hashcat/OpenCL/m16100_a1-optimized.cl

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

//#define NEW_SIMD_CODE

#include "inc_vendor.cl"
#include "inc_hash_constants.h"
#include "inc_hash_functions.cl"
#include "inc_types.cl"
#include "inc_common.cl"
#include "inc_simd.cl"

__kernel void m16100_m04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const tacacs_plus_t *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
{
  /**
   * modifier
   */

  const u64 lid = get_local_id (0);

  /**
   * base
   */

  const u64 gid = get_global_id (0);

  if (gid >= gid_max) 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_l_len = pws[gid].pw_len;

  /**
   * salt
   */

  const u32 session_buf  = esalt_bufs[digests_offset].session_buf[0];
  const u32 sequence_buf = esalt_bufs[digests_offset].sequence_buf[0];

  /**
   * digest
   */

  const u32 ct_len = esalt_bufs[digests_offset].ct_data_len;

  u32 ct_buf[2];

  ct_buf[0] = esalt_bufs[digests_offset].ct_data_buf[0];
  ct_buf[1] = esalt_bufs[digests_offset].ct_data_buf[1];

  /**
   * loop
   */

  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
  {
    const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);

    const u32x pw_len = pw_l_len + pw_r_len;

    /**
     * concat password candidate
     */

    u32x wordl0[4] = { 0 };
    u32x wordl1[4] = { 0 };
    u32x wordl2[4] = { 0 };
    u32x wordl3[4] = { 0 };

    wordl0[0] = pw_buf0[0];
    wordl0[1] = pw_buf0[1];
    wordl0[2] = pw_buf0[2];
    wordl0[3] = pw_buf0[3];
    wordl1[0] = pw_buf1[0];
    wordl1[1] = pw_buf1[1];
    wordl1[2] = pw_buf1[2];
    wordl1[3] = pw_buf1[3];

    u32x wordr0[4] = { 0 };
    u32x wordr1[4] = { 0 };
    u32x wordr2[4] = { 0 };
    u32x wordr3[4] = { 0 };

    wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
    wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
    wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
    wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
    wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
    wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
    wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
    wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);

    if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
    {
      switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
    }
    else
    {
      switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
    }

    u32x w0[4];
    u32x w1[4];
    u32x w2[4];
    u32x w3[4];

    w0[0] = wordl0[0] | wordr0[0];
    w0[1] = wordl0[1] | wordr0[1];
    w0[2] = wordl0[2] | wordr0[2];
    w0[3] = wordl0[3] | wordr0[3];
    w1[0] = wordl1[0] | wordr1[0];
    w1[1] = wordl1[1] | wordr1[1];
    w1[2] = wordl1[2] | wordr1[2];
    w1[3] = wordl1[3] | wordr1[3];
    w2[0] = wordl2[0] | wordr2[0];
    w2[1] = wordl2[1] | wordr2[1];
    w2[2] = wordl2[2] | wordr2[2];
    w2[3] = wordl2[3] | wordr2[3];
    w3[0] = wordl3[0] | wordr3[0];
    w3[1] = wordl3[1] | wordr3[1];
    w3[2] = wordl3[2] | wordr3[2];
    w3[3] = wordl3[3] | wordr3[3];

    /**
     * append salt
     */

    u32 s0[4];
    u32 s1[4];
    u32 s2[4];
    u32 s3[4];

    s0[0] = sequence_buf | 0x00800000;
    s0[1] = 0;
    s0[2] = 0;
    s0[3] = 0;
    s1[0] = 0;
    s1[1] = 0;
    s1[2] = 0;
    s1[3] = 0;
    s2[0] = 0;
    s2[1] = 0;
    s2[2] = 0;
    s2[3] = 0;
    s3[0] = 0;
    s3[1] = 0;
    s3[2] = 0;
    s3[3] = 0;

    switch_buffer_by_offset_le (s0, s1, s2, s3, 4 + pw_len);

    const u32x pw_salt_len = 4 + pw_len + 2;

    s0[0]  = session_buf;
    s0[1] |= w0[0];
    s0[2] |= w0[1];
    s0[3] |= w0[2];
    s1[0] |= w0[3];
    s1[1] |= w1[0];
    s1[2] |= w1[1];
    s1[3] |= w1[2];
    s2[0] |= w1[3];
    s2[1] |= 0;
    s2[2] |= 0;
    s2[3] |= 0;
    s3[0] |= 0;
    s3[1] |= 0;
    s3[2]  = pw_salt_len * 8;
    s3[3]  = 0;

    w0[0] = s0[0];
    w0[1] = s0[1];
    w0[2] = s0[2];
    w0[3] = s0[3];
    w1[0] = s1[0];
    w1[1] = s1[1];
    w1[2] = s1[2];
    w1[3] = s1[3];
    w2[0] = s2[0];
    w2[1] = s2[1];
    w2[2] = s2[2];
    w2[3] = s2[3];
    w3[0] = s3[0];
    w3[1] = s3[1];
    w3[2] = s3[2];
    w3[3] = s3[3];

    /**
     * md5
     */

    u32x a = MD5M_A;
    u32x b = MD5M_B;
    u32x c = MD5M_C;
    u32x d = MD5M_D;

    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);

    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);

    u32x t;

    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);

    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);


    a += MD5M_A;
    b += MD5M_B;
    c += MD5M_C;
    d += MD5M_D;

    u32 test[2];

    test[0] = a ^ ct_buf[0];
    test[1] = b ^ ct_buf[1];

    if (sequence_buf == 0x01c0)
    {
      const u32 action          = ((test[0] >>  0) & 0xff);
      // can have more than predefined ones
      // const u32 priv_lvl        = ((test[0] >>  8) & 0xff);
      const u32 authen_type     = ((test[0] >> 16) & 0xff);
      const u32 authen_service  = ((test[0] >> 24) & 0xff);
      const u32 user_len        = ((test[1] >>  0) & 0xff);
      const u32 port_len        = ((test[1] >>  8) & 0xff);
      const u32 rem_addr_len    = ((test[1] >> 16) & 0xff);
      const u32 data_len        = ((test[1] >> 24) & 0xff);

      if (((action == 0x01) || (action == 0x02) || (action == 0x04))
      &&  ((authen_type    >= 0x01) && (authen_type    <= 0x06))
      &&  ((authen_service >= 0x00) && (authen_service <= 0x09))
      &&  ((8 + user_len + port_len + rem_addr_len + data_len) == ct_len))
      {
        if (atomic_inc (&hashes_shown[digests_offset]) == 0)
        {
          mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, 0, digests_offset + 0, gid, il_pos);
        }
      }
    }
    else if ((sequence_buf == 0x03c0) || (sequence_buf == 0x05c0))
    {
      const u32 msg_len   = ((test[0] >>  0) & 0xff) << 8
                          | ((test[0] >>  8) & 0xff) << 0;
      const u32 data_len  = ((test[0] >> 16) & 0xff) << 8
                          | ((test[0] >> 24) & 0xff) << 0;
      const u32 flags     = ((test[1] >>  0) & 0xff);

      if (((5 + msg_len) == ct_len)
       &&  (data_len == 0)
       &&  (flags == 0))
      {
        if (atomic_inc (&hashes_shown[digests_offset]) == 0)
        {
          mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, 0, digests_offset + 0, gid, il_pos);
        }
      }
    }
    else
    {
      const u32 status    = ((test[0] >>  0) & 0xff);
      const u32 flags     = ((test[0] >>  8) & 0xff);
      const u32 msg_len   = ((test[0] >> 16) & 0xff) << 8
                          | ((test[0] >> 24) & 0xff) << 0;
      const u32 data_len  = ((test[1] >>  0) & 0xff) << 8
                          | ((test[1] >>  8) & 0xff) << 0;

      if (((status >= 0x01 && status <= 0x07) || status == 0x21)
       &&  (flags == 0x01 || flags == 0x00)
       &&  (6 + msg_len + data_len == ct_len))
      {
        if (atomic_inc (&hashes_shown[digests_offset]) == 0)
        {
          mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, 0, digests_offset + 0, gid, il_pos);
        }
      }
    }
  }
}

__kernel void m16100_m08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const tacacs_plus_t *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
{
}

__kernel void m16100_m16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const tacacs_plus_t *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
{
}

__kernel void m16100_s04 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const tacacs_plus_t *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
{
  /**
   * modifier
   */

  const u64 lid = get_local_id (0);

  /**
   * base
   */

  const u64 gid = get_global_id (0);

  if (gid >= gid_max) 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_l_len = pws[gid].pw_len;

  /**
   * salt
   */

  const u32 session_buf  = esalt_bufs[digests_offset].session_buf[0];
  const u32 sequence_buf = esalt_bufs[digests_offset].sequence_buf[0];

  /**
   * digest
   */

  const u32 ct_len = esalt_bufs[digests_offset].ct_data_len;

  u32 ct_buf[2];

  ct_buf[0] = esalt_bufs[digests_offset].ct_data_buf[0];
  ct_buf[1] = esalt_bufs[digests_offset].ct_data_buf[1];

  /**
   * loop
   */

  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
  {
    const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);

    const u32x pw_len = pw_l_len + pw_r_len;

    /**
     * concat password candidate
     */

    u32x wordl0[4] = { 0 };
    u32x wordl1[4] = { 0 };
    u32x wordl2[4] = { 0 };
    u32x wordl3[4] = { 0 };

    wordl0[0] = pw_buf0[0];
    wordl0[1] = pw_buf0[1];
    wordl0[2] = pw_buf0[2];
    wordl0[3] = pw_buf0[3];
    wordl1[0] = pw_buf1[0];
    wordl1[1] = pw_buf1[1];
    wordl1[2] = pw_buf1[2];
    wordl1[3] = pw_buf1[3];

    u32x wordr0[4] = { 0 };
    u32x wordr1[4] = { 0 };
    u32x wordr2[4] = { 0 };
    u32x wordr3[4] = { 0 };

    wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
    wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
    wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
    wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
    wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
    wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
    wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
    wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);

    if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
    {
      switch_buffer_by_offset_le_VV (wordr0, wordr1, wordr2, wordr3, pw_l_len);
    }
    else
    {
      switch_buffer_by_offset_le_VV (wordl0, wordl1, wordl2, wordl3, pw_r_len);
    }

    u32x w0[4];
    u32x w1[4];
    u32x w2[4];
    u32x w3[4];

    w0[0] = wordl0[0] | wordr0[0];
    w0[1] = wordl0[1] | wordr0[1];
    w0[2] = wordl0[2] | wordr0[2];
    w0[3] = wordl0[3] | wordr0[3];
    w1[0] = wordl1[0] | wordr1[0];
    w1[1] = wordl1[1] | wordr1[1];
    w1[2] = wordl1[2] | wordr1[2];
    w1[3] = wordl1[3] | wordr1[3];
    w2[0] = wordl2[0] | wordr2[0];
    w2[1] = wordl2[1] | wordr2[1];
    w2[2] = wordl2[2] | wordr2[2];
    w2[3] = wordl2[3] | wordr2[3];
    w3[0] = wordl3[0] | wordr3[0];
    w3[1] = wordl3[1] | wordr3[1];
    w3[2] = wordl3[2] | wordr3[2];
    w3[3] = wordl3[3] | wordr3[3];

    /**
     * append salt
     */

    u32 s0[4];
    u32 s1[4];
    u32 s2[4];
    u32 s3[4];

    s0[0] = sequence_buf | 0x00800000;
    s0[1] = 0;
    s0[2] = 0;
    s0[3] = 0;
    s1[0] = 0;
    s1[1] = 0;
    s1[2] = 0;
    s1[3] = 0;
    s2[0] = 0;
    s2[1] = 0;
    s2[2] = 0;
    s2[3] = 0;
    s3[0] = 0;
    s3[1] = 0;
    s3[2] = 0;
    s3[3] = 0;

    switch_buffer_by_offset_le (s0, s1, s2, s3, 4 + pw_len);

    const u32x pw_salt_len = 4 + pw_len + 2;

    s0[0]  = session_buf;
    s0[1] |= w0[0];
    s0[2] |= w0[1];
    s0[3] |= w0[2];
    s1[0] |= w0[3];
    s1[1] |= w1[0];
    s1[2] |= w1[1];
    s1[3] |= w1[2];
    s2[0] |= w1[3];
    s2[1] |= 0;
    s2[2] |= 0;
    s2[3] |= 0;
    s3[0] |= 0;
    s3[1] |= 0;
    s3[2]  = pw_salt_len * 8;
    s3[3]  = 0;

    w0[0] = s0[0];
    w0[1] = s0[1];
    w0[2] = s0[2];
    w0[3] = s0[3];
    w1[0] = s1[0];
    w1[1] = s1[1];
    w1[2] = s1[2];
    w1[3] = s1[3];
    w2[0] = s2[0];
    w2[1] = s2[1];
    w2[2] = s2[2];
    w2[3] = s2[3];
    w3[0] = s3[0];
    w3[1] = s3[1];
    w3[2] = s3[2];
    w3[3] = s3[3];

    /**
     * md5
     */

    u32x a = MD5M_A;
    u32x b = MD5M_B;
    u32x c = MD5M_C;
    u32x d = MD5M_D;

    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);

    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);

    u32x t;

    MD5_STEP (MD5_H1, a, b, c, d, w1[1], MD5C20, MD5S20);
    MD5_STEP (MD5_H2, d, a, b, c, w2[0], MD5C21, MD5S21);
    MD5_STEP (MD5_H1, c, d, a, b, w2[3], MD5C22, MD5S22);
    MD5_STEP (MD5_H2, b, c, d, a, w3[2], MD5C23, MD5S23);
    MD5_STEP (MD5_H1, a, b, c, d, w0[1], MD5C24, MD5S20);
    MD5_STEP (MD5_H2, d, a, b, c, w1[0], MD5C25, MD5S21);
    MD5_STEP (MD5_H1, c, d, a, b, w1[3], MD5C26, MD5S22);
    MD5_STEP (MD5_H2, b, c, d, a, w2[2], MD5C27, MD5S23);
    MD5_STEP (MD5_H1, a, b, c, d, w3[1], MD5C28, MD5S20);
    MD5_STEP (MD5_H2, d, a, b, c, w0[0], MD5C29, MD5S21);
    MD5_STEP (MD5_H1, c, d, a, b, w0[3], MD5C2a, MD5S22);
    MD5_STEP (MD5_H2, b, c, d, a, w1[2], MD5C2b, MD5S23);
    MD5_STEP (MD5_H1, a, b, c, d, w2[1], MD5C2c, MD5S20);
    MD5_STEP (MD5_H2, d, a, b, c, w3[0], MD5C2d, MD5S21);
    MD5_STEP (MD5_H1, c, d, a, b, w3[3], MD5C2e, MD5S22);
    MD5_STEP (MD5_H2, b, c, d, a, w0[2], MD5C2f, MD5S23);

    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);


    a += MD5M_A;
    b += MD5M_B;
    c += MD5M_C;
    d += MD5M_D;

    u32 test[2];

    test[0] = a ^ ct_buf[0];
    test[1] = b ^ ct_buf[1];

    if (sequence_buf == 0x01c0)
    {
      const u32 action          = ((test[0] >>  0) & 0xff);
      // can have more than predefined ones
      // const u32 priv_lvl        = ((test[0] >>  8) & 0xff);
      const u32 authen_type     = ((test[0] >> 16) & 0xff);
      const u32 authen_service  = ((test[0] >> 24) & 0xff);
      const u32 user_len        = ((test[1] >>  0) & 0xff);
      const u32 port_len        = ((test[1] >>  8) & 0xff);
      const u32 rem_addr_len    = ((test[1] >> 16) & 0xff);
      const u32 data_len        = ((test[1] >> 24) & 0xff);

      if (((action == 0x01) || (action == 0x02) || (action == 0x04))
      &&  ((authen_type    >= 0x01) && (authen_type    <= 0x06))
      &&  ((authen_service >= 0x00) && (authen_service <= 0x09))
      &&  ((8 + user_len + port_len + rem_addr_len + data_len) == ct_len))
      {
        if (atomic_inc (&hashes_shown[digests_offset]) == 0)
        {
          mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, 0, digests_offset + 0, gid, il_pos);
        }
      }
    }
    else if ((sequence_buf == 0x03c0) || (sequence_buf == 0x05c0))
    {
      const u32 msg_len   = ((test[0] >>  0) & 0xff) << 8
                          | ((test[0] >>  8) & 0xff) << 0;
      const u32 data_len  = ((test[0] >> 16) & 0xff) << 8
                          | ((test[0] >> 24) & 0xff) << 0;
      const u32 flags     = ((test[1] >>  0) & 0xff);

      if (((5 + msg_len) == ct_len)
       &&  (data_len == 0)
       &&  (flags == 0))
      {
        if (atomic_inc (&hashes_shown[digests_offset]) == 0)
        {
          mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, 0, digests_offset + 0, gid, il_pos);
        }
      }
    }
    else
    {
      const u32 status    = ((test[0] >>  0) & 0xff);
      const u32 flags     = ((test[0] >>  8) & 0xff);
      const u32 msg_len   = ((test[0] >> 16) & 0xff) << 8
                          | ((test[0] >> 24) & 0xff) << 0;
      const u32 data_len  = ((test[1] >>  0) & 0xff) << 8
                          | ((test[1] >>  8) & 0xff) << 0;

      if (((status >= 0x01 && status <= 0x07) || status == 0x21)
       &&  (flags == 0x01 || flags == 0x00)
       &&  (6 + msg_len + data_len == ct_len))
      {
        if (atomic_inc (&hashes_shown[digests_offset]) == 0)
        {
          mark_hash (plains_buf, d_return_buf, salt_pos, digests_cnt, 0, digests_offset + 0, gid, il_pos);
        }
      }
    }
  }
}

__kernel void m16100_s08 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const tacacs_plus_t *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
{
}

__kernel void m16100_s16 (__global pw_t *pws, __global const kernel_rule_t *rules_buf, __global const pw_t *combs_buf, __global const bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global const u32 *bitmaps_buf_s1_a, __global const u32 *bitmaps_buf_s1_b, __global const u32 *bitmaps_buf_s1_c, __global const u32 *bitmaps_buf_s1_d, __global const u32 *bitmaps_buf_s2_a, __global const u32 *bitmaps_buf_s2_b, __global const u32 *bitmaps_buf_s2_c, __global const u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global const digest_t *digests_buf, __global u32 *hashes_shown, __global const salt_t *salt_bufs, __global const tacacs_plus_t *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u64 gid_max)
{
}