| 1 | /* inffast_chunk.c -- fast decoding |
| 2 | * Copyright (C) 1995-2017 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 "contrib/optimizations/inffast_chunk.h" |
| 10 | #include "contrib/optimizations/chunkcopy.h" |
| 11 | |
| 12 | #ifdef ASMINF |
| 13 | # pragma message("Assembler code may have bugs -- use at your own risk") |
| 14 | #else |
| 15 | |
| 16 | /* |
| 17 | Decode literal, length, and distance codes and write out the resulting |
| 18 | literal and match bytes until either not enough input or output is |
| 19 | available, an end-of-block is encountered, or a data error is encountered. |
| 20 | When large enough input and output buffers are supplied to inflate(), for |
| 21 | example, a 16K input buffer and a 64K output buffer, more than 95% of the |
| 22 | inflate() execution time is spent in this routine. |
| 23 | |
| 24 | Entry assumptions: |
| 25 | |
| 26 | state->mode == LEN |
| 27 | strm->avail_in >= INFLATE_FAST_MIN_INPUT (6 or 8 bytes) |
| 28 | strm->avail_out >= INFLATE_FAST_MIN_OUTPUT (258 bytes) |
| 29 | start >= strm->avail_out |
| 30 | state->bits < 8 |
| 31 | (state->hold >> state->bits) == 0 |
| 32 | strm->next_out[0..strm->avail_out] does not overlap with |
| 33 | strm->next_in[0..strm->avail_in] |
| 34 | strm->state->window is allocated with an additional |
| 35 | CHUNKCOPY_CHUNK_SIZE-1 bytes of padding beyond strm->state->wsize |
| 36 | |
| 37 | On return, state->mode is one of: |
| 38 | |
| 39 | LEN -- ran out of enough output space or enough available input |
| 40 | TYPE -- reached end of block code, inflate() to interpret next block |
| 41 | BAD -- error in block data |
| 42 | |
| 43 | Notes: |
| 44 | |
| 45 | INFLATE_FAST_MIN_INPUT: 6 or 8 bytes |
| 46 | |
| 47 | - The maximum input bits used by a length/distance pair is 15 bits for the |
| 48 | length code, 5 bits for the length extra, 15 bits for the distance code, |
| 49 | and 13 bits for the distance extra. This totals 48 bits, or six bytes. |
| 50 | Therefore if strm->avail_in >= 6, then there is enough input to avoid |
| 51 | checking for available input while decoding. |
| 52 | |
| 53 | - The wide input data reading option reads 64 input bits at a time. Thus, |
| 54 | if strm->avail_in >= 8, then there is enough input to avoid checking for |
| 55 | available input while decoding. Reading consumes the input with: |
| 56 | |
| 57 | hold |= read64le(in) << bits; |
| 58 | in += 6; |
| 59 | bits += 48; |
| 60 | |
| 61 | reporting 6 bytes of new input because |bits| is 0..15 (2 bytes rounded |
| 62 | up, worst case) and 6 bytes is enough to decode as noted above. At exit, |
| 63 | hold &= (1U << bits) - 1 drops excess input to keep the invariant: |
| 64 | |
| 65 | (state->hold >> state->bits) == 0 |
| 66 | |
| 67 | INFLATE_FAST_MIN_OUTPUT: 258 bytes |
| 68 | |
| 69 | - The maximum bytes that a single length/distance pair can output is 258 |
| 70 | bytes, which is the maximum length that can be coded. inflate_fast() |
| 71 | requires strm->avail_out >= 258 for each loop to avoid checking for |
| 72 | available output space while decoding. |
| 73 | */ |
| 74 | void ZLIB_INTERNAL inflate_fast_chunk_(strm, start) |
| 75 | z_streamp strm; |
| 76 | unsigned start; /* inflate()'s starting value for strm->avail_out */ |
| 77 | { |
| 78 | struct inflate_state FAR *state; |
| 79 | z_const unsigned char FAR *in; /* local strm->next_in */ |
| 80 | z_const unsigned char FAR *last; /* have enough input while in < last */ |
| 81 | unsigned char FAR *out; /* local strm->next_out */ |
| 82 | unsigned char FAR *beg; /* inflate()'s initial strm->next_out */ |
| 83 | unsigned char FAR *end; /* while out < end, enough space available */ |
| 84 | unsigned char FAR *limit; /* safety limit for chunky copies */ |
| 85 | #ifdef INFLATE_STRICT |
| 86 | unsigned dmax; /* maximum distance from zlib header */ |
| 87 | #endif |
| 88 | unsigned wsize; /* window size or zero if not using window */ |
| 89 | unsigned whave; /* valid bytes in the window */ |
| 90 | unsigned wnext; /* window write index */ |
| 91 | unsigned char FAR *window; /* allocated sliding window, if wsize != 0 */ |
| 92 | inflate_holder_t hold; /* local strm->hold */ |
| 93 | unsigned bits; /* local strm->bits */ |
| 94 | code const FAR *lcode; /* local strm->lencode */ |
| 95 | code const FAR *dcode; /* local strm->distcode */ |
| 96 | unsigned lmask; /* mask for first level of length codes */ |
| 97 | unsigned dmask; /* mask for first level of distance codes */ |
| 98 | code here; /* retrieved table entry */ |
| 99 | unsigned op; /* code bits, operation, extra bits, or */ |
| 100 | /* window position, window bytes to copy */ |
| 101 | unsigned len; /* match length, unused bytes */ |
| 102 | unsigned dist; /* match distance */ |
| 103 | unsigned char FAR *from; /* where to copy match from */ |
| 104 | |
| 105 | /* copy state to local variables */ |
| 106 | state = (struct inflate_state FAR *)strm->state; |
| 107 | in = strm->next_in; |
| 108 | last = in + (strm->avail_in - (INFLATE_FAST_MIN_INPUT - 1)); |
| 109 | out = strm->next_out; |
| 110 | beg = out - (start - strm->avail_out); |
| 111 | end = out + (strm->avail_out - (INFLATE_FAST_MIN_OUTPUT - 1)); |
| 112 | limit = out + strm->avail_out; |
| 113 | #ifdef INFLATE_STRICT |
| 114 | dmax = state->dmax; |
| 115 | #endif |
| 116 | wsize = state->wsize; |
| 117 | whave = state->whave; |
| 118 | wnext = (state->wnext == 0 && whave >= wsize) ? wsize : state->wnext; |
| 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 | #ifdef INFLATE_CHUNK_READ_64LE |
| 132 | hold |= read64le(in) << bits; |
| 133 | in += 6; |
| 134 | bits += 48; |
| 135 | #else |
| 136 | hold += (unsigned long)(*in++) << bits; |
| 137 | bits += 8; |
| 138 | hold += (unsigned long)(*in++) << bits; |
| 139 | bits += 8; |
| 140 | #endif |
| 141 | } |
| 142 | here = lcode[hold & lmask]; |
| 143 | dolen: |
| 144 | op = (unsigned)(here.bits); |
| 145 | hold >>= op; |
| 146 | bits -= op; |
| 147 | op = (unsigned)(here.op); |
| 148 | if (op == 0) { /* literal */ |
| 149 | Tracevv((stderr, here.val >= 0x20 && here.val < 0x7f ? |
| 150 | "inflate: literal '%c'\n" : |
| 151 | "inflate: literal 0x%02x\n" , here.val)); |
| 152 | *out++ = (unsigned char)(here.val); |
| 153 | } |
| 154 | else if (op & 16) { /* length base */ |
| 155 | len = (unsigned)(here.val); |
| 156 | op &= 15; /* number of extra bits */ |
| 157 | if (op) { |
| 158 | if (bits < op) { |
| 159 | #ifdef INFLATE_CHUNK_READ_64LE |
| 160 | hold |= read64le(in) << bits; |
| 161 | in += 6; |
| 162 | bits += 48; |
| 163 | #else |
| 164 | hold += (unsigned long)(*in++) << bits; |
| 165 | bits += 8; |
| 166 | #endif |
| 167 | } |
| 168 | len += (unsigned)hold & ((1U << op) - 1); |
| 169 | hold >>= op; |
| 170 | bits -= op; |
| 171 | } |
| 172 | Tracevv((stderr, "inflate: length %u\n" , len)); |
| 173 | if (bits < 15) { |
| 174 | #ifdef INFLATE_CHUNK_READ_64LE |
| 175 | hold |= read64le(in) << bits; |
| 176 | in += 6; |
| 177 | bits += 48; |
| 178 | #else |
| 179 | hold += (unsigned long)(*in++) << bits; |
| 180 | bits += 8; |
| 181 | hold += (unsigned long)(*in++) << bits; |
| 182 | bits += 8; |
| 183 | #endif |
| 184 | } |
| 185 | here = dcode[hold & dmask]; |
| 186 | dodist: |
| 187 | op = (unsigned)(here.bits); |
| 188 | hold >>= op; |
| 189 | bits -= op; |
| 190 | op = (unsigned)(here.op); |
| 191 | if (op & 16) { /* distance base */ |
| 192 | dist = (unsigned)(here.val); |
| 193 | op &= 15; /* number of extra bits */ |
| 194 | if (bits < op) { |
| 195 | #ifdef INFLATE_CHUNK_READ_64LE |
| 196 | hold |= read64le(in) << bits; |
| 197 | in += 6; |
| 198 | bits += 48; |
| 199 | #else |
| 200 | hold += (unsigned long)(*in++) << bits; |
| 201 | bits += 8; |
| 202 | if (bits < op) { |
| 203 | hold += (unsigned long)(*in++) << bits; |
| 204 | bits += 8; |
| 205 | } |
| 206 | #endif |
| 207 | } |
| 208 | dist += (unsigned)hold & ((1U << op) - 1); |
| 209 | #ifdef INFLATE_STRICT |
| 210 | if (dist > dmax) { |
| 211 | strm->msg = (char *)"invalid distance too far back" ; |
| 212 | state->mode = BAD; |
| 213 | break; |
| 214 | } |
| 215 | #endif |
| 216 | hold >>= op; |
| 217 | bits -= op; |
| 218 | Tracevv((stderr, "inflate: distance %u\n" , dist)); |
| 219 | op = (unsigned)(out - beg); /* max distance in output */ |
| 220 | if (dist > op) { /* see if copy from window */ |
| 221 | op = dist - op; /* distance back in window */ |
| 222 | if (op > whave) { |
| 223 | if (state->sane) { |
| 224 | strm->msg = |
| 225 | (char *)"invalid distance too far back" ; |
| 226 | state->mode = BAD; |
| 227 | break; |
| 228 | } |
| 229 | #ifdef INFLATE_ALLOW_INVALID_DISTANCE_TOOFAR_ARRR |
| 230 | if (len <= op - whave) { |
| 231 | do { |
| 232 | *out++ = 0; |
| 233 | } while (--len); |
| 234 | continue; |
| 235 | } |
| 236 | len -= op - whave; |
| 237 | do { |
| 238 | *out++ = 0; |
| 239 | } while (--op > whave); |
| 240 | if (op == 0) { |
| 241 | from = out - dist; |
| 242 | do { |
| 243 | *out++ = *from++; |
| 244 | } while (--len); |
| 245 | continue; |
| 246 | } |
| 247 | #endif |
| 248 | } |
| 249 | from = window; |
| 250 | if (wnext >= op) { /* contiguous in window */ |
| 251 | from += wnext - op; |
| 252 | } |
| 253 | else { /* wrap around window */ |
| 254 | op -= wnext; |
| 255 | from += wsize - op; |
| 256 | if (op < len) { /* some from end of window */ |
| 257 | len -= op; |
| 258 | out = chunkcopy_safe(out, from, op, limit); |
| 259 | from = window; /* more from start of window */ |
| 260 | op = wnext; |
| 261 | /* This (rare) case can create a situation where |
| 262 | the first chunkcopy below must be checked. |
| 263 | */ |
| 264 | } |
| 265 | } |
| 266 | if (op < len) { /* still need some from output */ |
| 267 | out = chunkcopy_safe(out, from, op, limit); |
| 268 | len -= op; |
| 269 | /* When dist is small the amount of data that can be |
| 270 | copied from the window is also small, and progress |
| 271 | towards the dangerous end of the output buffer is |
| 272 | also small. This means that for trivial memsets and |
| 273 | for chunkunroll_relaxed() a safety check is |
| 274 | unnecessary. However, these conditions may not be |
| 275 | entered at all, and in that case it's possible that |
| 276 | the main copy is near the end. |
| 277 | */ |
| 278 | out = chunkunroll_relaxed(out, &dist, &len); |
| 279 | out = chunkcopy_safe(out, out - dist, len, limit); |
| 280 | } else { |
| 281 | /* from points to window, so there is no risk of |
| 282 | overlapping pointers requiring memset-like behaviour |
| 283 | */ |
| 284 | out = chunkcopy_safe(out, from, len, limit); |
| 285 | } |
| 286 | } |
| 287 | else { |
| 288 | /* Whole reference is in range of current output. No |
| 289 | range checks are necessary because we start with room |
| 290 | for at least 258 bytes of output, so unroll and roundoff |
| 291 | operations can write beyond `out+len` so long as they |
| 292 | stay within 258 bytes of `out`. |
| 293 | */ |
| 294 | out = chunkcopy_lapped_relaxed(out, dist, len); |
| 295 | } |
| 296 | } |
| 297 | else if ((op & 64) == 0) { /* 2nd level distance code */ |
| 298 | here = dcode[here.val + (hold & ((1U << op) - 1))]; |
| 299 | goto dodist; |
| 300 | } |
| 301 | else { |
| 302 | strm->msg = (char *)"invalid distance code" ; |
| 303 | state->mode = BAD; |
| 304 | break; |
| 305 | } |
| 306 | } |
| 307 | else if ((op & 64) == 0) { /* 2nd level length code */ |
| 308 | here = lcode[here.val + (hold & ((1U << op) - 1))]; |
| 309 | goto dolen; |
| 310 | } |
| 311 | else if (op & 32) { /* end-of-block */ |
| 312 | Tracevv((stderr, "inflate: end of block\n" )); |
| 313 | state->mode = TYPE; |
| 314 | break; |
| 315 | } |
| 316 | else { |
| 317 | strm->msg = (char *)"invalid literal/length code" ; |
| 318 | state->mode = BAD; |
| 319 | break; |
| 320 | } |
| 321 | } while (in < last && out < end); |
| 322 | |
| 323 | /* return unused bytes (on entry, bits < 8, so in won't go too far back) */ |
| 324 | len = bits >> 3; |
| 325 | in -= len; |
| 326 | bits -= len << 3; |
| 327 | hold &= (1U << bits) - 1; |
| 328 | |
| 329 | /* update state and return */ |
| 330 | strm->next_in = in; |
| 331 | strm->next_out = out; |
| 332 | strm->avail_in = (unsigned)(in < last ? |
| 333 | (INFLATE_FAST_MIN_INPUT - 1) + (last - in) : |
| 334 | (INFLATE_FAST_MIN_INPUT - 1) - (in - last)); |
| 335 | strm->avail_out = (unsigned)(out < end ? |
| 336 | (INFLATE_FAST_MIN_OUTPUT - 1) + (end - out) : |
| 337 | (INFLATE_FAST_MIN_OUTPUT - 1) - (out - end)); |
| 338 | state->hold = hold; |
| 339 | state->bits = bits; |
| 340 | |
| 341 | Assert((state->hold >> state->bits) == 0, "invalid input data state" ); |
| 342 | return; |
| 343 | } |
| 344 | |
| 345 | /* |
| 346 | inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe): |
| 347 | - Using bit fields for code structure |
| 348 | - Different op definition to avoid & for extra bits (do & for table bits) |
| 349 | - Three separate decoding do-loops for direct, window, and wnext == 0 |
| 350 | - Special case for distance > 1 copies to do overlapped load and store copy |
| 351 | - Explicit branch predictions (based on measured branch probabilities) |
| 352 | - Deferring match copy and interspersed it with decoding subsequent codes |
| 353 | - Swapping literal/length else |
| 354 | - Swapping window/direct else |
| 355 | - Larger unrolled copy loops (three is about right) |
| 356 | - Moving len -= 3 statement into middle of loop |
| 357 | */ |
| 358 | |
| 359 | #endif /* !ASMINF */ |
| 360 | |