; | |
242 | *out++ = *from++; | |
87a2b23a | 243 | len -= 3; |
244 | } | |
245 | if (len) { | |
1e5dce21 | 246 | *out++ = *from++; |
87a2b23a | 247 | if (len > 1) |
1e5dce21 | 248 | *out++ = *from++; |
87a2b23a | 249 | } |
250 | } | |
251 | else { | |
252 | from = out - dist; /* copy direct from output */ | |
253 | do { /* minimum length is three */ | |
1e5dce21 | 254 | *out++ = *from++; |
255 | *out++ = *from++; | |
256 | *out++ = *from++; | |
87a2b23a | 257 | len -= 3; |
258 | } while (len > 2); | |
259 | if (len) { | |
1e5dce21 | 260 | *out++ = *from++; |
87a2b23a | 261 | if (len > 1) |
1e5dce21 | 262 | *out++ = *from++; |
87a2b23a | 263 | } |
264 | } | |
d8b6dda4 | 265 | } |
87a2b23a | 266 | else if ((op & 64) == 0) { /* 2nd level distance code */ |
1f54c5b6 | 267 | here = dcode[here.val + (hold & ((1U << op) - 1))]; |
87a2b23a | 268 | goto dodist; |
269 | } | |
270 | else { | |
271 | strm->msg = (char *)"invalid distance code"; | |
272 | state->mode = BAD; | |
273 | break; | |
274 | } | |
275 | } | |
276 | else if ((op & 64) == 0) { /* 2nd level length code */ | |
1f54c5b6 | 277 | here = lcode[here.val + (hold & ((1U << op) - 1))]; |
87a2b23a | 278 | goto dolen; |
279 | } | |
280 | else if (op & 32) { /* end-of-block */ | |
281 | Tracevv((stderr, "inflate: end of block\n")); | |
282 | state->mode = TYPE; | |
283 | break; | |
284 | } | |
285 | else { | |
286 | strm->msg = (char *)"invalid literal/length code"; | |
287 | state->mode = BAD; | |
d8b6dda4 | 288 | break; |
d8b6dda4 | 289 | } |
87a2b23a | 290 | } while (in < last && out < end); |
291 | ||
292 | /* return unused bytes (on entry, bits < 8, so in won't go too far back) */ | |
293 | len = bits >> 3; | |
294 | in -= len; | |
295 | bits -= len << 3; | |
296 | hold &= (1U << bits) - 1; | |
297 | ||
298 | /* update state and return */ | |
1e5dce21 | 299 | strm->next_in = in; |
300 | strm->next_out = out; | |
87a2b23a | 301 | strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last)); |
302 | strm->avail_out = (unsigned)(out < end ? | |
303 | 257 + (end - out) : 257 - (out - end)); | |
304 | state->hold = hold; | |
305 | state->bits = bits; | |
306 | return; | |
d8b6dda4 | 307 | } |
87a2b23a | 308 | |
309 | /* | |
310 | inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe): | |
311 | - Using bit fields for code structure | |
312 | - Different op definition to avoid & for extra bits (do & for table bits) | |
1f54c5b6 | 313 | - Three separate decoding do-loops for direct, window, and wnext == 0 |
87a2b23a | 314 | - Special case for distance > 1 copies to do overlapped load and store copy |
315 | - Explicit branch predictions (based on measured branch probabilities) | |
316 | - Deferring match copy and interspersed it with decoding subsequent codes | |
317 | - Swapping literal/length else | |
318 | - Swapping window/direct else | |
319 | - Larger unrolled copy loops (three is about right) | |
320 | - Moving len -= 3 statement into middle of loop | |
321 | */ | |
322 | ||
323 | #endif /* !ASMINF */ |