rjw | 1f88458 | 2022-01-06 17:20:42 +0800 | [diff] [blame^] | 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 |
| 4 | */ |
| 5 | |
| 6 | #include <lib/zutil.h> |
| 7 | #include <lib/inftrees.h> |
| 8 | #include <lib/inflate.h> |
| 9 | #include <lib/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 | /* |
| 34 | Decode literal, length, and distance codes and write out the resulting |
| 35 | literal and match bytes until either not enough input or output is |
| 36 | available, an end-of-block is encountered, or a data error is encountered. |
| 37 | When large enough input and output buffers are supplied to inflate(), for |
| 38 | example, a 16K input buffer and a 64K output buffer, more than 95% of the |
| 39 | inflate execution time is spent in this routine. |
| 40 | |
| 41 | Entry assumptions: |
| 42 | |
| 43 | state->mode == LEN |
| 44 | strm->avail_in >= 6 |
| 45 | strm->avail_out >= 258 |
| 46 | start >= strm->avail_out |
| 47 | state->bits < 8 |
| 48 | |
| 49 | On return, state->mode is one of: |
| 50 | |
| 51 | LEN -- ran out of enough output space or enough available input |
| 52 | TYPE -- reached end of block code, inflate() to interpret next block |
| 53 | BAD -- error in block data |
| 54 | |
| 55 | Notes: |
| 56 | |
| 57 | - The maximum input bits used by a length/distance pair is 15 bits for the |
| 58 | length code, 5 bits for the length extra, 15 bits for the distance code, |
| 59 | and 13 bits for the distance extra. This totals 48 bits, or six bytes. |
| 60 | Therefore if strm->avail_in >= 6, then there is enough input to avoid |
| 61 | checking for available input while decoding. |
| 62 | |
| 63 | - The maximum bytes that a single length/distance pair can output is 258 |
| 64 | bytes, which is the maximum length that can be coded. inflate_fast() |
| 65 | requires strm->avail_out >= 258 for each loop to avoid checking for |
| 66 | output space. |
| 67 | */ |
| 68 | void inflate_fast(strm, start) |
| 69 | z_streamp strm; |
| 70 | unsigned start; /* inflate()'s starting value for strm->avail_out */ |
| 71 | { |
| 72 | struct inflate_state FAR *state; |
| 73 | unsigned char FAR *in; /* local strm->next_in */ |
| 74 | unsigned char FAR *last; /* while in < last, enough input available */ |
| 75 | unsigned char FAR *out; /* local strm->next_out */ |
| 76 | unsigned char FAR *beg; /* inflate()'s initial strm->next_out */ |
| 77 | unsigned char FAR *end; /* while out < end, enough space available */ |
| 78 | #ifdef INFLATE_STRICT |
| 79 | unsigned dmax; /* maximum distance from zlib header */ |
| 80 | #endif |
| 81 | unsigned wsize; /* window size or zero if not using window */ |
| 82 | unsigned whave; /* valid bytes in the window */ |
| 83 | unsigned write; /* window write index */ |
| 84 | unsigned char FAR *window; /* allocated sliding window, if wsize != 0 */ |
| 85 | unsigned long hold; /* local strm->hold */ |
| 86 | unsigned bits; /* local strm->bits */ |
| 87 | code const FAR *lcode; /* local strm->lencode */ |
| 88 | code const FAR *dcode; /* local strm->distcode */ |
| 89 | unsigned lmask; /* mask for first level of length codes */ |
| 90 | unsigned dmask; /* mask for first level of distance codes */ |
| 91 | code this; /* retrieved table entry */ |
| 92 | unsigned op; /* code bits, operation, extra bits, or */ |
| 93 | /* window position, window bytes to copy */ |
| 94 | unsigned len; /* match length, unused bytes */ |
| 95 | unsigned dist; /* match distance */ |
| 96 | unsigned char FAR *from; /* where to copy match from */ |
| 97 | |
| 98 | /* copy state to local variables */ |
| 99 | state = (struct inflate_state FAR *)strm->state; |
| 100 | in = strm->next_in - OFF; |
| 101 | last = in + (strm->avail_in - 5); |
| 102 | out = strm->next_out - OFF; |
| 103 | beg = out - (start - strm->avail_out); |
| 104 | end = out + (strm->avail_out - 257); |
| 105 | #ifdef INFLATE_STRICT |
| 106 | dmax = state->dmax; |
| 107 | #endif |
| 108 | wsize = state->wsize; |
| 109 | whave = state->whave; |
| 110 | write = state->write; |
| 111 | window = state->window; |
| 112 | hold = state->hold; |
| 113 | bits = state->bits; |
| 114 | lcode = state->lencode; |
| 115 | dcode = state->distcode; |
| 116 | lmask = (1U << state->lenbits) - 1; |
| 117 | dmask = (1U << state->distbits) - 1; |
| 118 | |
| 119 | /* decode literals and length/distances until end-of-block or not enough |
| 120 | input data or output space */ |
| 121 | do { |
| 122 | if (bits < 15) { |
| 123 | hold += (unsigned long)(PUP(in)) << bits; |
| 124 | bits += 8; |
| 125 | hold += (unsigned long)(PUP(in)) << bits; |
| 126 | bits += 8; |
| 127 | } |
| 128 | this = lcode[hold & lmask]; |
| 129 | dolen: |
| 130 | op = (unsigned)(this.bits); |
| 131 | hold >>= op; |
| 132 | bits -= op; |
| 133 | op = (unsigned)(this.op); |
| 134 | if (op == 0) { /* literal */ |
| 135 | PUP(out) = (unsigned char)(this.val); |
| 136 | } else if (op & 16) { /* length base */ |
| 137 | len = (unsigned)(this.val); |
| 138 | op &= 15; /* number of extra bits */ |
| 139 | if (op) { |
| 140 | if (bits < op) { |
| 141 | hold += (unsigned long)(PUP(in)) << bits; |
| 142 | bits += 8; |
| 143 | } |
| 144 | len += (unsigned)hold & ((1U << op) - 1); |
| 145 | hold >>= op; |
| 146 | bits -= op; |
| 147 | } |
| 148 | if (bits < 15) { |
| 149 | hold += (unsigned long)(PUP(in)) << bits; |
| 150 | bits += 8; |
| 151 | hold += (unsigned long)(PUP(in)) << bits; |
| 152 | bits += 8; |
| 153 | } |
| 154 | this = dcode[hold & dmask]; |
| 155 | dodist: |
| 156 | op = (unsigned)(this.bits); |
| 157 | hold >>= op; |
| 158 | bits -= op; |
| 159 | op = (unsigned)(this.op); |
| 160 | if (op & 16) { /* distance base */ |
| 161 | dist = (unsigned)(this.val); |
| 162 | op &= 15; /* number of extra bits */ |
| 163 | if (bits < op) { |
| 164 | hold += (unsigned long)(PUP(in)) << bits; |
| 165 | bits += 8; |
| 166 | if (bits < op) { |
| 167 | hold += (unsigned long)(PUP(in)) << bits; |
| 168 | bits += 8; |
| 169 | } |
| 170 | } |
| 171 | dist += (unsigned)hold & ((1U << op) - 1); |
| 172 | #ifdef INFLATE_STRICT |
| 173 | if (dist > dmax) { |
| 174 | strm->msg = (char *)"invalid distance too far back"; |
| 175 | state->mode = BAD; |
| 176 | break; |
| 177 | } |
| 178 | #endif |
| 179 | hold >>= op; |
| 180 | bits -= op; |
| 181 | op = (unsigned)(out - beg); /* max distance in output */ |
| 182 | if (dist > op) { /* see if copy from window */ |
| 183 | op = dist - op; /* distance back in window */ |
| 184 | if (op > whave) { |
| 185 | strm->msg = (char *)"invalid distance too far back"; |
| 186 | state->mode = BAD; |
| 187 | break; |
| 188 | } |
| 189 | from = window - OFF; |
| 190 | if (write == 0) { /* very common case */ |
| 191 | from += wsize - op; |
| 192 | if (op < len) { /* some from window */ |
| 193 | len -= op; |
| 194 | do { |
| 195 | PUP(out) = PUP(from); |
| 196 | } while (--op); |
| 197 | from = out - dist; /* rest from output */ |
| 198 | } |
| 199 | } else if (write < op) { /* wrap around window */ |
| 200 | from += wsize + write - op; |
| 201 | op -= write; |
| 202 | if (op < len) { /* some from end of window */ |
| 203 | len -= op; |
| 204 | do { |
| 205 | PUP(out) = PUP(from); |
| 206 | } while (--op); |
| 207 | from = window - OFF; |
| 208 | if (write < len) { /* some from start of window */ |
| 209 | op = write; |
| 210 | len -= op; |
| 211 | do { |
| 212 | PUP(out) = PUP(from); |
| 213 | } while (--op); |
| 214 | from = out - dist; /* rest from output */ |
| 215 | } |
| 216 | } |
| 217 | } else { /* contiguous in window */ |
| 218 | from += write - op; |
| 219 | if (op < len) { /* some from window */ |
| 220 | len -= op; |
| 221 | do { |
| 222 | PUP(out) = PUP(from); |
| 223 | } while (--op); |
| 224 | from = out - dist; /* rest from output */ |
| 225 | } |
| 226 | } |
| 227 | while (len > 2) { |
| 228 | PUP(out) = PUP(from); |
| 229 | PUP(out) = PUP(from); |
| 230 | PUP(out) = PUP(from); |
| 231 | len -= 3; |
| 232 | } |
| 233 | if (len) { |
| 234 | PUP(out) = PUP(from); |
| 235 | if (len > 1) |
| 236 | PUP(out) = PUP(from); |
| 237 | } |
| 238 | } else { |
| 239 | from = out - dist; /* copy direct from output */ |
| 240 | do { /* minimum length is three */ |
| 241 | PUP(out) = PUP(from); |
| 242 | PUP(out) = PUP(from); |
| 243 | PUP(out) = PUP(from); |
| 244 | len -= 3; |
| 245 | } while (len > 2); |
| 246 | if (len) { |
| 247 | PUP(out) = PUP(from); |
| 248 | if (len > 1) |
| 249 | PUP(out) = PUP(from); |
| 250 | } |
| 251 | } |
| 252 | } else if ((op & 64) == 0) { /* 2nd level distance code */ |
| 253 | this = dcode[this.val + (hold & ((1U << op) - 1))]; |
| 254 | goto dodist; |
| 255 | } else { |
| 256 | strm->msg = (char *)"invalid distance code"; |
| 257 | state->mode = BAD; |
| 258 | break; |
| 259 | } |
| 260 | } else if ((op & 64) == 0) { /* 2nd level length code */ |
| 261 | this = lcode[this.val + (hold & ((1U << op) - 1))]; |
| 262 | goto dolen; |
| 263 | } else if (op & 32) { /* end-of-block */ |
| 264 | state->mode = TYPE; |
| 265 | break; |
| 266 | } else { |
| 267 | strm->msg = (char *)"invalid literal/length code"; |
| 268 | state->mode = BAD; |
| 269 | break; |
| 270 | } |
| 271 | } while (in < last && out < end); |
| 272 | |
| 273 | /* return unused bytes (on entry, bits < 8, so in won't go too far back) */ |
| 274 | len = bits >> 3; |
| 275 | in -= len; |
| 276 | bits -= len << 3; |
| 277 | hold &= (1U << bits) - 1; |
| 278 | |
| 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)); |
| 285 | state->hold = hold; |
| 286 | state->bits = bits; |
| 287 | return; |
| 288 | } |
| 289 | |
| 290 | /* |
| 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 |
| 302 | */ |
| 303 | |
| 304 | #endif /* !ASMINF */ |