#    undef W
#    define W 4
     typedef Z_U4 z_word_t;
#  else
#    undef W
#  endif
#endif

/* If available, use the ARM processor CRC32 instruction. */
#if defined(__aarch64__) && defined(__ARM_FEATURE_CRC32) && W == 8
#  define ARMCRC32
#endif

/* Local functions. */
local z_crc_t multmodp OF((z_crc_t a, z_crc_t b));
local z_crc_t x2nmodp OF((z_off64_t n, unsigned k));

#if defined(W) && (!defined(ARMCRC32) || defined(DYNAMIC_CRC_TABLE))
    local z_word_t byte_swap OF((z_word_t word));
#endif

#if defined(W) && !defined(ARMCRC32)
    local z_crc_t crc_word OF((z_word_t data));
    local z_word_t crc_word_big OF((z_word_t data));
#endif

#if defined(W) && (!defined(ARMCRC32) || defined(DYNAMIC_CRC_TABLE))
/*
  Swap the bytes in a z_word_t to convert between little and big endian. Any
  self-respecting compiler will optimize this to a single machine byte-swap
  instruction, if one is available. This assumes that word_t is either 32 bits

    if (buf == Z_NULL) return 0;

#ifdef DYNAMIC_CRC_TABLE
    once(&made, make_crc_table);
#endif /* DYNAMIC_CRC_TABLE */

    /* Pre-condition the CRC */
    crc = (~crc) & 0xffffffff;

    /* Compute the CRC up to a word boundary. */
    while (len && ((z_size_t)buf & 7) != 0) {
        len--;
        val = *buf++;
        __asm__ volatile("crc32b %w0, %w0, %w1" : "+r"(crc) : "r"(val));
    }

    /* Prepare to compute the CRC on full 64-bit words word[0..num-1]. */
    word = (z_word_t const *)buf;
    num = len >> 3;
    len &= 7;

    /* Do three interleaved CRCs to realize the throughput of one crc32x
       instruction per cycle. Each CRC is calculated on Z_BATCH words. The
       three CRCs are combined into a single CRC after each set of batches. */
    while (num >= 3 * Z_BATCH) {
        crc1 = 0;
        crc2 = 0;
        for (i = 0; i < Z_BATCH; i++) {
            val0 = word[i];
            val1 = word[i + Z_BATCH];
            val2 = word[i + 2 * Z_BATCH];

    if (buf == Z_NULL) return 0;

#ifdef DYNAMIC_CRC_TABLE
    once(&made, make_crc_table);
#endif /* DYNAMIC_CRC_TABLE */

    /* Pre-condition the CRC */
    crc = (~crc) & 0xffffffff;

#ifdef W

    /* If provided enough bytes, do a braided CRC calculation. */
    if (len >= N * W + W - 1) {
        z_size_t blks;
        z_word_t const *words;

    uLong crc1;
    uLong crc2;
    z_off64_t len2;
{
#ifdef DYNAMIC_CRC_TABLE
    once(&made, make_crc_table);
#endif /* DYNAMIC_CRC_TABLE */
    return multmodp(x2nmodp(len2, 3), crc1) ^ (crc2 & 0xffffffff);
}

/* ========================================================================= */
uLong ZEXPORT crc32_combine(crc1, crc2, len2)
    uLong crc1;
    uLong crc2;
    z_off_t len2;
{
    return crc32_combine64(crc1, crc2, (z_off64_t)len2);
}

/* ========================================================================= */
uLong ZEXPORT crc32_combine_gen64(len2)
    z_off64_t len2;
{
#ifdef DYNAMIC_CRC_TABLE
    once(&made, make_crc_table);
#endif /* DYNAMIC_CRC_TABLE */
    return x2nmodp(len2, 3);
}

/* ========================================================================= */
uLong ZEXPORT crc32_combine_gen(len2)
    z_off_t len2;
{
    return crc32_combine_gen64((z_off64_t)len2);
}

/* ========================================================================= */
uLong ZEXPORT crc32_combine_op(crc1, crc2, op)
    uLong crc1;
    uLong crc2;
    uLong op;
{
    return multmodp(op, crc1) ^ (crc2 & 0xffffffff);
}