1 /* inffast.c -- fast decoding
2 * Copyright (C) 1995-2004 Mark Adler
3 * For conditions of distribution and use, see copyright notice in zlib.h
6 #pragma ident "%Z%%M% %I% %E% SMI"
15 /* Allow machine dependent optimization for post-increment or pre-increment.
16 Based on testing to date,
17 Pre-increment preferred for:
19 - MIPS R5000 (Randers-Pehrson)
20 Post-increment preferred for:
22 No measurable difference:
23 - Pentium III (Anderson)
28 # define PUP(a) *(a)++
31 # define PUP(a) *++(a)
35 Decode literal, length, and distance codes and write out the resulting
36 literal and match bytes until either not enough input or output is
37 available, an end-of-block is encountered, or a data error is encountered.
38 When large enough input and output buffers are supplied to inflate(), for
39 example, a 16K input buffer and a 64K output buffer, more than 95% of the
40 inflate execution time is spent in this routine.
46 strm->avail_out >= 258
47 start >= strm->avail_out
50 On return, state->mode is one of:
52 LEN -- ran out of enough output space or enough available input
53 TYPE -- reached end of block code, inflate() to interpret next block
54 BAD -- error in block data
58 - The maximum input bits used by a length/distance pair is 15 bits for the
59 length code, 5 bits for the length extra, 15 bits for the distance code,
60 and 13 bits for the distance extra. This totals 48 bits, or six bytes.
61 Therefore if strm->avail_in >= 6, then there is enough input to avoid
62 checking for available input while decoding.
64 - The maximum bytes that a single length/distance pair can output is 258
65 bytes, which is the maximum length that can be coded. inflate_fast()
66 requires strm->avail_out >= 258 for each loop to avoid checking for
69 void inflate_fast(strm, start)
71 unsigned start; /* inflate()'s starting value for strm->avail_out */
73 struct inflate_state FAR *state;
74 unsigned char FAR *in; /* local strm->next_in */
75 unsigned char FAR *last; /* while in < last, enough input available */
76 unsigned char FAR *out; /* local strm->next_out */
77 unsigned char FAR *beg; /* inflate()'s initial strm->next_out */
78 unsigned char FAR *end; /* while out < end, enough space available */
80 unsigned dmax; /* maximum distance from zlib header */
82 unsigned wsize; /* window size or zero if not using window */
83 unsigned whave; /* valid bytes in the window */
84 unsigned write; /* window write index */
85 unsigned char FAR *window; /* allocated sliding window, if wsize != 0 */
86 unsigned long hold; /* local strm->hold */
87 unsigned bits; /* local strm->bits */
88 code const FAR *lcode; /* local strm->lencode */
89 code const FAR *dcode; /* local strm->distcode */
90 unsigned lmask; /* mask for first level of length codes */
91 unsigned dmask; /* mask for first level of distance codes */
92 code this; /* retrieved table entry */
93 unsigned op; /* code bits, operation, extra bits, or */
94 /* window position, window bytes to copy */
95 unsigned len; /* match length, unused bytes */
96 unsigned dist; /* match distance */
97 unsigned char FAR *from; /* where to copy match from */
99 /* copy state to local variables */
100 state = (struct inflate_state FAR *)strm->state;
101 in = strm->next_in - OFF;
102 last = in + (strm->avail_in - 5);
103 out = strm->next_out - OFF;
104 beg = out - (start - strm->avail_out);
105 end = out + (strm->avail_out - 257);
106 #ifdef INFLATE_STRICT
109 wsize = state->wsize;
110 whave = state->whave;
111 write = state->write;
112 window = state->window;
115 lcode = state->lencode;
116 dcode = state->distcode;
117 lmask = (1U << state->lenbits) - 1;
118 dmask = (1U << state->distbits) - 1;
120 /* decode literals and length/distances until end-of-block or not enough
121 input data or output space */
124 hold += (unsigned long)(PUP(in)) << bits;
126 hold += (unsigned long)(PUP(in)) << bits;
129 this = lcode[hold & lmask];
131 op = (unsigned)(this.bits);
134 op = (unsigned)(this.op);
135 if (op == 0) { /* literal */
136 Tracevv((stderr, this.val >= 0x20 && this.val < 0x7f ?
137 "inflate: literal '%c'\n" :
138 "inflate: literal 0x%02x\n", this.val));
139 PUP(out) = (unsigned char)(this.val);
141 else if (op & 16) { /* length base */
142 len = (unsigned)(this.val);
143 op &= 15; /* number of extra bits */
146 hold += (unsigned long)(PUP(in)) << bits;
149 len += (unsigned)hold & ((1U << op) - 1);
153 Tracevv((stderr, "inflate: length %u\n", len));
155 hold += (unsigned long)(PUP(in)) << bits;
157 hold += (unsigned long)(PUP(in)) << bits;
160 this = dcode[hold & dmask];
162 op = (unsigned)(this.bits);
165 op = (unsigned)(this.op);
166 if (op & 16) { /* distance base */
167 dist = (unsigned)(this.val);
168 op &= 15; /* number of extra bits */
170 hold += (unsigned long)(PUP(in)) << bits;
173 hold += (unsigned long)(PUP(in)) << bits;
177 dist += (unsigned)hold & ((1U << op) - 1);
178 #ifdef INFLATE_STRICT
180 strm->msg = (char *)"invalid distance too far back";
187 Tracevv((stderr, "inflate: distance %u\n", dist));
188 op = (unsigned)(out - beg); /* max distance in output */
189 if (dist > op) { /* see if copy from window */
190 op = dist - op; /* distance back in window */
192 strm->msg = (char *)"invalid distance too far back";
197 if (write == 0) { /* very common case */
199 if (op < len) { /* some from window */
202 PUP(out) = PUP(from);
204 from = out - dist; /* rest from output */
207 else if (write < op) { /* wrap around window */
208 from += wsize + write - op;
210 if (op < len) { /* some from end of window */
213 PUP(out) = PUP(from);
216 if (write < len) { /* some from start of window */
220 PUP(out) = PUP(from);
222 from = out - dist; /* rest from output */
226 else { /* contiguous in window */
228 if (op < len) { /* some from window */
231 PUP(out) = PUP(from);
233 from = out - dist; /* rest from output */
237 PUP(out) = PUP(from);
238 PUP(out) = PUP(from);
239 PUP(out) = PUP(from);
243 PUP(out) = PUP(from);
245 PUP(out) = PUP(from);
249 from = out - dist; /* copy direct from output */
250 do { /* minimum length is three */
251 PUP(out) = PUP(from);
252 PUP(out) = PUP(from);
253 PUP(out) = PUP(from);
257 PUP(out) = PUP(from);
259 PUP(out) = PUP(from);
263 else if ((op & 64) == 0) { /* 2nd level distance code */
264 this = dcode[this.val + (hold & ((1U << op) - 1))];
268 strm->msg = (char *)"invalid distance code";
273 else if ((op & 64) == 0) { /* 2nd level length code */
274 this = lcode[this.val + (hold & ((1U << op) - 1))];
277 else if (op & 32) { /* end-of-block */
278 Tracevv((stderr, "inflate: end of block\n"));
283 strm->msg = (char *)"invalid literal/length code";
287 } while (in < last && out < end);
289 /* return unused bytes (on entry, bits < 8, so in won't go too far back) */
293 hold &= (1U << bits) - 1;
295 /* update state and return */
296 strm->next_in = in + OFF;
297 strm->next_out = out + OFF;
298 strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last));
299 strm->avail_out = (unsigned)(out < end ?
300 257 + (end - out) : 257 - (out - end));
307 inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe):
308 - Using bit fields for code structure
309 - Different op definition to avoid & for extra bits (do & for table bits)
310 - Three separate decoding do-loops for direct, window, and write == 0
311 - Special case for distance > 1 copies to do overlapped load and store copy
312 - Explicit branch predictions (based on measured branch probabilities)
313 - Deferring match copy and interspersed it with decoding subsequent codes
314 - Swapping literal/length else
315 - Swapping window/direct else
316 - Larger unrolled copy loops (three is about right)
317 - Moving len -= 3 statement into middle of loop