Added hash-mode: AES-128/192/256-ECB NOKDF

This mode is probably very rare in real-life scenarios,
but it is a nice template for kernels which do
not use a KDF,
or use AES,
or simple fast hashes with lookup tables
or simple optimized kernels in general

OpenCL/m26401_a0-optimized.cl

@@ -0,0 +1,297 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_rp_optimized.h"
+#include "inc_rp_optimized.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+KERNEL_FQ void m26401_m04 (KERN_ATTR_RULES ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * modifier
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    u32 w0[4] = { 0 };
+    u32 w1[4] = { 0 };
+
+    // ignore output length
+    apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    u32 ukey[4];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+
+    #define KEYLEN 44
+
+    u32 ks[KEYLEN];
+
+    aes128_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes128_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32x r0 = ct[0];
+    const u32x r1 = ct[1];
+    const u32x r2 = ct[2];
+    const u32x r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26401_m08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26401_m16 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26401_s04 (KERN_ATTR_RULES ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * modifier
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    u32 w0[4] = { 0 };
+    u32 w1[4] = { 0 };
+
+    // ignore output length
+    apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    u32 ukey[4];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+
+    #define KEYLEN 44
+
+    u32 ks[KEYLEN];
+
+    aes128_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes128_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32x r0 = ct[0];
+    const u32x r1 = ct[1];
+    const u32x r2 = ct[2];
+    const u32x r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26401_s08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26401_s16 (KERN_ATTR_RULES ())
+{
+}

OpenCL/m26401_a1-optimized.cl

@@ -0,0 +1,402 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_scalar.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+KERNEL_FQ void m26401_m04 (KERN_ATTR_BASIC ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * 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) & 15;
+
+    const u32x pw_len = (pw_l_len + pw_r_len) & 15;
+
+    /**
+     * 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];
+
+    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);
+
+    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] = 0;
+    w1[1] = 0;
+    w1[2] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 ukey[4];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+
+    #define KEYLEN 44
+
+    u32 ks[KEYLEN];
+
+    aes128_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes128_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26401_m08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26401_m16 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26401_s04 (KERN_ATTR_BASIC ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * 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) & 15;
+
+    const u32x pw_len = (pw_l_len + pw_r_len) & 15;
+
+    /**
+     * 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];
+
+    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);
+
+    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] = 0;
+    w1[1] = 0;
+    w1[2] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 ukey[4];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+
+    #define KEYLEN 44
+
+    u32 ks[KEYLEN];
+
+    aes128_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes128_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26401_s08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26401_s16 (KERN_ATTR_BASIC ())
+{
+}

OpenCL/m26401_a3-optimized.cl

@@ -0,0 +1,515 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+DECLSPEC void m26401m (SHM_TYPE u32a *s_te0, SHM_TYPE u32a *s_te1, SHM_TYPE u32a *s_te2, SHM_TYPE u32a *s_te3, SHM_TYPE u32a *s_te4, SHM_TYPE u32a *s_td0, SHM_TYPE u32a *s_td1, SHM_TYPE u32a *s_td2, SHM_TYPE u32a *s_td3, SHM_TYPE u32a *s_td4, u32 *w, const u32 pw_len, KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * loop
+   */
+
+  u32 w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    u32 ukey[4];
+
+    ukey[0] = w0;
+    ukey[1] = w[1];
+    ukey[2] = w[2];
+    ukey[3] = w[3];
+
+    #define KEYLEN 44
+
+    u32 ks[KEYLEN];
+
+    aes128_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes128_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+DECLSPEC void m26401s (SHM_TYPE u32a *s_te0, SHM_TYPE u32a *s_te1, SHM_TYPE u32a *s_te2, SHM_TYPE u32a *s_te3, SHM_TYPE u32a *s_te4, SHM_TYPE u32a *s_td0, SHM_TYPE u32a *s_td1, SHM_TYPE u32a *s_td2, SHM_TYPE u32a *s_td3, SHM_TYPE u32a *s_td4, u32 *w, const u32 pw_len, KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  u32 w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    u32 ukey[4];
+
+    ukey[0] = w0;
+    ukey[1] = w[1];
+    ukey[2] = w[2];
+    ukey[3] = w[3];
+
+    #define KEYLEN 44
+
+    u32 ks[KEYLEN];
+
+    aes128_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes128_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26401_m04 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26401m (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26401_m08 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = pws[gid].i[ 4];
+  w[ 5] = pws[gid].i[ 5];
+  w[ 6] = pws[gid].i[ 6];
+  w[ 7] = pws[gid].i[ 7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26401m (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26401_m16 (KERN_ATTR_VECTOR ())
+{
+}
+
+KERNEL_FQ void m26401_s04 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26401s (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26401_s08 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = pws[gid].i[ 4];
+  w[ 5] = pws[gid].i[ 5];
+  w[ 6] = pws[gid].i[ 6];
+  w[ 7] = pws[gid].i[ 7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26401s (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26401_s16 (KERN_ATTR_VECTOR ())
+{
+}

OpenCL/m26402_a0-optimized.cl

@@ -0,0 +1,301 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_rp_optimized.h"
+#include "inc_rp_optimized.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+KERNEL_FQ void m26402_m04 (KERN_ATTR_RULES ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * modifier
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    u32 w0[4] = { 0 };
+    u32 w1[4] = { 0 };
+
+    // ignore output length
+    apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    u32 ukey[6];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+    ukey[4] = w1[0];
+    ukey[5] = w1[1];
+
+    #define KEYLEN 52
+
+    u32 ks[KEYLEN];
+
+    aes192_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes192_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32x r0 = ct[0];
+    const u32x r1 = ct[1];
+    const u32x r2 = ct[2];
+    const u32x r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26402_m08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26402_m16 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26402_s04 (KERN_ATTR_RULES ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * modifier
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    u32 w0[4] = { 0 };
+    u32 w1[4] = { 0 };
+
+    // ignore output length
+    apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    u32 ukey[6];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+    ukey[4] = w1[0];
+    ukey[5] = w1[1];
+
+    #define KEYLEN 52
+
+    u32 ks[KEYLEN];
+
+    aes192_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes192_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32x r0 = ct[0];
+    const u32x r1 = ct[1];
+    const u32x r2 = ct[2];
+    const u32x r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26402_s08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26402_s16 (KERN_ATTR_RULES ())
+{
+}

OpenCL/m26402_a1-optimized.cl

@@ -0,0 +1,414 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_scalar.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+KERNEL_FQ void m26402_m04 (KERN_ATTR_BASIC ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * 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) & 15;
+
+    const u32x pw_len = (pw_l_len + pw_r_len) & 15;
+
+    /**
+     * 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];
+
+    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);
+
+    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] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 ukey[6];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+    ukey[4] = w1[0];
+    ukey[5] = w1[1];
+
+    #define KEYLEN 52
+
+    u32 ks[KEYLEN];
+
+    aes192_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes192_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26402_m08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26402_m16 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26402_s04 (KERN_ATTR_BASIC ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * 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) & 15;
+
+    const u32x pw_len = (pw_l_len + pw_r_len) & 15;
+
+    /**
+     * 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];
+
+    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);
+
+    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] = 0;
+    w1[3] = 0;
+    w2[0] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 ukey[6];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+    ukey[4] = w1[0];
+    ukey[5] = w1[1];
+
+    #define KEYLEN 52
+
+    u32 ks[KEYLEN];
+
+    aes192_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes192_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26402_s08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26402_s16 (KERN_ATTR_BASIC ())
+{
+}

OpenCL/m26402_a3-optimized.cl

@@ -0,0 +1,519 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+DECLSPEC void m26402m (SHM_TYPE u32a *s_te0, SHM_TYPE u32a *s_te1, SHM_TYPE u32a *s_te2, SHM_TYPE u32a *s_te3, SHM_TYPE u32a *s_te4, SHM_TYPE u32a *s_td0, SHM_TYPE u32a *s_td1, SHM_TYPE u32a *s_td2, SHM_TYPE u32a *s_td3, SHM_TYPE u32a *s_td4, u32 *w, const u32 pw_len, KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * loop
+   */
+
+  u32 w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    u32 ukey[6];
+
+    ukey[0] = w0;
+    ukey[1] = w[1];
+    ukey[2] = w[2];
+    ukey[3] = w[3];
+    ukey[4] = w[4];
+    ukey[5] = w[5];
+
+    #define KEYLEN 52
+
+    u32 ks[KEYLEN];
+
+    aes192_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes192_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+DECLSPEC void m26402s (SHM_TYPE u32a *s_te0, SHM_TYPE u32a *s_te1, SHM_TYPE u32a *s_te2, SHM_TYPE u32a *s_te3, SHM_TYPE u32a *s_te4, SHM_TYPE u32a *s_td0, SHM_TYPE u32a *s_td1, SHM_TYPE u32a *s_td2, SHM_TYPE u32a *s_td3, SHM_TYPE u32a *s_td4, u32 *w, const u32 pw_len, KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  u32 w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    u32 ukey[6];
+
+    ukey[0] = w0;
+    ukey[1] = w[1];
+    ukey[2] = w[2];
+    ukey[3] = w[3];
+    ukey[4] = w[4];
+    ukey[5] = w[5];
+
+    #define KEYLEN 52
+
+    u32 ks[KEYLEN];
+
+    aes192_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes192_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26402_m04 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26402m (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26402_m08 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = pws[gid].i[ 4];
+  w[ 5] = pws[gid].i[ 5];
+  w[ 6] = pws[gid].i[ 6];
+  w[ 7] = pws[gid].i[ 7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26402m (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26402_m16 (KERN_ATTR_VECTOR ())
+{
+}
+
+KERNEL_FQ void m26402_s04 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26402s (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26402_s08 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = pws[gid].i[ 4];
+  w[ 5] = pws[gid].i[ 5];
+  w[ 6] = pws[gid].i[ 6];
+  w[ 7] = pws[gid].i[ 7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26402s (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26402_s16 (KERN_ATTR_VECTOR ())
+{
+}

OpenCL/m26403_a0-optimized.cl

@@ -0,0 +1,305 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_rp_optimized.h"
+#include "inc_rp_optimized.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+KERNEL_FQ void m26403_m04 (KERN_ATTR_RULES ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * modifier
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    u32 w0[4] = { 0 };
+    u32 w1[4] = { 0 };
+
+    // ignore output length
+    apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    u32 ukey[8];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+    ukey[4] = w1[0];
+    ukey[5] = w1[1];
+    ukey[6] = w1[2];
+    ukey[7] = w1[3];
+
+    #define KEYLEN 60
+
+    u32 ks[KEYLEN];
+
+    aes256_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes256_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32x r0 = ct[0];
+    const u32x r1 = ct[1];
+    const u32x r2 = ct[2];
+    const u32x r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26403_m08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26403_m16 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26403_s04 (KERN_ATTR_RULES ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * modifier
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    u32 w0[4] = { 0 };
+    u32 w1[4] = { 0 };
+
+    // ignore output length
+    apply_rules_vect_optimized (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
+
+    u32 ukey[8];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+    ukey[4] = w1[0];
+    ukey[5] = w1[1];
+    ukey[6] = w1[2];
+    ukey[7] = w1[3];
+
+    #define KEYLEN 60
+
+    u32 ks[KEYLEN];
+
+    aes256_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes256_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32x r0 = ct[0];
+    const u32x r1 = ct[1];
+    const u32x r2 = ct[2];
+    const u32x r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26403_s08 (KERN_ATTR_RULES ())
+{
+}
+
+KERNEL_FQ void m26403_s16 (KERN_ATTR_RULES ())
+{
+}

OpenCL/m26403_a1-optimized.cl

@@ -0,0 +1,426 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_scalar.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+KERNEL_FQ void m26403_m04 (KERN_ATTR_BASIC ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * 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) & 15;
+
+    const u32x pw_len = (pw_l_len + pw_r_len) & 15;
+
+    /**
+     * 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] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 ukey[8];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+    ukey[4] = w1[0];
+    ukey[5] = w1[1];
+    ukey[6] = w1[2];
+    ukey[7] = w1[3];
+
+    #define KEYLEN 60
+
+    u32 ks[KEYLEN];
+
+    aes256_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes256_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26403_m08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26403_m16 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26403_s04 (KERN_ATTR_BASIC ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  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 & 63;
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * 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) & 15;
+
+    const u32x pw_len = (pw_l_len + pw_r_len) & 15;
+
+    /**
+     * 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] = 0;
+    w2[1] = 0;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 0;
+    w3[3] = 0;
+
+    u32 ukey[8];
+
+    ukey[0] = w0[0];
+    ukey[1] = w0[1];
+    ukey[2] = w0[2];
+    ukey[3] = w0[3];
+    ukey[4] = w1[0];
+    ukey[5] = w1[1];
+    ukey[6] = w1[2];
+    ukey[7] = w1[3];
+
+    #define KEYLEN 60
+
+    u32 ks[KEYLEN];
+
+    aes256_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes256_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26403_s08 (KERN_ATTR_BASIC ())
+{
+}
+
+KERNEL_FQ void m26403_s16 (KERN_ATTR_BASIC ())
+{
+}

OpenCL/m26403_a3-optimized.cl

@@ -0,0 +1,697 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
+#include "inc_cipher_aes.cl"
+#endif
+
+DECLSPEC void m26403m (SHM_TYPE u32a *s_te0, SHM_TYPE u32a *s_te1, SHM_TYPE u32a *s_te2, SHM_TYPE u32a *s_te3, SHM_TYPE u32a *s_te4, SHM_TYPE u32a *s_td0, SHM_TYPE u32a *s_td1, SHM_TYPE u32a *s_td2, SHM_TYPE u32a *s_td3, SHM_TYPE u32a *s_td4, u32 *w, const u32 pw_len, KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * loop
+   */
+
+  u32 w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    u32 ukey[8];
+
+    ukey[0] = w0;
+    ukey[1] = w[1];
+    ukey[2] = w[2];
+    ukey[3] = w[3];
+    ukey[4] = w[4];
+    ukey[5] = w[5];
+    ukey[6] = w[6];
+    ukey[7] = w[7];
+
+    #define KEYLEN 60
+
+    u32 ks[KEYLEN];
+
+    aes256_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes256_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+DECLSPEC void m26403s (SHM_TYPE u32a *s_te0, SHM_TYPE u32a *s_te1, SHM_TYPE u32a *s_te2, SHM_TYPE u32a *s_te3, SHM_TYPE u32a *s_te4, SHM_TYPE u32a *s_td0, SHM_TYPE u32a *s_td1, SHM_TYPE u32a *s_td2, SHM_TYPE u32a *s_td3, SHM_TYPE u32a *s_td4, u32 *w, const u32 pw_len, KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+
+  /**
+   * Salt prep
+   */
+
+  u32 pt[4];
+
+  pt[0] = salt_bufs[SALT_POS].salt_buf[0];
+  pt[1] = salt_bufs[SALT_POS].salt_buf[1];
+  pt[2] = salt_bufs[SALT_POS].salt_buf[2];
+  pt[3] = salt_bufs[SALT_POS].salt_buf[3];
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R1],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R2],
+    digests_buf[DIGESTS_OFFSET].digest_buf[DGST_R3]
+  };
+
+  /**
+   * loop
+   */
+
+  u32 w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    u32 ukey[8];
+
+    ukey[0] = w0;
+    ukey[1] = w[1];
+    ukey[2] = w[2];
+    ukey[3] = w[3];
+    ukey[4] = w[4];
+    ukey[5] = w[5];
+    ukey[6] = w[6];
+    ukey[7] = w[7];
+
+    #define KEYLEN 60
+
+    u32 ks[KEYLEN];
+
+    aes256_set_encrypt_key (ks, ukey, s_te0, s_te1, s_te2, s_te3);
+
+    u32 ct[4];
+
+    aes256_encrypt (ks, pt, ct, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+    const u32 r0 = ct[0];
+    const u32 r1 = ct[1];
+    const u32 r2 = ct[2];
+    const u32 r3 = ct[3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26403_m04 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26403m (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26403_m08 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = pws[gid].i[ 4];
+  w[ 5] = pws[gid].i[ 5];
+  w[ 6] = pws[gid].i[ 6];
+  w[ 7] = pws[gid].i[ 7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26403m (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26403_m16 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = pws[gid].i[ 4];
+  w[ 5] = pws[gid].i[ 5];
+  w[ 6] = pws[gid].i[ 6];
+  w[ 7] = pws[gid].i[ 7];
+  w[ 8] = pws[gid].i[ 8];
+  w[ 9] = pws[gid].i[ 9];
+  w[10] = pws[gid].i[10];
+  w[11] = pws[gid].i[11];
+  w[12] = pws[gid].i[12];
+  w[13] = pws[gid].i[13];
+  w[14] = pws[gid].i[14];
+  w[15] = pws[gid].i[15];
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26403m (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26403_s04 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = 0;
+  w[ 5] = 0;
+  w[ 6] = 0;
+  w[ 7] = 0;
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26403s (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26403_s08 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = pws[gid].i[ 4];
+  w[ 5] = pws[gid].i[ 5];
+  w[ 6] = pws[gid].i[ 6];
+  w[ 7] = pws[gid].i[ 7];
+  w[ 8] = 0;
+  w[ 9] = 0;
+  w[10] = 0;
+  w[11] = 0;
+  w[12] = 0;
+  w[13] = 0;
+  w[14] = 0;
+  w[15] = 0;
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26403s (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}
+
+KERNEL_FQ void m26403_s16 (KERN_ATTR_VECTOR ())
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  /**
+   * aes shared
+   */
+
+  #ifdef REAL_SHM
+
+  LOCAL_VK u32 s_td0[256];
+  LOCAL_VK u32 s_td1[256];
+  LOCAL_VK u32 s_td2[256];
+  LOCAL_VK u32 s_td3[256];
+  LOCAL_VK u32 s_td4[256];
+
+  LOCAL_VK u32 s_te0[256];
+  LOCAL_VK u32 s_te1[256];
+  LOCAL_VK u32 s_te2[256];
+  LOCAL_VK u32 s_te3[256];
+  LOCAL_VK u32 s_te4[256];
+
+  for (u32 i = lid; i < 256; i += lsz)
+  {
+    s_td0[i] = td0[i];
+    s_td1[i] = td1[i];
+    s_td2[i] = td2[i];
+    s_td3[i] = td3[i];
+    s_td4[i] = td4[i];
+
+    s_te0[i] = te0[i];
+    s_te1[i] = te1[i];
+    s_te2[i] = te2[i];
+    s_te3[i] = te3[i];
+    s_te4[i] = te4[i];
+  }
+
+  SYNC_THREADS ();
+
+  #else
+
+  CONSTANT_AS u32a *s_td0 = td0;
+  CONSTANT_AS u32a *s_td1 = td1;
+  CONSTANT_AS u32a *s_td2 = td2;
+  CONSTANT_AS u32a *s_td3 = td3;
+  CONSTANT_AS u32a *s_td4 = td4;
+
+  CONSTANT_AS u32a *s_te0 = te0;
+  CONSTANT_AS u32a *s_te1 = te1;
+  CONSTANT_AS u32a *s_te2 = te2;
+  CONSTANT_AS u32a *s_te3 = te3;
+  CONSTANT_AS u32a *s_te4 = te4;
+
+  #endif
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 w[16];
+
+  w[ 0] = pws[gid].i[ 0];
+  w[ 1] = pws[gid].i[ 1];
+  w[ 2] = pws[gid].i[ 2];
+  w[ 3] = pws[gid].i[ 3];
+  w[ 4] = pws[gid].i[ 4];
+  w[ 5] = pws[gid].i[ 5];
+  w[ 6] = pws[gid].i[ 6];
+  w[ 7] = pws[gid].i[ 7];
+  w[ 8] = pws[gid].i[ 8];
+  w[ 9] = pws[gid].i[ 9];
+  w[10] = pws[gid].i[10];
+  w[11] = pws[gid].i[11];
+  w[12] = pws[gid].i[12];
+  w[13] = pws[gid].i[13];
+  w[14] = pws[gid].i[14];
+  w[15] = pws[gid].i[15];
+
+  const u32 pw_len = pws[gid].pw_len & 63;
+
+  /**
+   * main
+   */
+
+  m26403s (s_te0, s_te1, s_te2, s_te3, s_te4, s_td0, s_td1, s_td2, s_td3, s_td4, w, pw_len, pws, rules_buf, combs_buf, words_buf_r, tmps, hooks, bitmaps_buf_s1_a, bitmaps_buf_s1_b, bitmaps_buf_s1_c, bitmaps_buf_s1_d, bitmaps_buf_s2_a, bitmaps_buf_s2_b, bitmaps_buf_s2_c, bitmaps_buf_s2_d, plains_buf, digests_buf, hashes_shown, salt_bufs, esalt_bufs, d_return_buf, d_extra0_buf, d_extra1_buf, d_extra2_buf, d_extra3_buf, bitmap_mask, bitmap_shift1, bitmap_shift2, SALT_POS, loop_pos, loop_cnt, il_cnt, digests_cnt, DIGESTS_OFFSET, combs_mode, salt_repeat, pws_pos, gid_max);
+}

docs/changes.txt

@@ -7,6 +7,9 @@
 - Added hash-mode: VeraCrypt PBKDF2-HMAC-Streebog-512 + XTS 512 bit + boot-mode
 - Added hash-mode: VeraCrypt PBKDF2-HMAC-Streebog-512 + XTS 1024 bit + boot-mode
 - Added hash-mode: VeraCrypt PBKDF2-HMAC-Streebog-512 + XTS 1536 bit + boot-mode
+- Added hash-mode: AES-128-ECB NOKDF (PT = $salt, key = $pass)
+- Added hash-mode: AES-192-ECB NOKDF (PT = $salt, key = $pass)
+- Added hash-mode: AES-256-ECB NOKDF (PT = $salt, key = $pass)
 
 ##
 ## Bugs

docs/readme.txt

@@ -137,11 +137,14 @@ NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or
 - CRC32
 - Java Object hashCode()
 - MurmurHash
-- 3DES (PT = $salt, key = $pass)
+- 3DES
-- DES (PT = $salt, key = $pass)
+- DES
+- AES-128-ECB NOKDF
+- AES-192-ECB NOKDF
+- AES-256-ECB NOKDF
 - ChaCha20
 - Linux Kernel Crypto API (2.4)
-- Skip32 (PT = $salt, key = $pass)
+- Skip32
 - PBKDF2-HMAC-MD5
 - PBKDF2-HMAC-SHA1
 - PBKDF2-HMAC-SHA256

src/backend.c

@@ -3008,6 +3008,8 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
 
     if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
     {
+      // this is not perfectly right, only in case algorithm has to add 0x80 (most of the cases for fast optimized kernels)
+
       if (highest_pw_len < 16)
       {
         if (run_kernel (hashcat_ctx, device_param, KERN_RUN_1, pws_pos, pws_cnt, true, fast_iteration) == -1) return -1;

src/modules/module_26401.c

@@ -0,0 +1,204 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "emu_inc_cipher_des.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_INSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_RAW_CIPHER_KPA;
+static const char *HASH_NAME      = "AES-128-ECB NOKDF (PT = $salt, key = $pass)";
+static const u64   KERN_TYPE      = 26401;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_ST_HEX;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat";
+static const char *ST_HASH        = "e7a32f3210455cc044f26117c4612aab:86046627772965328523223752173724";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 0;  // Underlaying AES-128 min
+
+  return pw_min;
+}
+
+u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_max = 16; // Underlaying AES-128 max
+
+  return pw_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  token_t token;
+
+  token.token_cnt = 2;
+
+  token.sep[0]     = hashconfig->separator;
+  token.len_min[0] = 32;
+  token.len_max[0] = 32;
+  token.attr[0]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  token.len_min[1] = 32;
+  token.len_max[1] = 32;
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[1];
+
+  salt->salt_buf[0] = hex_to_u32 (salt_pos +  0);
+  salt->salt_buf[1] = hex_to_u32 (salt_pos +  8);
+  salt->salt_buf[2] = hex_to_u32 (salt_pos + 16);
+  salt->salt_buf[3] = hex_to_u32 (salt_pos + 24);
+
+  salt->salt_len = 16;
+
+  // hash
+
+  const u8 *hash_pos = token.buf[0];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = 0;
+
+  u32_to_hex (digest[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[3], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = hashconfig->separator;
+
+  out_len++;
+
+  u32_to_hex (salt->salt_buf[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[3], out_buf + out_len); out_len += 8;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = MODULE_DEFAULT;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = MODULE_DEFAULT;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}

src/modules/module_26402.c

@@ -0,0 +1,204 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "emu_inc_cipher_des.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_INSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_RAW_CIPHER_KPA;
+static const char *HASH_NAME      = "AES-192-ECB NOKDF (PT = $salt, key = $pass)";
+static const u64   KERN_TYPE      = 26402;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_ST_HEX;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat";
+static const char *ST_HASH        = "2995e91b798ef51232a91579edb1d176:49869364034411376791729962721320";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 0;  // Underlaying AES-192 min
+
+  return pw_min;
+}
+
+u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_max = 24; // Underlaying AES-192 max
+
+  return pw_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  token_t token;
+
+  token.token_cnt = 2;
+
+  token.sep[0]     = hashconfig->separator;
+  token.len_min[0] = 32;
+  token.len_max[0] = 32;
+  token.attr[0]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  token.len_min[1] = 32;
+  token.len_max[1] = 32;
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[1];
+
+  salt->salt_buf[0] = hex_to_u32 (salt_pos +  0);
+  salt->salt_buf[1] = hex_to_u32 (salt_pos +  8);
+  salt->salt_buf[2] = hex_to_u32 (salt_pos + 16);
+  salt->salt_buf[3] = hex_to_u32 (salt_pos + 24);
+
+  salt->salt_len = 16;
+
+  // hash
+
+  const u8 *hash_pos = token.buf[0];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = 0;
+
+  u32_to_hex (digest[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[3], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = hashconfig->separator;
+
+  out_len++;
+
+  u32_to_hex (salt->salt_buf[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[3], out_buf + out_len); out_len += 8;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = MODULE_DEFAULT;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = MODULE_DEFAULT;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}

src/modules/module_26403.c

@@ -0,0 +1,204 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "modules.h"
+#include "bitops.h"
+#include "convert.h"
+#include "shared.h"
+#include "emu_inc_cipher_des.h"
+
+static const u32   ATTACK_EXEC    = ATTACK_EXEC_INSIDE_KERNEL;
+static const u32   DGST_POS0      = 0;
+static const u32   DGST_POS1      = 1;
+static const u32   DGST_POS2      = 2;
+static const u32   DGST_POS3      = 3;
+static const u32   DGST_SIZE      = DGST_SIZE_4_4;
+static const u32   HASH_CATEGORY  = HASH_CATEGORY_RAW_CIPHER_KPA;
+static const char *HASH_NAME      = "AES-256-ECB NOKDF (PT = $salt, key = $pass)";
+static const u64   KERN_TYPE      = 26403;
+static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
+static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
+                                  | OPTS_TYPE_ST_HEX;
+static const u32   SALT_TYPE      = SALT_TYPE_EMBEDDED;
+static const char *ST_PASS        = "hashcat";
+static const char *ST_HASH        = "264a4248c9522cb74d33fe26cb596895:61270210011294880287232432636227";
+
+u32         module_attack_exec    (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ATTACK_EXEC;     }
+u32         module_dgst_pos0      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS0;       }
+u32         module_dgst_pos1      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS1;       }
+u32         module_dgst_pos2      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS2;       }
+u32         module_dgst_pos3      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_POS3;       }
+u32         module_dgst_size      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return DGST_SIZE;       }
+u32         module_hash_category  (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_CATEGORY;   }
+const char *module_hash_name      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return HASH_NAME;       }
+u64         module_kern_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return KERN_TYPE;       }
+u32         module_opti_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTI_TYPE;       }
+u64         module_opts_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return OPTS_TYPE;       }
+u32         module_salt_type      (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return SALT_TYPE;       }
+const char *module_st_hash        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_HASH;         }
+const char *module_st_pass        (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra) { return ST_PASS;         }
+
+u32 module_pw_min (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_min = 0;  // Underlaying AES-256 min
+
+  return pw_min;
+}
+
+u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 pw_max = 32; // Underlaying AES-256 max
+
+  return pw_max;
+}
+
+int module_hash_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED void *digest_buf, MAYBE_UNUSED salt_t *salt, MAYBE_UNUSED void *esalt_buf, MAYBE_UNUSED void *hook_salt_buf, MAYBE_UNUSED hashinfo_t *hash_info, const char *line_buf, MAYBE_UNUSED const int line_len)
+{
+  u32 *digest = (u32 *) digest_buf;
+
+  token_t token;
+
+  token.token_cnt = 2;
+
+  token.sep[0]     = hashconfig->separator;
+  token.len_min[0] = 32;
+  token.len_max[0] = 32;
+  token.attr[0]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  token.len_min[1] = 32;
+  token.len_max[1] = 32;
+  token.attr[1]    = TOKEN_ATTR_VERIFY_LENGTH
+                   | TOKEN_ATTR_VERIFY_HEX;
+
+  const int rc_tokenizer = input_tokenizer ((const u8 *) line_buf, line_len, &token);
+
+  if (rc_tokenizer != PARSER_OK) return (rc_tokenizer);
+
+  // salt
+
+  const u8 *salt_pos = token.buf[1];
+
+  salt->salt_buf[0] = hex_to_u32 (salt_pos +  0);
+  salt->salt_buf[1] = hex_to_u32 (salt_pos +  8);
+  salt->salt_buf[2] = hex_to_u32 (salt_pos + 16);
+  salt->salt_buf[3] = hex_to_u32 (salt_pos + 24);
+
+  salt->salt_len = 16;
+
+  // hash
+
+  const u8 *hash_pos = token.buf[0];
+
+  digest[0] = hex_to_u32 (hash_pos +  0);
+  digest[1] = hex_to_u32 (hash_pos +  8);
+  digest[2] = hex_to_u32 (hash_pos + 16);
+  digest[3] = hex_to_u32 (hash_pos + 24);
+
+  return (PARSER_OK);
+}
+
+int module_hash_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const void *digest_buf, MAYBE_UNUSED const salt_t *salt, MAYBE_UNUSED const void *esalt_buf, MAYBE_UNUSED const void *hook_salt_buf, MAYBE_UNUSED const hashinfo_t *hash_info, char *line_buf, MAYBE_UNUSED const int line_size)
+{
+  const u32 *digest = (const u32 *) digest_buf;
+
+  u8 *out_buf = (u8 *) line_buf;
+
+  int out_len = 0;
+
+  u32_to_hex (digest[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (digest[3], out_buf + out_len); out_len += 8;
+
+  out_buf[out_len] = hashconfig->separator;
+
+  out_len++;
+
+  u32_to_hex (salt->salt_buf[0], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[1], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[2], out_buf + out_len); out_len += 8;
+  u32_to_hex (salt->salt_buf[3], out_buf + out_len); out_len += 8;
+
+  return out_len;
+}
+
+void module_init (module_ctx_t *module_ctx)
+{
+  module_ctx->module_context_size             = MODULE_CONTEXT_SIZE_CURRENT;
+  module_ctx->module_interface_version        = MODULE_INTERFACE_VERSION_CURRENT;
+
+  module_ctx->module_attack_exec              = module_attack_exec;
+  module_ctx->module_benchmark_esalt          = MODULE_DEFAULT;
+  module_ctx->module_benchmark_hook_salt      = MODULE_DEFAULT;
+  module_ctx->module_benchmark_mask           = MODULE_DEFAULT;
+  module_ctx->module_benchmark_salt           = MODULE_DEFAULT;
+  module_ctx->module_build_plain_postprocess  = MODULE_DEFAULT;
+  module_ctx->module_deep_comp_kernel         = MODULE_DEFAULT;
+  module_ctx->module_dgst_pos0                = module_dgst_pos0;
+  module_ctx->module_dgst_pos1                = module_dgst_pos1;
+  module_ctx->module_dgst_pos2                = module_dgst_pos2;
+  module_ctx->module_dgst_pos3                = module_dgst_pos3;
+  module_ctx->module_dgst_size                = module_dgst_size;
+  module_ctx->module_dictstat_disable         = MODULE_DEFAULT;
+  module_ctx->module_esalt_size               = MODULE_DEFAULT;
+  module_ctx->module_extra_buffer_size        = MODULE_DEFAULT;
+  module_ctx->module_extra_tmp_size           = MODULE_DEFAULT;
+  module_ctx->module_forced_outfile_format    = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_count        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_parse        = MODULE_DEFAULT;
+  module_ctx->module_hash_binary_save         = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_decode_zero_hash    = MODULE_DEFAULT;
+  module_ctx->module_hash_decode              = module_hash_decode;
+  module_ctx->module_hash_encode_status       = MODULE_DEFAULT;
+  module_ctx->module_hash_encode_potfile      = MODULE_DEFAULT;
+  module_ctx->module_hash_encode              = module_hash_encode;
+  module_ctx->module_hash_init_selftest       = MODULE_DEFAULT;
+  module_ctx->module_hash_mode                = MODULE_DEFAULT;
+  module_ctx->module_hash_category            = module_hash_category;
+  module_ctx->module_hash_name                = module_hash_name;
+  module_ctx->module_hashes_count_min         = MODULE_DEFAULT;
+  module_ctx->module_hashes_count_max         = MODULE_DEFAULT;
+  module_ctx->module_hlfmt_disable            = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_size    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_init    = MODULE_DEFAULT;
+  module_ctx->module_hook_extra_param_term    = MODULE_DEFAULT;
+  module_ctx->module_hook12                   = MODULE_DEFAULT;
+  module_ctx->module_hook23                   = MODULE_DEFAULT;
+  module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
+  module_ctx->module_hook_size                = MODULE_DEFAULT;
+  module_ctx->module_jit_build_options        = MODULE_DEFAULT;
+  module_ctx->module_jit_cache_disable        = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_max         = MODULE_DEFAULT;
+  module_ctx->module_kernel_loops_min         = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
+  module_ctx->module_kern_type                = module_kern_type;
+  module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
+  module_ctx->module_opti_type                = module_opti_type;
+  module_ctx->module_opts_type                = module_opts_type;
+  module_ctx->module_outfile_check_disable    = MODULE_DEFAULT;
+  module_ctx->module_outfile_check_nocomp     = MODULE_DEFAULT;
+  module_ctx->module_potfile_custom_check     = MODULE_DEFAULT;
+  module_ctx->module_potfile_disable          = MODULE_DEFAULT;
+  module_ctx->module_potfile_keep_all_hashes  = MODULE_DEFAULT;
+  module_ctx->module_pwdump_column            = MODULE_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
+  module_ctx->module_pw_min                   = module_pw_min;
+  module_ctx->module_salt_max                 = MODULE_DEFAULT;
+  module_ctx->module_salt_min                 = MODULE_DEFAULT;
+  module_ctx->module_salt_type                = module_salt_type;
+  module_ctx->module_separator                = MODULE_DEFAULT;
+  module_ctx->module_st_hash                  = module_st_hash;
+  module_ctx->module_st_pass                  = module_st_pass;
+  module_ctx->module_tmp_size                 = MODULE_DEFAULT;
+  module_ctx->module_unstable_warning         = MODULE_DEFAULT;
+  module_ctx->module_warmup_disable           = MODULE_DEFAULT;
+}

tools/test_modules/m26401.pm

@@ -0,0 +1,53 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Crypt::Mode::ECB;
+
+sub module_constraints { [[0, 16], [32, 32], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+
+  my $cipher = Crypt::Mode::ECB->new ('AES', 0);
+
+  my $key_bin = $word;
+  my $pt_bin  = pack ("H*", $salt);
+
+  $key_bin .= "\x00" x 16;
+
+  $key_bin = substr ($key_bin, 0, 16);
+
+  my $ct_bin = $cipher->encrypt ($pt_bin, $key_bin);
+
+  my $hash = sprintf ("%s:%s", unpack ("H*", $ct_bin), $salt);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my ($hash, $salt, $word) = split (':', $line);
+
+  return unless defined $hash;
+  return unless defined $salt;
+  return unless defined $word;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt);
+
+  return ($new_hash, $word);
+}
+
+1;

tools/test_modules/m26402.pm

@@ -0,0 +1,53 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Crypt::Mode::ECB;
+
+sub module_constraints { [[0, 24], [32, 32], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+
+  my $cipher = Crypt::Mode::ECB->new ('AES', 0);
+
+  my $key_bin = $word;
+  my $pt_bin  = pack ("H*", $salt);
+
+  $key_bin .= "\x00" x 24;
+
+  $key_bin = substr ($key_bin, 0, 24);
+
+  my $ct_bin = $cipher->encrypt ($pt_bin, $key_bin);
+
+  my $hash = sprintf ("%s:%s", unpack ("H*", $ct_bin), $salt);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my ($hash, $salt, $word) = split (':', $line);
+
+  return unless defined $hash;
+  return unless defined $salt;
+  return unless defined $word;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt);
+
+  return ($new_hash, $word);
+}
+
+1;

tools/test_modules/m26403.pm

@@ -0,0 +1,53 @@
+#!/usr/bin/env perl
+
+##
+## Author......: See docs/credits.txt
+## License.....: MIT
+##
+
+use strict;
+use warnings;
+
+use Crypt::Mode::ECB;
+
+sub module_constraints { [[0, 32], [32, 32], [-1, -1], [-1, -1], [-1, -1]] }
+
+sub module_generate_hash
+{
+  my $word = shift;
+  my $salt = shift;
+
+  my $cipher = Crypt::Mode::ECB->new ('AES', 0);
+
+  my $key_bin = $word;
+  my $pt_bin  = pack ("H*", $salt);
+
+  $key_bin .= "\x00" x 32;
+
+  $key_bin = substr ($key_bin, 0, 32);
+
+  my $ct_bin = $cipher->encrypt ($pt_bin, $key_bin);
+
+  my $hash = sprintf ("%s:%s", unpack ("H*", $ct_bin), $salt);
+
+  return $hash;
+}
+
+sub module_verify_hash
+{
+  my $line = shift;
+
+  my ($hash, $salt, $word) = split (':', $line);
+
+  return unless defined $hash;
+  return unless defined $salt;
+  return unless defined $word;
+
+  my $word_packed = pack_if_HEX_notation ($word);
+
+  my $new_hash = module_generate_hash ($word_packed, $salt);
+
+  return ($new_hash, $word);
+}
+
+1;