1 /* inffast.cpp -- fast decoding |
|
2 * Copyright (C) 1995-2004 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 #ifndef ASMINF |
|
12 |
|
13 /* Allow machine dependent optimization for post-increment or pre-increment. |
|
14 Based on testing to date, |
|
15 Pre-increment preferred for: |
|
16 - PowerPC G3 (Adler) |
|
17 - MIPS R5000 (Randers-Pehrson) |
|
18 Post-increment preferred for: |
|
19 - none |
|
20 No measurable difference: |
|
21 - Pentium III (Anderson) |
|
22 - M68060 (Nikl) |
|
23 */ |
|
24 #ifdef POSTINC |
|
25 # define OFF 0 |
|
26 # define PUP(a) *(a)++ |
|
27 #else |
|
28 # define OFF 1 |
|
29 # define PUP(a) *++(a) |
|
30 #endif |
|
31 |
|
32 /* |
|
33 Decode literal, length, and distance codes and write out the resulting |
|
34 literal and match bytes until either not enough input or output is |
|
35 available, an end-of-block is encountered, or a data error is encountered. |
|
36 When large enough input and output buffers are supplied to inflate(), for |
|
37 example, a 16K input buffer and a 64K output buffer, more than 95% of the |
|
38 inflate execution time is spent in this routine. |
|
39 |
|
40 Entry assumptions: |
|
41 |
|
42 state->mode == LEN |
|
43 strm->avail_in >= 6 |
|
44 strm->avail_out >= 258 |
|
45 start >= strm->avail_out |
|
46 state->bits < 8 |
|
47 |
|
48 On return, state->mode is one of: |
|
49 |
|
50 LEN -- ran out of enough output space or enough available input |
|
51 TYPE -- reached end of block code, inflate() to interpret next block |
|
52 BAD -- error in block data |
|
53 |
|
54 Notes: |
|
55 |
|
56 - The maximum input bits used by a length/distance pair is 15 bits for the |
|
57 length code, 5 bits for the length extra, 15 bits for the distance code, |
|
58 and 13 bits for the distance extra. This totals 48 bits, or six bytes. |
|
59 Therefore if strm->avail_in >= 6, then there is enough input to avoid |
|
60 checking for available input while decoding. |
|
61 |
|
62 - The maximum bytes that a single length/distance pair can output is 258 |
|
63 bytes, which is the maximum length that can be coded. inflate_fast() |
|
64 requires strm->avail_out >= 258 for each loop to avoid checking for |
|
65 output space. |
|
66 */ |
|
67 #ifdef __SYMBIAN32__ |
|
68 void inflate_fast(z_streamp strm,unsigned start) |
|
69 #else |
|
70 void inflate_fast(strm, start) |
|
71 z_streamp strm; |
|
72 unsigned start; /* inflate()'s starting value for strm->avail_out */ |
|
73 #endif //__SYMBIAN32__ |
|
74 { |
|
75 struct inflate_state FAR *state; |
|
76 unsigned char FAR *in; /* local strm->next_in */ |
|
77 unsigned char FAR *last; /* while in < last, enough input available */ |
|
78 unsigned char FAR *out; /* local strm->next_out */ |
|
79 unsigned char FAR *beg; /* inflate()'s initial strm->next_out */ |
|
80 unsigned char FAR *end; /* while out < end, enough space available */ |
|
81 #ifdef INFLATE_STRICT |
|
82 unsigned dmax; /* maximum distance from zlib header */ |
|
83 #endif |
|
84 unsigned wsize; /* window size or zero if not using window */ |
|
85 unsigned whave; /* valid bytes in the window */ |
|
86 unsigned write; /* window write index */ |
|
87 unsigned char FAR *window; /* allocated sliding window, if wsize != 0 */ |
|
88 unsigned long hold; /* local strm->hold */ |
|
89 unsigned bits; /* local strm->bits */ |
|
90 code const FAR *lcode; /* local strm->lencode */ |
|
91 code const FAR *dcode; /* local strm->distcode */ |
|
92 unsigned lmask; /* mask for first level of length codes */ |
|
93 unsigned dmask; /* mask for first level of distance codes */ |
|
94 /* Need to replace "this" variable with "current" as "this" is a reserved |
|
95 * keyword in C++ which is prefectly fine for a c code. As this file |
|
96 * has been changed to C++ "this" needs to be changed. |
|
97 */ |
|
98 # define this current |
|
99 code this; /* retrieved table entry */ |
|
100 unsigned op; /* code bits, operation, extra bits, or */ |
|
101 /* window position, window bytes to copy */ |
|
102 unsigned len; /* match length, unused bytes */ |
|
103 unsigned dist; /* match distance */ |
|
104 unsigned char FAR *from; /* where to copy match from */ |
|
105 |
|
106 /* copy state to local variables */ |
|
107 state = (struct inflate_state FAR *)strm->state; |
|
108 in = strm->next_in - OFF; |
|
109 last = in + (strm->avail_in - 5); |
|
110 out = strm->next_out - OFF; |
|
111 beg = out - (start - strm->avail_out); |
|
112 end = out + (strm->avail_out - 257); |
|
113 #ifdef INFLATE_STRICT |
|
114 dmax = state->dmax; |
|
115 #endif |
|
116 wsize = state->wsize; |
|
117 whave = state->whave; |
|
118 write = state->write; |
|
119 window = state->window; |
|
120 hold = state->hold; |
|
121 bits = state->bits; |
|
122 lcode = state->lencode; |
|
123 dcode = state->distcode; |
|
124 lmask = (1U << state->lenbits) - 1; |
|
125 dmask = (1U << state->distbits) - 1; |
|
126 |
|
127 /* decode literals and length/distances until end-of-block or not enough |
|
128 input data or output space */ |
|
129 do { |
|
130 if (bits < 15) { |
|
131 hold += (unsigned long)(PUP(in)) << bits; |
|
132 bits += 8; |
|
133 hold += (unsigned long)(PUP(in)) << bits; |
|
134 bits += 8; |
|
135 } |
|
136 this = lcode[hold & lmask]; |
|
137 dolen: |
|
138 op = (unsigned)(this.bits); |
|
139 hold >>= op; |
|
140 bits -= op; |
|
141 op = (unsigned)(this.op); |
|
142 if (op == 0) { /* literal */ |
|
143 Tracevv((stderr, this.val >= 0x20 && this.val < 0x7f ? |
|
144 "inflate: literal '%c'\n" : |
|
145 "inflate: literal 0x%02x\n", this.val)); |
|
146 PUP(out) = (unsigned char)(this.val); |
|
147 } |
|
148 else if (op & 16) { /* length base */ |
|
149 len = (unsigned)(this.val); |
|
150 op &= 15; /* number of extra bits */ |
|
151 if (op) { |
|
152 if (bits < op) { |
|
153 hold += (unsigned long)(PUP(in)) << bits; |
|
154 bits += 8; |
|
155 } |
|
156 len += (unsigned)hold & ((1U << op) - 1); |
|
157 hold >>= op; |
|
158 bits -= op; |
|
159 } |
|
160 Tracevv((stderr, "inflate: length %u\n", len)); |
|
161 if (bits < 15) { |
|
162 hold += (unsigned long)(PUP(in)) << bits; |
|
163 bits += 8; |
|
164 hold += (unsigned long)(PUP(in)) << bits; |
|
165 bits += 8; |
|
166 } |
|
167 this = dcode[hold & dmask]; |
|
168 dodist: |
|
169 op = (unsigned)(this.bits); |
|
170 hold >>= op; |
|
171 bits -= op; |
|
172 op = (unsigned)(this.op); |
|
173 if (op & 16) { /* distance base */ |
|
174 dist = (unsigned)(this.val); |
|
175 op &= 15; /* number of extra bits */ |
|
176 if (bits < op) { |
|
177 hold += (unsigned long)(PUP(in)) << bits; |
|
178 bits += 8; |
|
179 if (bits < op) { |
|
180 hold += (unsigned long)(PUP(in)) << bits; |
|
181 bits += 8; |
|
182 } |
|
183 } |
|
184 dist += (unsigned)hold & ((1U << op) - 1); |
|
185 #ifdef INFLATE_STRICT |
|
186 if (dist > dmax) { |
|
187 strm->msg = (char *)"invalid distance too far back"; |
|
188 state->mode = BAD; |
|
189 break; |
|
190 } |
|
191 #endif |
|
192 hold >>= op; |
|
193 bits -= op; |
|
194 Tracevv((stderr, "inflate: distance %u\n", dist)); |
|
195 op = (unsigned)(out - beg); /* max distance in output */ |
|
196 if (dist > op) { /* see if copy from window */ |
|
197 op = dist - op; /* distance back in window */ |
|
198 if (op > whave) { |
|
199 strm->msg = (char *)"invalid distance too far back"; |
|
200 state->mode = BAD; |
|
201 break; |
|
202 } |
|
203 from = window - OFF; |
|
204 if (write == 0) { /* very common case */ |
|
205 from += wsize - op; |
|
206 if (op < len) { /* some from window */ |
|
207 len -= op; |
|
208 do { |
|
209 PUP(out) = PUP(from); |
|
210 } while (--op); |
|
211 from = out - dist; /* rest from output */ |
|
212 } |
|
213 } |
|
214 else if (write < op) { /* wrap around window */ |
|
215 from += wsize + write - op; |
|
216 op -= write; |
|
217 if (op < len) { /* some from end of window */ |
|
218 len -= op; |
|
219 do { |
|
220 PUP(out) = PUP(from); |
|
221 } while (--op); |
|
222 from = window - OFF; |
|
223 if (write < len) { /* some from start of window */ |
|
224 op = write; |
|
225 len -= op; |
|
226 do { |
|
227 PUP(out) = PUP(from); |
|
228 } while (--op); |
|
229 from = out - dist; /* rest from output */ |
|
230 } |
|
231 } |
|
232 } |
|
233 else { /* contiguous in window */ |
|
234 from += write - op; |
|
235 if (op < len) { /* some from window */ |
|
236 len -= op; |
|
237 do { |
|
238 PUP(out) = PUP(from); |
|
239 } while (--op); |
|
240 from = out - dist; /* rest from output */ |
|
241 } |
|
242 } |
|
243 while (len > 2) { |
|
244 PUP(out) = PUP(from); |
|
245 PUP(out) = PUP(from); |
|
246 PUP(out) = PUP(from); |
|
247 len -= 3; |
|
248 } |
|
249 if (len) { |
|
250 PUP(out) = PUP(from); |
|
251 if (len > 1) |
|
252 PUP(out) = PUP(from); |
|
253 } |
|
254 } |
|
255 else { |
|
256 from = out - dist; /* copy direct from output */ |
|
257 do { /* minimum length is three */ |
|
258 PUP(out) = PUP(from); |
|
259 PUP(out) = PUP(from); |
|
260 PUP(out) = PUP(from); |
|
261 len -= 3; |
|
262 } while (len > 2); |
|
263 if (len) { |
|
264 PUP(out) = PUP(from); |
|
265 if (len > 1) |
|
266 PUP(out) = PUP(from); |
|
267 } |
|
268 } |
|
269 } |
|
270 else if ((op & 64) == 0) { /* 2nd level distance code */ |
|
271 this = dcode[this.val + (hold & ((1U << op) - 1))]; |
|
272 goto dodist; |
|
273 } |
|
274 else { |
|
275 strm->msg = (char *)"invalid distance code"; |
|
276 state->mode = BAD; |
|
277 break; |
|
278 } |
|
279 } |
|
280 else if ((op & 64) == 0) { /* 2nd level length code */ |
|
281 this = lcode[this.val + (hold & ((1U << op) - 1))]; |
|
282 goto dolen; |
|
283 } |
|
284 else if (op & 32) { /* end-of-block */ |
|
285 Tracevv((stderr, "inflate: end of block\n")); |
|
286 state->mode = TYPE; |
|
287 break; |
|
288 } |
|
289 else { |
|
290 strm->msg = (char *)"invalid literal/length code"; |
|
291 state->mode = BAD; |
|
292 break; |
|
293 } |
|
294 } while (in < last && out < end); |
|
295 |
|
296 /* return unused bytes (on entry, bits < 8, so in won't go too far back) */ |
|
297 len = bits >> 3; |
|
298 in -= len; |
|
299 bits -= len << 3; |
|
300 hold &= (1U << bits) - 1; |
|
301 |
|
302 /* update state and return */ |
|
303 strm->next_in = in + OFF; |
|
304 strm->next_out = out + OFF; |
|
305 strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last)); |
|
306 strm->avail_out = (unsigned)(out < end ? |
|
307 257 + (end - out) : 257 - (out - end)); |
|
308 state->hold = hold; |
|
309 state->bits = bits; |
|
310 return; |
|
311 } |
|
312 |
|
313 /* |
|
314 inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe): |
|
315 - Using bit fields for code structure |
|
316 - Different op definition to avoid & for extra bits (do & for table bits) |
|
317 - Three separate decoding do-loops for direct, window, and write == 0 |
|
318 - Special case for distance > 1 copies to do overlapped load and store copy |
|
319 - Explicit branch predictions (based on measured branch probabilities) |
|
320 - Deferring match copy and interspersed it with decoding subsequent codes |
|
321 - Swapping literal/length else |
|
322 - Swapping window/direct else |
|
323 - Larger unrolled copy loops (three is about right) |
|
324 - Moving len -= 3 statement into middle of loop |
|
325 */ |
|
326 |
|
327 #endif /* !ASMINF */ |
|
328 |
|
329 |
|
330 |
|
331 |
|
332 |
|