src/zlib/crc32.c

   1 /* crc32.c -- compute the CRC-32 of a data stream
   2  * Copyright (C) 1995-2006, 2010, 2011, 2012 Mark Adler
   3  * For conditions of distribution and use, see copyright notice in zlib.h
   4  *
   5  * Thanks to Rodney Brown <rbrown64@csc.com.au> for his contribution of faster
   6  * CRC methods: exclusive-oring 32 bits of data at a time, and pre-computing
   7  * tables for updating the shift register in one step with three exclusive-ors
   8  * instead of four steps with four exclusive-ors.  This results in about a
   9  * factor of two increase in speed on a Power PC G4 (PPC7455) using gcc -O3.
  10  */
  11
  12
  13 /*
  14   Note on the use of DYNAMIC_CRC_TABLE: there is no mutex or semaphore
  15   protection on the static variables used to control the first-use generation
  16   of the crc tables.  Therefore, if you #define DYNAMIC_CRC_TABLE, you should
  17   first call get_crc_table() to initialize the tables before allowing more than
  18   one thread to use crc32().
  19
  20   DYNAMIC_CRC_TABLE and MAKECRCH can be #defined to write out crc32.h.
  21  */
  22
  23 #ifdef MAKECRCH
  24 #  include <stdio.h>
  25 #  ifndef DYNAMIC_CRC_TABLE
  26 #    define DYNAMIC_CRC_TABLE
  27 #  endif /* !DYNAMIC_CRC_TABLE */
  28 #endif /* MAKECRCH */
  29
  30 #include "zutil.h"      /* for STDC and FAR definitions */
  31
  32 #define local static
  33
  34 /* Definitions for doing the crc four data bytes at a time. */
  35 #if !defined(NOBYFOUR) && defined(Z_U4)
  36 #  define BYFOUR
  37 #endif
  38 #ifdef BYFOUR
  39    local unsigned long crc32_little OF((unsigned long,
  40                         const unsigned char FAR *, unsigned));
  41    local unsigned long crc32_big OF((unsigned long,
  42                         const unsigned char FAR *, unsigned));
  43 #  define TBLS 8
  44 #else
  45 #  define TBLS 1
  46 #endif /* BYFOUR */
  47
  48 /* Local functions for crc concatenation */
  49 local unsigned long gf2_matrix_times OF((unsigned long *mat,
  50                                          unsigned long vec));
  51 local void gf2_matrix_square OF((unsigned long *square, unsigned long *mat));
  52 local uLong crc32_combine_ OF((uLong crc1, uLong crc2, z_off64_t len2));
  53
  54
  55 #ifdef DYNAMIC_CRC_TABLE
  56
  57 local volatile int crc_table_empty = 1;
  58 local z_crc_t FAR crc_table[TBLS][256];
  59 local void make_crc_table OF((void));
  60 #ifdef MAKECRCH
  61    local void write_table OF((FILE *, const z_crc_t FAR *));
  62 #endif /* MAKECRCH */
  63 /*
  64   Generate tables for a byte-wise 32-bit CRC calculation on the polynomial:
  65   x^32+x^26+x^23+x^22+x^16+x^12+x^11+x^10+x^8+x^7+x^5+x^4+x^2+x+1.
  66
  67   Polynomials over GF(2) are represented in binary, one bit per coefficient,
  68   with the lowest powers in the most significant bit.  Then adding polynomials
  69   is just exclusive-or, and multiplying a polynomial by x is a right shift by
  70   one.  If we call the above polynomial p, and represent a byte as the
  71   polynomial q, also with the lowest power in the most significant bit (so the
  72   byte 0xb1 is the polynomial x^7+x^3+x+1), then the CRC is (q*x^32) mod p,
  73   where a mod b means the remainder after dividing a by b.
  74
  75   This calculation is done using the shift-register method of multiplying and
  76   taking the remainder.  The register is initialized to zero, and for each
  77   incoming bit, x^32 is added mod p to the register if the bit is a one (where
  78   x^32 mod p is p+x^32 = x^26+...+1), and the register is multiplied mod p by
  79   x (which is shifting right by one and adding x^32 mod p if the bit shifted
  80   out is a one).  We start with the highest power (least significant bit) of
  81   q and repeat for all eight bits of q.
  82
  83   The first table is simply the CRC of all possible eight bit values.  This is
  84   all the information needed to generate CRCs on data a byte at a time for all
  85   combinations of CRC register values and incoming bytes.  The remaining tables
  86   allow for word-at-a-time CRC calculation for both big-endian and little-
  87   endian machines, where a word is four bytes.
  88 */
  89 local void make_crc_table()
  90 {
  91     z_crc_t c;
  92     int n, k;
  93     z_crc_t poly;                       /* polynomial exclusive-or pattern */
  94     /* terms of polynomial defining this crc (except x^32): */
  95     static volatile int first = 1;      /* flag to limit concurrent making */
  96     static const unsigned char p[] = {0,1,2,4,5,7,8,10,11,12,16,22,23,26};
  97
  98     /* See if another task is already doing this (not thread-safe, but better
  99        than nothing -- significantly reduces duration of vulnerability in
 100        case the advice about DYNAMIC_CRC_TABLE is ignored) */
 101     if (first) {
 102         first = 0;
 103
 104         /* make exclusive-or pattern from polynomial (0xedb88320UL) */
 105         poly = 0;
 106         for (n = 0; n < (int)(sizeof(p)/sizeof(unsigned char)); n++)
 107             poly |= (z_crc_t)1 << (31 - p[n]);
 108
 109         /* generate a crc for every 8-bit value */
 110         for (n = 0; n < 256; n++) {
 111             c = (z_crc_t)n;
 112             for (k = 0; k < 8; k++)
 113                 c = c & 1 ? poly ^ (c >> 1) : c >> 1;
 114             crc_table[0][n] = c;
 115         }
 116
 117 #ifdef BYFOUR
 118         /* generate crc for each value followed by one, two, and three zeros,
 119            and then the byte reversal of those as well as the first table */
 120         for (n = 0; n < 256; n++) {
 121             c = crc_table[0][n];
 122             crc_table[4][n] = ZSWAP32(c);
 123             for (k = 1; k < 4; k++) {
 124                 c = crc_table[0][c & 0xff] ^ (c >> 8);
 125                 crc_table[k][n] = c;
 126                 crc_table[k + 4][n] = ZSWAP32(c);
 127             }
 128         }
 129 #endif /* BYFOUR */
 130
 131         crc_table_empty = 0;
 132     }
 133     else {      /* not first */
 134         /* wait for the other guy to finish (not efficient, but rare) */
 135         while (crc_table_empty)
 136             ;
 137     }
 138
 139 #ifdef MAKECRCH
 140     /* write out CRC tables to crc32.h */
 141     {
 142         FILE *out;
 143
 144         out = fopen("crc32.h", "w");
 145         if (out == NULL) return;
 146         fprintf(out, "/* crc32.h -- tables for rapid CRC calculation\n");
 147         fprintf(out, " * Generated automatically by crc32.c\n */\n\n");
 148         fprintf(out, "local const z_crc_t FAR ");
 149         fprintf(out, "crc_table[TBLS][256] =\n{\n  {\n");
 150         write_table(out, crc_table[0]);
 151 #  ifdef BYFOUR
 152         fprintf(out, "#ifdef BYFOUR\n");
 153         for (k = 1; k < 8; k++) {
 154             fprintf(out, "  },\n  {\n");
 155             write_table(out, crc_table[k]);
 156         }
 157         fprintf(out, "#endif\n");
 158 #  endif /* BYFOUR */
 159         fprintf(out, "  }\n};\n");
 160         fclose(out);
 161     }
 162 #endif /* MAKECRCH */
 163 }
 164
 165 #ifdef MAKECRCH
 166 local void write_table(out, table)
 167     FILE *out;
 168     const z_crc_t FAR *table;
 169 {
 170     int n;
 171
 172     for (n = 0; n < 256; n++)
 173         fprintf(out, "%s0x%08lxUL%s", n % 5 ? "" : "    ",
 174                 (unsigned long)(table[n]),
 175                 n == 255 ? "\n" : (n % 5 == 4 ? ",\n" : ", "));
 176 }
 177 #endif /* MAKECRCH */
 178
 179 #else /* !DYNAMIC_CRC_TABLE */
 180 /* ========================================================================
 181  * Tables of CRC-32s of all single-byte values, made by make_crc_table().
 182  */
 183 #include "crc32.h"
 184 #endif /* DYNAMIC_CRC_TABLE */
 185
 186 /* =========================================================================
 187  * This function can be used by asm versions of crc32()
 188  */
 189 const z_crc_t FAR * ZEXPORT get_crc_table()
 190 {
 191 #ifdef DYNAMIC_CRC_TABLE
 192     if (crc_table_empty)
 193         make_crc_table();
 194 #endif /* DYNAMIC_CRC_TABLE */
 195     return (const z_crc_t FAR *)crc_table;
 196 }
 197
 198 /* ========================================================================= */
 199 #define DO1 crc = crc_table[0][((int)crc ^ (*buf++)) & 0xff] ^ (crc >> 8)
 200 #define DO8 DO1; DO1; DO1; DO1; DO1; DO1; DO1; DO1
 201
 202 /* ========================================================================= */
 203 unsigned long ZEXPORT crc32(crc, buf, len)
 204     unsigned long crc;
 205     const unsigned char FAR *buf;
 206     uInt len;
 207 {
 208     if (buf == Z_NULL) return 0UL;
 209
 210 #ifdef DYNAMIC_CRC_TABLE
 211     if (crc_table_empty)
 212         make_crc_table();
 213 #endif /* DYNAMIC_CRC_TABLE */
 214
 215 #ifdef BYFOUR
 216     if (sizeof(void *) == sizeof(ptrdiff_t)) {
 217         z_crc_t endian;
 218
 219         endian = 1;
 220         if (*((unsigned char *)(&endian)))
 221             return crc32_little(crc, buf, len);
 222         else
 223             return crc32_big(crc, buf, len);
 224     }
 225 #endif /* BYFOUR */
 226     crc = crc ^ 0xffffffffUL;
 227     while (len >= 8) {
 228         DO8;
 229         len -= 8;
 230     }
 231     if (len) do {
 232         DO1;
 233     } while (--len);
 234     return crc ^ 0xffffffffUL;
 235 }
 236
 237 #ifdef BYFOUR
 238
 239 /* ========================================================================= */
 240 #define DOLIT4 c ^= *buf4++; \
 241         c = crc_table[3][c & 0xff] ^ crc_table[2][(c >> 8) & 0xff] ^ \
 242             crc_table[1][(c >> 16) & 0xff] ^ crc_table[0][c >> 24]
 243 #define DOLIT32 DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4
 244
 245 /* ========================================================================= */
 246 local unsigned long crc32_little(crc, buf, len)
 247     unsigned long crc;
 248     const unsigned char FAR *buf;
 249     unsigned len;
 250 {
 251     register z_crc_t c;
 252     register const z_crc_t FAR *buf4;
 253
 254     c = (z_crc_t)crc;
 255     c = ~c;
 256     while (len && ((ptrdiff_t)buf & 3)) {
 257         c = crc_table[0][(c ^ *buf++) & 0xff] ^ (c >> 8);
 258         len--;
 259     }
 260
 261     buf4 = (const z_crc_t FAR *)(const void FAR *)buf;
 262     while (len >= 32) {
 263         DOLIT32;
 264         len -= 32;
 265     }
 266     while (len >= 4) {
 267         DOLIT4;
 268         len -= 4;
 269     }
 270     buf = (const unsigned char FAR *)buf4;
 271
 272     if (len) do {
 273         c = crc_table[0][(c ^ *buf++) & 0xff] ^ (c >> 8);
 274     } while (--len);
 275     c = ~c;
 276     return (unsigned long)c;
 277 }
 278
 279 /* ========================================================================= */
 280 #define DOBIG4 c ^= *++buf4; \
 281         c = crc_table[4][c & 0xff] ^ crc_table[5][(c >> 8) & 0xff] ^ \
 282             crc_table[6][(c >> 16) & 0xff] ^ crc_table[7][c >> 24]
 283 #define DOBIG32 DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4
 284
 285 /* ========================================================================= */
 286 local unsigned long crc32_big(crc, buf, len)
 287     unsigned long crc;
 288     const unsigned char FAR *buf;
 289     unsigned len;
 290 {
 291     register z_crc_t c;
 292     register const z_crc_t FAR *buf4;
 293
 294     c = ZSWAP32((z_crc_t)crc);
 295     c = ~c;
 296     while (len && ((ptrdiff_t)buf & 3)) {
 297         c = crc_table[4][(c >> 24) ^ *buf++] ^ (c << 8);
 298         len--;
 299     }
 300
 301     buf4 = (const z_crc_t FAR *)(const void FAR *)buf;
 302     buf4--;
 303     while (len >= 32) {
 304         DOBIG32;
 305         len -= 32;
 306     }
 307     while (len >= 4) {
 308         DOBIG4;
 309         len -= 4;
 310     }
 311     buf4++;
 312     buf = (const unsigned char FAR *)buf4;
 313
 314     if (len) do {
 315         c = crc_table[4][(c >> 24) ^ *buf++] ^ (c << 8);
 316     } while (--len);
 317     c = ~c;
 318     return (unsigned long)(ZSWAP32(c));
 319 }
 320
 321 #endif /* BYFOUR */
 322
 323 #define GF2_DIM 32      /* dimension of GF(2) vectors (length of CRC) */
 324
 325 /* ========================================================================= */
 326 local unsigned long gf2_matrix_times(mat, vec)
 327     unsigned long *mat;
 328     unsigned long vec;
 329 {
 330     unsigned long sum;
 331
 332     sum = 0;
 333     while (vec) {
 334         if (vec & 1)
 335             sum ^= *mat;
 336         vec >>= 1;
 337         mat++;
 338     }
 339     return sum;
 340 }
 341
 342 /* ========================================================================= */
 343 local void gf2_matrix_square(square, mat)
 344     unsigned long *square;
 345     unsigned long *mat;
 346 {
 347     int n;
 348
 349     for (n = 0; n < GF2_DIM; n++)
 350         square[n] = gf2_matrix_times(mat, mat[n]);
 351 }
 352
 353 /* ========================================================================= */
 354 local uLong crc32_combine_(crc1, crc2, len2)
 355     uLong crc1;
 356     uLong crc2;
 357     z_off64_t len2;
 358 {
 359     int n;
 360     unsigned long row;
 361     unsigned long even[GF2_DIM];    /* even-power-of-two zeros operator */
 362     unsigned long odd[GF2_DIM];     /* odd-power-of-two zeros operator */
 363
 364     /* degenerate case (also disallow negative lengths) */
 365     if (len2 <= 0)
 366         return crc1;
 367
 368     /* put operator for one zero bit in odd */
 369     odd[0] = 0xedb88320UL;          /* CRC-32 polynomial */
 370     row = 1;
 371     for (n = 1; n < GF2_DIM; n++) {
 372         odd[n] = row;
 373         row <<= 1;
 374     }
 375
 376     /* put operator for two zero bits in even */
 377     gf2_matrix_square(even, odd);
 378
 379     /* put operator for four zero bits in odd */
 380     gf2_matrix_square(odd, even);
 381
 382     /* apply len2 zeros to crc1 (first square will put the operator for one
 383        zero byte, eight zero bits, in even) */
 384     do {
 385         /* apply zeros operator for this bit of len2 */
 386         gf2_matrix_square(even, odd);
 387         if (len2 & 1)
 388             crc1 = gf2_matrix_times(even, crc1);
 389         len2 >>= 1;
 390
 391         /* if no more bits set, then done */
 392         if (len2 == 0)
 393             break;
 394
 395         /* another iteration of the loop with odd and even swapped */
 396         gf2_matrix_square(odd, even);
 397         if (len2 & 1)
 398             crc1 = gf2_matrix_times(odd, crc1);
 399         len2 >>= 1;
 400
 401         /* if no more bits set, then done */
 402     } while (len2 != 0);
 403
 404     /* return combined crc */
 405     crc1 ^= crc2;
 406     return crc1;
 407 }
 408
 409 /* ========================================================================= */
 410 uLong ZEXPORT crc32_combine(crc1, crc2, len2)
 411     uLong crc1;
 412     uLong crc2;
 413     z_off_t len2;
 414 {
 415     return crc32_combine_(crc1, crc2, len2);
 416 }
 417
 418 uLong ZEXPORT crc32_combine64(crc1, crc2, len2)
 419     uLong crc1;
 420     uLong crc2;
 421     z_off64_t len2;
 422 {
 423     return crc32_combine_(crc1, crc2, len2);
 424 }