/**
 * Author......: See docs/credits.txt
 * License.....: MIT
 */

//#define NEW_SIMD_CODE

#ifdef KERNEL_STATIC
#include M2S(INCLUDE_PATH/inc_vendor.h)
#include M2S(INCLUDE_PATH/inc_types.h)
#include M2S(INCLUDE_PATH/inc_platform.cl)
#include M2S(INCLUDE_PATH/inc_common.cl)
#include M2S(INCLUDE_PATH/inc_rp_optimized.h)
#include M2S(INCLUDE_PATH/inc_rp_optimized.cl)
#include M2S(INCLUDE_PATH/inc_simd.cl)
#endif

DECLSPEC u32 MurmurHash64A_truncated (PRIVATE_AS const u32 *data, const u32 len)
{
#define M 0xc6a4a7935bd1e995
#define R 47

  // Initialize hash
  u64 hash = len * M;

  // Twice the number of u64 blocks
  const u32 num_u32_blocks = (len / 8) * 2;

  // Loop over one u64 at a time
  u32 i = 0;
  while (i < num_u32_blocks)
  {
    // Reconstruct u64 from two u32s
    u64 k = hl32_to_64 (data[i + 1], data[i]);

    k *= M;
    k ^= k >> R;
    k *= M;

    hash ^= k;
    hash *= M;

    i += 2;
  }

  // Up to 7 overflow bytes
  const u32 overflow = len & 7;

  if (overflow > 4)
  {
    hash ^= hl32_to_64 (data[i + 1], data[i]);
    hash *= M;
  }
  else if (overflow > 0)
  {
    hash ^= hl32_to_64 (0, data[i]);
    hash *= M;
  }

  hash ^= hash >> R;
  hash *= M;
  hash ^= hash >> R;

#undef M
#undef R

  // Truncate to high 4 bytes
  return (u32) (hash >> 32);
}

KERNEL_FQ KERNEL_FA void m34211_m04 (KERN_ATTR_RULES ())
{
  /**
   * modifier
   */

  const u64 lid = get_local_id (0);

  /**
   * base
   */

  const u64 gid = get_global_id (0);

  if (gid >= GID_CNT) return;

  u32 pw_buf0[4];
  u32 pw_buf1[4];

  pw_buf0[0] = pws[gid].i[0];
  pw_buf0[1] = pws[gid].i[1];
  pw_buf0[2] = pws[gid].i[2];
  pw_buf0[3] = pws[gid].i[3];
  pw_buf1[0] = pws[gid].i[4];
  pw_buf1[1] = pws[gid].i[5];
  pw_buf1[2] = pws[gid].i[6];
  pw_buf1[3] = pws[gid].i[7];

  const u32 pw_len = pws[gid].pw_len;

  /**
   * loop
   */

  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos += VECT_SIZE)
  {
    u32x w[16] = { 0 };

    const u32x out_len = apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w + 0, w + 4);

    u32x hash = MurmurHash64A_truncated (w, out_len);

    const u32x z = 0;

    COMPARE_M_SIMD (hash, z, z, z);
  }
}

KERNEL_FQ KERNEL_FA void m34211_m08 (KERN_ATTR_RULES ())
{
}

KERNEL_FQ KERNEL_FA void m34211_m16 (KERN_ATTR_RULES ())
{
}

KERNEL_FQ KERNEL_FA void m34211_s04 (KERN_ATTR_RULES ())
{
  /**
   * modifier
   */

  const u64 lid = get_local_id (0);

  /**
   * base
   */

  const u64 gid = get_global_id (0);

  if (gid >= GID_CNT) return;

  u32 pw_buf0[4];
  u32 pw_buf1[4];

  pw_buf0[0] = pws[gid].i[0];
  pw_buf0[1] = pws[gid].i[1];
  pw_buf0[2] = pws[gid].i[2];
  pw_buf0[3] = pws[gid].i[3];
  pw_buf1[0] = pws[gid].i[4];
  pw_buf1[1] = pws[gid].i[5];
  pw_buf1[2] = pws[gid].i[6];
  pw_buf1[3] = pws[gid].i[7];

  const u32 pw_len = pws[gid].pw_len;

  /**
   * digest
   */

  const u32 search[4] =
  {
    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
    0,
    0,
    0
  };

  /**
   * loop
   */

  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos += VECT_SIZE)
  {
    u32x w[16] = { 0 };

    const u32x out_len = apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w + 0, w + 4);

    u32x hash = MurmurHash64A_truncated (w, out_len);

    const u32x z = 0;

    COMPARE_S_SIMD (hash, z, z, z);
  }
}

KERNEL_FQ KERNEL_FA void m34211_s08 (KERN_ATTR_RULES ())
{
}

KERNEL_FQ KERNEL_FA void m34211_s16 (KERN_ATTR_RULES ())
{
}