1 /* inffast.c -- fast decoding
2 * Copyright (C) 1995-2003 Mark Adler
3 * For conditions of distribution and use, see copyright notice in zlib.h
13 /* Allow machine dependent optimization for post-increment or pre-increment.
14 Based on testing to date,
15 Pre-increment preferred for:
17 - MIPS R5000 (Randers-Pehrson)
18 Post-increment preferred for:
20 No measurable difference:
21 - Pentium III (Anderson)
25 # define PUP(a) *(a)++
28 # define PUP(a) *++(a)
32 Decode literal, length, and distance codes and write out the resulting
33 literal and match bytes until either not enough input or output is
34 available, an end-of-block is encountered, or a data error is encountered.
35 When large enough input and output buffers are supplied to inflate(), for
36 example, a 16K input buffer and a 64K output buffer, more than 95% of the
37 inflate execution time is spent in this routine.
43 strm->avail_out >= 258
44 start >= strm->avail_out
47 On return, state->mode is one of:
49 LEN -- ran out of enough output space or enough available input
50 TYPE -- reached end of block code, inflate() to interpret next block
51 BAD -- error in block data
55 - The maximum input bits used by a length/distance pair is 15 bits for the
56 length code, 5 bits for the length extra, 15 bits for the distance code,
57 and 13 bits for the distance extra. This totals 48 bits, or six bytes.
58 Therefore if strm->avail_in >= 6, then there is enough input to avoid
59 checking for available input while decoding.
61 - The maximum bytes that a single length/distance pair can output is 258
62 bytes, which is the maximum length that can be coded. inflate_fast()
63 requires strm->avail_out >= 258 for each loop to avoid checking for
66 void inflate_fast(strm, start)
68 unsigned start; /* inflate()'s starting value for strm->avail_out */
70 struct inflate_state FAR *state;
71 unsigned char FAR *in; /* local strm->next_in */
72 unsigned char FAR *last; /* while in < last, enough input available */
73 unsigned char FAR *out; /* local strm->next_out */
74 unsigned char FAR *beg; /* inflate()'s initial strm->next_out */
75 unsigned char FAR *end; /* while out < end, enough space available */
76 unsigned wsize; /* window size or zero if not using window */
77 unsigned whave; /* valid bytes in the window */
78 unsigned write; /* window write index */
79 unsigned char FAR *window; /* allocated sliding window, if wsize != 0 */
80 unsigned long hold; /* local strm->hold */
81 unsigned bits; /* local strm->bits */
82 code const FAR *lcode; /* local strm->lencode */
83 code const FAR *dcode; /* local strm->distcode */
84 unsigned lmask; /* mask for first level of length codes */
85 unsigned dmask; /* mask for first level of distance codes */
86 code this; /* retrieved table entry */
87 unsigned op; /* code bits, operation, extra bits, or */
88 /* window position, window bytes to copy */
89 unsigned len; /* match length, unused bytes */
90 unsigned dist; /* match distance */
91 unsigned char FAR *from; /* where to copy match from */
93 /* copy state to local variables */
94 state = (struct inflate_state FAR *)strm->state;
95 in = strm->next_in - OFF;
96 last = in + (strm->avail_in - 5);
97 out = strm->next_out - OFF;
98 beg = out - (start - strm->avail_out);
99 end = out + (strm->avail_out - 257);
100 wsize = state->wsize;
101 whave = state->whave;
102 write = state->write;
103 window = state->window;
106 lcode = state->lencode;
107 dcode = state->distcode;
108 lmask = (1U << state->lenbits) - 1;
109 dmask = (1U << state->distbits) - 1;
111 /* decode literals and length/distances until end-of-block or not enough
112 input data or output space */
115 hold += (unsigned long)(PUP(in)) << bits;
117 hold += (unsigned long)(PUP(in)) << bits;
120 this = lcode[hold & lmask];
122 op = (unsigned)(this.bits);
125 op = (unsigned)(this.op);
126 if (op == 0) { /* literal */
127 Tracevv((stderr, this.val >= 0x20 && this.val < 0x7f ?
128 "inflate: literal '%c'\n" :
129 "inflate: literal 0x%02x\n", this.val));
130 PUP(out) = (unsigned char)(this.val);
132 else if (op & 16) { /* length base */
133 len = (unsigned)(this.val);
134 op &= 15; /* number of extra bits */
137 hold += (unsigned long)(PUP(in)) << bits;
140 len += (unsigned)hold & ((1U << op) - 1);
144 Tracevv((stderr, "inflate: length %u\n", len));
146 hold += (unsigned long)(PUP(in)) << bits;
148 hold += (unsigned long)(PUP(in)) << bits;
151 this = dcode[hold & dmask];
153 op = (unsigned)(this.bits);
156 op = (unsigned)(this.op);
157 if (op & 16) { /* distance base */
158 dist = (unsigned)(this.val);
159 op &= 15; /* number of extra bits */
161 hold += (unsigned long)(PUP(in)) << bits;
164 hold += (unsigned long)(PUP(in)) << bits;
168 dist += (unsigned)hold & ((1U << op) - 1);
171 Tracevv((stderr, "inflate: distance %u\n", dist));
172 op = (unsigned)(out - beg); /* max distance in output */
173 if (dist > op) { /* see if copy from window */
174 op = dist - op; /* distance back in window */
176 strm->msg = (char *)"invalid distance too far back";
181 if (write == 0) { /* very common case */
183 if (op < len) { /* some from window */
186 PUP(out) = PUP(from);
188 from = out - dist; /* rest from output */
191 else if (write < op) { /* wrap around window */
192 from += wsize + write - op;
194 if (op < len) { /* some from end of window */
197 PUP(out) = PUP(from);
200 if (write < len) { /* some from start of window */
204 PUP(out) = PUP(from);
206 from = out - dist; /* rest from output */
210 else { /* contiguous in window */
212 if (op < len) { /* some from window */
215 PUP(out) = PUP(from);
217 from = out - dist; /* rest from output */
221 PUP(out) = PUP(from);
222 PUP(out) = PUP(from);
223 PUP(out) = PUP(from);
227 PUP(out) = PUP(from);
229 PUP(out) = PUP(from);
233 from = out - dist; /* copy direct from output */
234 do { /* minimum length is three */
235 PUP(out) = PUP(from);
236 PUP(out) = PUP(from);
237 PUP(out) = PUP(from);
241 PUP(out) = PUP(from);
243 PUP(out) = PUP(from);
247 else if ((op & 64) == 0) { /* 2nd level distance code */
248 this = dcode[this.val + (hold & ((1U << op) - 1))];
252 strm->msg = (char *)"invalid distance code";
257 else if ((op & 64) == 0) { /* 2nd level length code */
258 this = lcode[this.val + (hold & ((1U << op) - 1))];
261 else if (op & 32) { /* end-of-block */
262 Tracevv((stderr, "inflate: end of block\n"));
267 strm->msg = (char *)"invalid literal/length code";
271 } while (in < last && out < end);
273 /* return unused bytes (on entry, bits < 8, so in won't go too far back) */
277 hold &= (1U << bits) - 1;
279 /* update state and return */
280 strm->next_in = in + OFF;
281 strm->next_out = out + OFF;
282 strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last));
283 strm->avail_out = (unsigned)(out < end ?
284 257 + (end - out) : 257 - (out - end));
291 inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe):
292 - Using bit fields for code structure
293 - Different op definition to avoid & for extra bits (do & for table bits)
294 - Three separate decoding do-loops for direct, window, and write == 0
295 - Special case for distance > 1 copies to do overlapped load and store copy
296 - Explicit branch predictions (based on measured branch probabilities)
297 - Deferring match copy and interspersed it with decoding subsequent codes
298 - Swapping literal/length else
299 - Swapping window/direct else
300 - Larger unrolled copy loops (three is about right)
301 - Moving len -= 3 statement into middle of loop