1/* inffast.c -- fast decoding 2 * Copyright (C) 1995-2008, 2010 Mark Adler 3 * For conditions of distribution and use, see copyright notice in zlib.h 4 */ 5 6#if defined __arm__ 7#include <arm/arch.h> 8#endif 9 10 11#include "zutil.h" 12#include "inftrees.h" 13#include "inflate.h" 14#include "inffast.h" 15 16#ifndef ASMINF 17 18/* Allow machine dependent optimization for post-increment or pre-increment. 19 Based on testing to date, 20 Pre-increment preferred for: 21 - PowerPC G3 (Adler) 22 - MIPS R5000 (Randers-Pehrson) 23 Post-increment preferred for: 24 - none 25 No measurable difference: 26 - Pentium III (Anderson) 27 - M68060 (Nikl) 28 */ 29#ifdef POSTINC 30# define OFF 0 31# define PUP(a) *(a)++ 32#else 33# define OFF 1 34# define PUP(a) *++(a) 35#endif 36 37/* 38 Decode literal, length, and distance codes and write out the resulting 39 literal and match bytes until either not enough input or output is 40 available, an end-of-block is encountered, or a data error is encountered. 41 When large enough input and output buffers are supplied to inflate(), for 42 example, a 16K input buffer and a 64K output buffer, more than 95% of the 43 inflate execution time is spent in this routine. 44 45 Entry assumptions: 46 47 state->mode == LEN 48 strm->avail_in >= 6 49 strm->avail_out >= 258 50 start >= strm->avail_out 51 state->bits < 8 52 53 On return, state->mode is one of: 54 55 LEN -- ran out of enough output space or enough available input 56 TYPE -- reached end of block code, inflate() to interpret next block 57 BAD -- error in block data 58 59 Notes: 60 61 - The maximum input bits used by a length/distance pair is 15 bits for the 62 length code, 5 bits for the length extra, 15 bits for the distance code, 63 and 13 bits for the distance extra. This totals 48 bits, or six bytes. 64 Therefore if strm->avail_in >= 6, then there is enough input to avoid 65 checking for available input while decoding. 66 67 - The maximum bytes that a single length/distance pair can output is 258 68 bytes, which is the maximum length that can be coded. inflate_fast() 69 requires strm->avail_out >= 258 for each loop to avoid checking for 70 output space. 71 */ 72void ZLIB_INTERNAL inflate_fast(strm, start) 73z_streamp strm; 74unsigned start; /* inflate()'s starting value for strm->avail_out */ 75{ 76 struct inflate_state FAR *state; 77 unsigned char FAR *in; /* local strm->next_in */ 78 unsigned char FAR *last; /* while in < last, enough input available */ 79 unsigned char FAR *out; /* local strm->next_out */ 80 unsigned char FAR *beg; /* inflate()'s initial strm->next_out */ 81 unsigned char FAR *end; /* while out < end, enough space available */ 82#ifdef INFLATE_STRICT 83 unsigned dmax; /* maximum distance from zlib header */ 84#endif 85 unsigned wsize; /* window size or zero if not using window */ 86 unsigned whave; /* valid bytes in the window */ 87 unsigned wnext; /* window write index */ 88 unsigned char FAR *window; /* allocated sliding window, if wsize != 0 */ 89 unsigned long hold; /* local strm->hold */ 90 unsigned bits; /* local strm->bits */ 91 code const FAR *lcode; /* local strm->lencode */ 92 code const FAR *dcode; /* local strm->distcode */ 93 unsigned lmask; /* mask for first level of length codes */ 94 unsigned dmask; /* mask for first level of distance codes */ 95 code here; /* retrieved table entry */ 96 unsigned op; /* code bits, operation, extra bits, or */ 97 /* window position, window bytes to copy */ 98 unsigned len; /* match length, unused bytes */ 99 unsigned dist; /* match distance */ 100 unsigned char FAR *from; /* where to copy match from */ 101 102 /* copy state to local variables */ 103 state = (struct inflate_state FAR *)strm->state; 104 in = strm->next_in - OFF; 105 last = in + (strm->avail_in - 5); 106 out = strm->next_out - OFF; 107 beg = out - (start - strm->avail_out); 108 end = out + (strm->avail_out - 257); 109#ifdef INFLATE_STRICT 110 dmax = state->dmax; 111#endif 112 wsize = state->wsize; 113 whave = state->whave; 114 wnext = state->wnext; 115 window = state->window; 116 hold = state->hold; 117 bits = state->bits; 118 lcode = state->lencode; 119 dcode = state->distcode; 120 lmask = (1U << state->lenbits) - 1; 121 dmask = (1U << state->distbits) - 1; 122 123 /* decode literals and length/distances until end-of-block or not enough 124 input data or output space */ 125 do { 126 if (bits < 15) { 127 hold += (unsigned long)(PUP(in)) << bits; 128 bits += 8; 129 hold += (unsigned long)(PUP(in)) << bits; 130 bits += 8; 131 } 132 here = lcode[hold & lmask]; 133 dolen: 134 op = (unsigned)(here.bits); 135 hold >>= op; 136 bits -= op; 137 op = (unsigned)(here.op); 138 if (op == 0) { /* literal */ 139 Tracevv((stderr, here.val >= 0x20 && here.val < 0x7f ? 140 "inflate: literal '%c'\n" : 141 "inflate: literal 0x%02x\n", here.val)); 142 PUP(out) = (unsigned char)(here.val); 143 } 144 else if (op & 16) { /* length base */ 145 len = (unsigned)(here.val); 146 op &= 15; /* number of extra bits */ 147 if (op) { 148 if (bits < op) { 149 hold += (unsigned long)(PUP(in)) << bits; 150 bits += 8; 151 } 152 len += (unsigned)hold & ((1U << op) - 1); 153 hold >>= op; 154 bits -= op; 155 } 156 Tracevv((stderr, "inflate: length %u\n", len)); 157 if (bits < 15) { 158 hold += (unsigned long)(PUP(in)) << bits; 159 bits += 8; 160 hold += (unsigned long)(PUP(in)) << bits; 161 bits += 8; 162 } 163 here = dcode[hold & dmask]; 164 dodist: 165 op = (unsigned)(here.bits); 166 hold >>= op; 167 bits -= op; 168 op = (unsigned)(here.op); 169 if (op & 16) { /* distance base */ 170 dist = (unsigned)(here.val); 171 op &= 15; /* number of extra bits */ 172 if (bits < op) { 173 hold += (unsigned long)(PUP(in)) << bits; 174 bits += 8; 175 if (bits < op) { 176 hold += (unsigned long)(PUP(in)) << bits; 177 bits += 8; 178 } 179 } 180 dist += (unsigned)hold & ((1U << op) - 1); 181#ifdef INFLATE_STRICT 182 if (dist > dmax) { 183 strm->msg = (char *)"invalid distance too far back"; 184 state->mode = BAD; 185 break; 186 } 187#endif 188 hold >>= op; 189 bits -= op; 190 Tracevv((stderr, "inflate: distance %u\n", dist)); 191 op = (unsigned)(out - beg); /* max distance in output */ 192 if (dist > op) { /* see if copy from window */ 193 op = dist - op; /* distance back in window */ 194 if (op > whave) { 195 if (state->sane) { 196 strm->msg = 197 (char *)"invalid distance too far back"; 198 state->mode = BAD; 199 break; 200 } 201#ifdef INFLATE_ALLOW_INVALID_DISTANCE_TOOFAR_ARRR 202 if (len <= op - whave) { 203 do { 204 PUP(out) = 0; 205 } while (--len); 206 continue; 207 } 208 len -= op - whave; 209 do { 210 PUP(out) = 0; 211 } while (--op > whave); 212 if (op == 0) { 213 from = out - dist; 214 do { 215 PUP(out) = PUP(from); 216 } while (--len); 217 continue; 218 } 219#endif 220 } 221 from = window - OFF; 222 if (wnext == 0) { /* very common case */ 223 from += wsize - op; 224 if (op < len) { /* some from window */ 225 len -= op; 226 do { 227 PUP(out) = PUP(from); 228 } while (--op); 229 from = out - dist; /* rest from output */ 230 } 231 } 232 else if (wnext < op) { /* wrap around window */ 233 from += wsize + wnext - op; 234 op -= wnext; 235 if (op < len) { /* some from end of window */ 236 len -= op; 237 do { 238 PUP(out) = PUP(from); 239 } while (--op); 240 from = window - OFF; 241 if (wnext < len) { /* some from start of window */ 242 op = wnext; 243 len -= op; 244 do { 245 PUP(out) = PUP(from); 246 } while (--op); 247 from = out - dist; /* rest from output */ 248 } 249 } 250 } 251 else { /* contiguous in window */ 252 from += wnext - op; 253 if (op < len) { /* some from window */ 254 len -= op; 255 do { 256 PUP(out) = PUP(from); 257 } while (--op); 258 from = out - dist; /* rest from output */ 259 } 260 } 261 while (len > 2) { 262 PUP(out) = PUP(from); 263 PUP(out) = PUP(from); 264 PUP(out) = PUP(from); 265 len -= 3; 266 } 267 if (len) { 268 PUP(out) = PUP(from); 269 if (len > 1) 270 PUP(out) = PUP(from); 271 } 272 } 273 else { 274 from = out - dist; /* copy direct from output */ 275 do { /* minimum length is three */ 276 PUP(out) = PUP(from); 277 PUP(out) = PUP(from); 278 PUP(out) = PUP(from); 279 len -= 3; 280 } while (len > 2); 281 if (len) { 282 PUP(out) = PUP(from); 283 if (len > 1) 284 PUP(out) = PUP(from); 285 } 286 } 287 } 288 else if ((op & 64) == 0) { /* 2nd level distance code */ 289 here = dcode[here.val + (hold & ((1U << op) - 1))]; 290 goto dodist; 291 } 292 else { 293 strm->msg = (char *)"invalid distance code"; 294 state->mode = BAD; 295 break; 296 } 297 } 298 else if ((op & 64) == 0) { /* 2nd level length code */ 299 here = lcode[here.val + (hold & ((1U << op) - 1))]; 300 goto dolen; 301 } 302 else if (op & 32) { /* end-of-block */ 303 Tracevv((stderr, "inflate: end of block\n")); 304 state->mode = TYPE; 305 break; 306 } 307 else { 308 strm->msg = (char *)"invalid literal/length code"; 309 state->mode = BAD; 310 break; 311 } 312 } while (in < last && out < end); 313 314 /* return unused bytes (on entry, bits < 8, so in won't go too far back) */ 315 len = bits >> 3; 316 in -= len; 317 bits -= len << 3; 318 hold &= (1U << bits) - 1; 319 320 /* update state and return */ 321 strm->next_in = in + OFF; 322 strm->next_out = out + OFF; 323 strm->avail_in = (unsigned)(in < last ? 5 + (last - in) : 5 - (in - last)); 324 strm->avail_out = (unsigned)(out < end ? 325 257 + (end - out) : 257 - (out - end)); 326 state->hold = hold; 327 state->bits = bits; 328 return; 329} 330 331/* 332 inflate_fast() speedups that turned out slower (on a PowerPC G3 750CXe): 333 - Using bit fields for code structure 334 - Different op definition to avoid & for extra bits (do & for table bits) 335 - Three separate decoding do-loops for direct, window, and wnext == 0 336 - Special case for distance > 1 copies to do overlapped load and store copy 337 - Explicit branch predictions (based on measured branch probabilities) 338 - Deferring match copy and interspersed it with decoding subsequent codes 339 - Swapping literal/length else 340 - Swapping window/direct else 341 - Larger unrolled copy loops (three is about right) 342 - Moving len -= 3 statement into middle of loop 343 */ 344 345#endif /* !ASMINF */ 346 347