michael@0: /* crc32.c -- compute the CRC-32 of a data stream
michael@0:  * Copyright (C) 1995-2006, 2010 Mark Adler
michael@0:  * For conditions of distribution and use, see copyright notice in zlib.h
michael@0:  *
michael@0:  * Thanks to Rodney Brown <rbrown64@csc.com.au> for his contribution of faster
michael@0:  * CRC methods: exclusive-oring 32 bits of data at a time, and pre-computing
michael@0:  * tables for updating the shift register in one step with three exclusive-ors
michael@0:  * instead of four steps with four exclusive-ors.  This results in about a
michael@0:  * factor of two increase in speed on a Power PC G4 (PPC7455) using gcc -O3.
michael@0:  */
michael@0: 
michael@0: /* @(#) $Id$ */
michael@0: 
michael@0: /*
michael@0:   Note on the use of DYNAMIC_CRC_TABLE: there is no mutex or semaphore
michael@0:   protection on the static variables used to control the first-use generation
michael@0:   of the crc tables.  Therefore, if you #define DYNAMIC_CRC_TABLE, you should
michael@0:   first call get_crc_table() to initialize the tables before allowing more than
michael@0:   one thread to use crc32().
michael@0:  */
michael@0: 
michael@0: #ifdef MAKECRCH
michael@0: #  include <stdio.h>
michael@0: #  ifndef DYNAMIC_CRC_TABLE
michael@0: #    define DYNAMIC_CRC_TABLE
michael@0: #  endif /* !DYNAMIC_CRC_TABLE */
michael@0: #endif /* MAKECRCH */
michael@0: 
michael@0: #include "zutil.h"      /* for STDC and FAR definitions */
michael@0: 
michael@0: #define local static
michael@0: 
michael@0: /* Find a four-byte integer type for crc32_little() and crc32_big(). */
michael@0: #ifndef NOBYFOUR
michael@0: #  ifdef STDC           /* need ANSI C limits.h to determine sizes */
michael@0: #    include <limits.h>
michael@0: #    define BYFOUR
michael@0: #    if (UINT_MAX == 0xffffffffUL)
michael@0:        typedef unsigned int u4;
michael@0: #    else
michael@0: #      if (ULONG_MAX == 0xffffffffUL)
michael@0:          typedef unsigned long u4;
michael@0: #      else
michael@0: #        if (USHRT_MAX == 0xffffffffUL)
michael@0:            typedef unsigned short u4;
michael@0: #        else
michael@0: #          undef BYFOUR     /* can't find a four-byte integer type! */
michael@0: #        endif
michael@0: #      endif
michael@0: #    endif
michael@0: #  endif /* STDC */
michael@0: #endif /* !NOBYFOUR */
michael@0: 
michael@0: /* Definitions for doing the crc four data bytes at a time. */
michael@0: #ifdef BYFOUR
michael@0: #  define REV(w) ((((w)>>24)&0xff)+(((w)>>8)&0xff00)+ \
michael@0:                 (((w)&0xff00)<<8)+(((w)&0xff)<<24))
michael@0:    local unsigned long crc32_little OF((unsigned long,
michael@0:                         const unsigned char FAR *, unsigned));
michael@0:    local unsigned long crc32_big OF((unsigned long,
michael@0:                         const unsigned char FAR *, unsigned));
michael@0: #  define TBLS 8
michael@0: #else
michael@0: #  define TBLS 1
michael@0: #endif /* BYFOUR */
michael@0: 
michael@0: /* Local functions for crc concatenation */
michael@0: local unsigned long gf2_matrix_times OF((unsigned long *mat,
michael@0:                                          unsigned long vec));
michael@0: local void gf2_matrix_square OF((unsigned long *square, unsigned long *mat));
michael@0: local uLong crc32_combine_(uLong crc1, uLong crc2, z_off64_t len2);
michael@0: 
michael@0: 
michael@0: #ifdef DYNAMIC_CRC_TABLE
michael@0: 
michael@0: local volatile int crc_table_empty = 1;
michael@0: local unsigned long FAR crc_table[TBLS][256];
michael@0: local void make_crc_table OF((void));
michael@0: #ifdef MAKECRCH
michael@0:    local void write_table OF((FILE *, const unsigned long FAR *));
michael@0: #endif /* MAKECRCH */
michael@0: /*
michael@0:   Generate tables for a byte-wise 32-bit CRC calculation on the polynomial:
michael@0:   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.
michael@0: 
michael@0:   Polynomials over GF(2) are represented in binary, one bit per coefficient,
michael@0:   with the lowest powers in the most significant bit.  Then adding polynomials
michael@0:   is just exclusive-or, and multiplying a polynomial by x is a right shift by
michael@0:   one.  If we call the above polynomial p, and represent a byte as the
michael@0:   polynomial q, also with the lowest power in the most significant bit (so the
michael@0:   byte 0xb1 is the polynomial x^7+x^3+x+1), then the CRC is (q*x^32) mod p,
michael@0:   where a mod b means the remainder after dividing a by b.
michael@0: 
michael@0:   This calculation is done using the shift-register method of multiplying and
michael@0:   taking the remainder.  The register is initialized to zero, and for each
michael@0:   incoming bit, x^32 is added mod p to the register if the bit is a one (where
michael@0:   x^32 mod p is p+x^32 = x^26+...+1), and the register is multiplied mod p by
michael@0:   x (which is shifting right by one and adding x^32 mod p if the bit shifted
michael@0:   out is a one).  We start with the highest power (least significant bit) of
michael@0:   q and repeat for all eight bits of q.
michael@0: 
michael@0:   The first table is simply the CRC of all possible eight bit values.  This is
michael@0:   all the information needed to generate CRCs on data a byte at a time for all
michael@0:   combinations of CRC register values and incoming bytes.  The remaining tables
michael@0:   allow for word-at-a-time CRC calculation for both big-endian and little-
michael@0:   endian machines, where a word is four bytes.
michael@0: */
michael@0: local void make_crc_table()
michael@0: {
michael@0:     unsigned long c;
michael@0:     int n, k;
michael@0:     unsigned long poly;                 /* polynomial exclusive-or pattern */
michael@0:     /* terms of polynomial defining this crc (except x^32): */
michael@0:     static volatile int first = 1;      /* flag to limit concurrent making */
michael@0:     static const unsigned char p[] = {0,1,2,4,5,7,8,10,11,12,16,22,23,26};
michael@0: 
michael@0:     /* See if another task is already doing this (not thread-safe, but better
michael@0:        than nothing -- significantly reduces duration of vulnerability in
michael@0:        case the advice about DYNAMIC_CRC_TABLE is ignored) */
michael@0:     if (first) {
michael@0:         first = 0;
michael@0: 
michael@0:         /* make exclusive-or pattern from polynomial (0xedb88320UL) */
michael@0:         poly = 0UL;
michael@0:         for (n = 0; n < sizeof(p)/sizeof(unsigned char); n++)
michael@0:             poly |= 1UL << (31 - p[n]);
michael@0: 
michael@0:         /* generate a crc for every 8-bit value */
michael@0:         for (n = 0; n < 256; n++) {
michael@0:             c = (unsigned long)n;
michael@0:             for (k = 0; k < 8; k++)
michael@0:                 c = c & 1 ? poly ^ (c >> 1) : c >> 1;
michael@0:             crc_table[0][n] = c;
michael@0:         }
michael@0: 
michael@0: #ifdef BYFOUR
michael@0:         /* generate crc for each value followed by one, two, and three zeros,
michael@0:            and then the byte reversal of those as well as the first table */
michael@0:         for (n = 0; n < 256; n++) {
michael@0:             c = crc_table[0][n];
michael@0:             crc_table[4][n] = REV(c);
michael@0:             for (k = 1; k < 4; k++) {
michael@0:                 c = crc_table[0][c & 0xff] ^ (c >> 8);
michael@0:                 crc_table[k][n] = c;
michael@0:                 crc_table[k + 4][n] = REV(c);
michael@0:             }
michael@0:         }
michael@0: #endif /* BYFOUR */
michael@0: 
michael@0:         crc_table_empty = 0;
michael@0:     }
michael@0:     else {      /* not first */
michael@0:         /* wait for the other guy to finish (not efficient, but rare) */
michael@0:         while (crc_table_empty)
michael@0:             ;
michael@0:     }
michael@0: 
michael@0: #ifdef MAKECRCH
michael@0:     /* write out CRC tables to crc32.h */
michael@0:     {
michael@0:         FILE *out;
michael@0: 
michael@0:         out = fopen("crc32.h", "w");
michael@0:         if (out == NULL) return;
michael@0:         fprintf(out, "/* crc32.h -- tables for rapid CRC calculation\n");
michael@0:         fprintf(out, " * Generated automatically by crc32.c\n */\n\n");
michael@0:         fprintf(out, "local const unsigned long FAR ");
michael@0:         fprintf(out, "crc_table[TBLS][256] =\n{\n  {\n");
michael@0:         write_table(out, crc_table[0]);
michael@0: #  ifdef BYFOUR
michael@0:         fprintf(out, "#ifdef BYFOUR\n");
michael@0:         for (k = 1; k < 8; k++) {
michael@0:             fprintf(out, "  },\n  {\n");
michael@0:             write_table(out, crc_table[k]);
michael@0:         }
michael@0:         fprintf(out, "#endif\n");
michael@0: #  endif /* BYFOUR */
michael@0:         fprintf(out, "  }\n};\n");
michael@0:         fclose(out);
michael@0:     }
michael@0: #endif /* MAKECRCH */
michael@0: }
michael@0: 
michael@0: #ifdef MAKECRCH
michael@0: local void write_table(out, table)
michael@0:     FILE *out;
michael@0:     const unsigned long FAR *table;
michael@0: {
michael@0:     int n;
michael@0: 
michael@0:     for (n = 0; n < 256; n++)
michael@0:         fprintf(out, "%s0x%08lxUL%s", n % 5 ? "" : "    ", table[n],
michael@0:                 n == 255 ? "\n" : (n % 5 == 4 ? ",\n" : ", "));
michael@0: }
michael@0: #endif /* MAKECRCH */
michael@0: 
michael@0: #else /* !DYNAMIC_CRC_TABLE */
michael@0: /* ========================================================================
michael@0:  * Tables of CRC-32s of all single-byte values, made by make_crc_table().
michael@0:  */
michael@0: #include "crc32.h"
michael@0: #endif /* DYNAMIC_CRC_TABLE */
michael@0: 
michael@0: /* =========================================================================
michael@0:  * This function can be used by asm versions of crc32()
michael@0:  */
michael@0: const unsigned long FAR * ZEXPORT get_crc_table()
michael@0: {
michael@0: #ifdef DYNAMIC_CRC_TABLE
michael@0:     if (crc_table_empty)
michael@0:         make_crc_table();
michael@0: #endif /* DYNAMIC_CRC_TABLE */
michael@0:     return (const unsigned long FAR *)crc_table;
michael@0: }
michael@0: 
michael@0: /* ========================================================================= */
michael@0: #define DO1 crc = crc_table[0][((int)crc ^ (*buf++)) & 0xff] ^ (crc >> 8)
michael@0: #define DO8 DO1; DO1; DO1; DO1; DO1; DO1; DO1; DO1
michael@0: 
michael@0: /* ========================================================================= */
michael@0: unsigned long ZEXPORT crc32(crc, buf, len)
michael@0:     unsigned long crc;
michael@0:     const unsigned char FAR *buf;
michael@0:     uInt len;
michael@0: {
michael@0:     if (buf == Z_NULL) return 0UL;
michael@0: 
michael@0: #ifdef DYNAMIC_CRC_TABLE
michael@0:     if (crc_table_empty)
michael@0:         make_crc_table();
michael@0: #endif /* DYNAMIC_CRC_TABLE */
michael@0: 
michael@0: #ifdef BYFOUR
michael@0:     if (sizeof(void *) == sizeof(ptrdiff_t)) {
michael@0:         u4 endian;
michael@0: 
michael@0:         endian = 1;
michael@0:         if (*((unsigned char *)(&endian)))
michael@0:             return crc32_little(crc, buf, len);
michael@0:         else
michael@0:             return crc32_big(crc, buf, len);
michael@0:     }
michael@0: #endif /* BYFOUR */
michael@0:     crc = crc ^ 0xffffffffUL;
michael@0:     while (len >= 8) {
michael@0:         DO8;
michael@0:         len -= 8;
michael@0:     }
michael@0:     if (len) do {
michael@0:         DO1;
michael@0:     } while (--len);
michael@0:     return crc ^ 0xffffffffUL;
michael@0: }
michael@0: 
michael@0: #ifdef BYFOUR
michael@0: 
michael@0: /* ========================================================================= */
michael@0: #define DOLIT4 c ^= *buf4++; \
michael@0:         c = crc_table[3][c & 0xff] ^ crc_table[2][(c >> 8) & 0xff] ^ \
michael@0:             crc_table[1][(c >> 16) & 0xff] ^ crc_table[0][c >> 24]
michael@0: #define DOLIT32 DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4; DOLIT4
michael@0: 
michael@0: /* ========================================================================= */
michael@0: local unsigned long crc32_little(crc, buf, len)
michael@0:     unsigned long crc;
michael@0:     const unsigned char FAR *buf;
michael@0:     unsigned len;
michael@0: {
michael@0:     register u4 c;
michael@0:     register const u4 FAR *buf4;
michael@0: 
michael@0:     c = (u4)crc;
michael@0:     c = ~c;
michael@0:     while (len && ((ptrdiff_t)buf & 3)) {
michael@0:         c = crc_table[0][(c ^ *buf++) & 0xff] ^ (c >> 8);
michael@0:         len--;
michael@0:     }
michael@0: 
michael@0:     buf4 = (const u4 FAR *)(const void FAR *)buf;
michael@0:     while (len >= 32) {
michael@0:         DOLIT32;
michael@0:         len -= 32;
michael@0:     }
michael@0:     while (len >= 4) {
michael@0:         DOLIT4;
michael@0:         len -= 4;
michael@0:     }
michael@0:     buf = (const unsigned char FAR *)buf4;
michael@0: 
michael@0:     if (len) do {
michael@0:         c = crc_table[0][(c ^ *buf++) & 0xff] ^ (c >> 8);
michael@0:     } while (--len);
michael@0:     c = ~c;
michael@0:     return (unsigned long)c;
michael@0: }
michael@0: 
michael@0: /* ========================================================================= */
michael@0: #define DOBIG4 c ^= *++buf4; \
michael@0:         c = crc_table[4][c & 0xff] ^ crc_table[5][(c >> 8) & 0xff] ^ \
michael@0:             crc_table[6][(c >> 16) & 0xff] ^ crc_table[7][c >> 24]
michael@0: #define DOBIG32 DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4; DOBIG4
michael@0: 
michael@0: /* ========================================================================= */
michael@0: local unsigned long crc32_big(crc, buf, len)
michael@0:     unsigned long crc;
michael@0:     const unsigned char FAR *buf;
michael@0:     unsigned len;
michael@0: {
michael@0:     register u4 c;
michael@0:     register const u4 FAR *buf4;
michael@0: 
michael@0:     c = REV((u4)crc);
michael@0:     c = ~c;
michael@0:     while (len && ((ptrdiff_t)buf & 3)) {
michael@0:         c = crc_table[4][(c >> 24) ^ *buf++] ^ (c << 8);
michael@0:         len--;
michael@0:     }
michael@0: 
michael@0:     buf4 = (const u4 FAR *)(const void FAR *)buf;
michael@0:     buf4--;
michael@0:     while (len >= 32) {
michael@0:         DOBIG32;
michael@0:         len -= 32;
michael@0:     }
michael@0:     while (len >= 4) {
michael@0:         DOBIG4;
michael@0:         len -= 4;
michael@0:     }
michael@0:     buf4++;
michael@0:     buf = (const unsigned char FAR *)buf4;
michael@0: 
michael@0:     if (len) do {
michael@0:         c = crc_table[4][(c >> 24) ^ *buf++] ^ (c << 8);
michael@0:     } while (--len);
michael@0:     c = ~c;
michael@0:     return (unsigned long)(REV(c));
michael@0: }
michael@0: 
michael@0: #endif /* BYFOUR */
michael@0: 
michael@0: #define GF2_DIM 32      /* dimension of GF(2) vectors (length of CRC) */
michael@0: 
michael@0: /* ========================================================================= */
michael@0: local unsigned long gf2_matrix_times(mat, vec)
michael@0:     unsigned long *mat;
michael@0:     unsigned long vec;
michael@0: {
michael@0:     unsigned long sum;
michael@0: 
michael@0:     sum = 0;
michael@0:     while (vec) {
michael@0:         if (vec & 1)
michael@0:             sum ^= *mat;
michael@0:         vec >>= 1;
michael@0:         mat++;
michael@0:     }
michael@0:     return sum;
michael@0: }
michael@0: 
michael@0: /* ========================================================================= */
michael@0: local void gf2_matrix_square(square, mat)
michael@0:     unsigned long *square;
michael@0:     unsigned long *mat;
michael@0: {
michael@0:     int n;
michael@0: 
michael@0:     for (n = 0; n < GF2_DIM; n++)
michael@0:         square[n] = gf2_matrix_times(mat, mat[n]);
michael@0: }
michael@0: 
michael@0: /* ========================================================================= */
michael@0: local uLong crc32_combine_(crc1, crc2, len2)
michael@0:     uLong crc1;
michael@0:     uLong crc2;
michael@0:     z_off64_t len2;
michael@0: {
michael@0:     int n;
michael@0:     unsigned long row;
michael@0:     unsigned long even[GF2_DIM];    /* even-power-of-two zeros operator */
michael@0:     unsigned long odd[GF2_DIM];     /* odd-power-of-two zeros operator */
michael@0: 
michael@0:     /* degenerate case (also disallow negative lengths) */
michael@0:     if (len2 <= 0)
michael@0:         return crc1;
michael@0: 
michael@0:     /* put operator for one zero bit in odd */
michael@0:     odd[0] = 0xedb88320UL;          /* CRC-32 polynomial */
michael@0:     row = 1;
michael@0:     for (n = 1; n < GF2_DIM; n++) {
michael@0:         odd[n] = row;
michael@0:         row <<= 1;
michael@0:     }
michael@0: 
michael@0:     /* put operator for two zero bits in even */
michael@0:     gf2_matrix_square(even, odd);
michael@0: 
michael@0:     /* put operator for four zero bits in odd */
michael@0:     gf2_matrix_square(odd, even);
michael@0: 
michael@0:     /* apply len2 zeros to crc1 (first square will put the operator for one
michael@0:        zero byte, eight zero bits, in even) */
michael@0:     do {
michael@0:         /* apply zeros operator for this bit of len2 */
michael@0:         gf2_matrix_square(even, odd);
michael@0:         if (len2 & 1)
michael@0:             crc1 = gf2_matrix_times(even, crc1);
michael@0:         len2 >>= 1;
michael@0: 
michael@0:         /* if no more bits set, then done */
michael@0:         if (len2 == 0)
michael@0:             break;
michael@0: 
michael@0:         /* another iteration of the loop with odd and even swapped */
michael@0:         gf2_matrix_square(odd, even);
michael@0:         if (len2 & 1)
michael@0:             crc1 = gf2_matrix_times(odd, crc1);
michael@0:         len2 >>= 1;
michael@0: 
michael@0:         /* if no more bits set, then done */
michael@0:     } while (len2 != 0);
michael@0: 
michael@0:     /* return combined crc */
michael@0:     crc1 ^= crc2;
michael@0:     return crc1;
michael@0: }
michael@0: 
michael@0: /* ========================================================================= */
michael@0: uLong ZEXPORT crc32_combine(crc1, crc2, len2)
michael@0:     uLong crc1;
michael@0:     uLong crc2;
michael@0:     z_off_t len2;
michael@0: {
michael@0:     return crc32_combine_(crc1, crc2, len2);
michael@0: }
michael@0: 
michael@0: uLong ZEXPORT crc32_combine64(crc1, crc2, len2)
michael@0:     uLong crc1;
michael@0:     uLong crc2;
michael@0:     z_off64_t len2;
michael@0: {
michael@0:     return crc32_combine_(crc1, crc2, len2);
michael@0: }