|
1
|
/* inffast.c -- fast decoding |
|
2
|
* Copyright (C) 1995-2026 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 *) |
|
159
|
"invalid distance too far back"; |
|
160
|
state->mode = BAD; |
|
161
|
break; |
|
162
|
} |
|
163
|
#endif |
|
164
|
hold >>= op; |
|
165
|
bits -= op; |
|
166
|
Tracevv((stderr, "inflate: distance %u\n", dist)); |
|
167
|
op = (unsigned)(out - beg); /* max distance in output */ |
|
168
|
if (dist > op) { /* see if copy from window */ |
|
169
|
op = dist - op; /* distance back in window */ |
|
170
|
if (op > whave) { |
|
171
|
if (state->sane) { |
|
172
|
strm->msg = (z_const char *) |
|
173
|
"invalid distance too far back"; |
|
174
|
state->mode = BAD; |
|
175
|
break; |
|
176
|
} |
|
177
|
#ifdef INFLATE_ALLOW_INVALID_DISTANCE_TOOFAR_ARRR |
|
178
|
if (len <= op - whave) { |
|
179
|
do { |
|
180
|
*out++ = 0; |
|
181
|
} while (--len); |
|
182
|
continue; |
|
183
|
} |
|
184
|
len -= op - whave; |
|
185
|
do { |
|
186
|
*out++ = 0; |
|
187
|
} while (--op > whave); |
|
188
|
if (op == 0) { |
|
189
|
from = out - dist; |
|
190
|
do { |
|
191
|
*out++ = *from++; |
|
192
|
} while (--len); |
|
193
|
continue; |
|
194
|
} |
|
195
|
#endif |
|
196
|
} |
|
197
|
from = window; |
|
198
|
if (wnext == 0) { /* very common case */ |
|
199
|
from += wsize - op; |
|
200
|
if (op < len) { /* some from window */ |
|
201
|
len -= op; |
|
202
|
do { |
|
203
|
*out++ = *from++; |
|
204
|
} while (--op); |
|
205
|
from = out - dist; /* rest from output */ |
|
206
|
} |
|
207
|
} |
|
208
|
else if (wnext < op) { /* wrap around window */ |
|
209
|
from += wsize + wnext - op; |
|
210
|
op -= wnext; |
|
211
|
if (op < len) { /* some from end of window */ |
|
212
|
len -= op; |
|
213
|
do { |
|
214
|
*out++ = *from++; |
|
215
|
} while (--op); |
|
216
|
from = window; |
|
217
|
if (wnext < len) { /* some from start of window */ |
|
218
|
op = wnext; |
|
219
|
len -= op; |
|
220
|
do { |
|
221
|
*out++ = *from++; |
|
222
|
} while (--op); |
|
223
|
from = out - dist; /* rest from output */ |
|
224
|
} |
|
225
|
} |
|
226
|
} |
|
227
|
else { /* contiguous in window */ |
|
228
|
from += wnext - op; |
|
229
|
if (op < len) { /* some from window */ |
|
230
|
len -= op; |
|
231
|
do { |
|
232
|
*out++ = *from++; |
|
233
|
} while (--op); |
|
234
|
from = out - dist; /* rest from output */ |
|
235
|
} |
|
236
|
} |
|
237
|
while (len > 2) { |
|
238
|
*out++ = *from++; |
|
239
|
*out++ = *from++; |
|
240
|
*out++ = *from++; |
|
241
|
len -= 3; |
|
242
|
} |
|
243
|
if (len) { |
|
244
|
*out++ = *from++; |
|
245
|
if (len > 1) |
|
246
|
*out++ = *from++; |
|
247
|
} |
|
248
|
} |
|
249
|
else { |
|
250
|
from = out - dist; /* copy direct from output */ |
|
251
|
do { /* minimum length is three */ |
|
252
|
*out++ = *from++; |
|
253
|
*out++ = *from++; |
|
254
|
*out++ = *from++; |
|
255
|
len -= 3; |
|
256
|
} while (len > 2); |
|
257
|
if (len) { |
|
258
|
*out++ = *from++; |
|
259
|
if (len > 1) |
|
260
|
*out++ = *from++; |
|
261
|
} |
|
262
|
} |
|
263
|
} |
|
264
|
else if ((op & 64) == 0) { /* 2nd level distance code */ |
|
265
|
here = dcode + here->val + (hold & ((1U << op) - 1)); |
|
266
|
goto dodist; |
|
267
|
} |
|
268
|
else { |
|
269
|
strm->msg = (z_const char *)"invalid distance code"; |
|
270
|
state->mode = BAD; |
|
271
|
break; |
|
272
|
} |
|
273
|
} |
|
274
|
else if ((op & 64) == 0) { /* 2nd level length code */ |
|
275
|
here = lcode + here->val + (hold & ((1U << op) - 1)); |
|
276
|
goto dolen; |
|
277
|
} |
|
278
|
else if (op & 32) { /* end-of-block */ |
|
279
|
Tracevv((stderr, "inflate: end of block\n")); |
|
280
|
state->mode = TYPE; |
|
281
|
break; |
|
282
|
} |
|
283
|
else { |
|
284
|
strm->msg = (z_const char *)"invalid literal/length code"; |
|
285
|
state->mode = BAD; |
|
286
|
break; |
|
287
|
} |
|
288
|
} while (in < last && out < end); |
|
289
|
|
|
290
|
/* return unused bytes (on entry, bits < 8, so in won't go too far back) */ |
|
291
|
len = bits >> 3; |
|
292
|
in -= len; |
|
293
|
bits -= len << 3; |
|
294
|
hold &= (1U << bits) - 1; |
|
295
|
|
|
296
|
/* update state and return */ |
|
297
|
strm->next_in = in; |
|
298
|
strm->next_out = out; |
|
299
|
strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last)); |
|
300
|
strm->avail_out = (unsigned)(out < end ? |
|
301
|
257 + (end - out) : 257 - (out - end)); |
|
302
|
state->hold = hold; |
|
303
|
state->bits = bits; |
|
304
|
return; |
|
305
|
} |
|
306
|
|
|
307
|
/* |
|
308
|
inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe): |
|
309
|
- Using bit fields for code structure |
|
310
|
- Different op definition to avoid & for extra bits (do & for table bits) |
|
311
|
- Three separate decoding do-loops for direct, window, and wnext == 0 |
|
312
|
- Special case for distance > 1 copies to do overlapped load and store copy |
|
313
|
- Explicit branch predictions (based on measured branch probabilities) |
|
314
|
- Deferring match copy and interspersed it with decoding subsequent codes |
|
315
|
- Swapping literal/length else |
|
316
|
- Swapping window/direct else |
|
317
|
- Larger unrolled copy loops (three is about right) |
|
318
|
- Moving len -= 3 statement into middle of loop |
|
319
|
*/ |
|
320
|
|
|
321
|
#endif /* !ASMINF */ |
|
322
|
|