Merge branch 'master' of https://github.com/roycewilliams/hashcat

.github/ISSUE_TEMPLATE/feature_request_-_add_new_algorithm.md

@@ -7,10 +7,10 @@ assignees: ''
 
 ---
 
-**Is your request based on a publicly know cryptographic schema and where can we find information about?**
+**Is your request based on a publicly known cryptographic schema and where can we find information about?**
 For instance:
 - The hashing function XY is a public known hash algorithm on Wikipedia. 
-- The password protection for the application XY was made by public by security researcher XY and there's a paper called ...
+- The password protection for the application XY was made public by security researcher XY and there's a paper called ...
 - The cryptographic schema used in protocol XY can be found in the OSS code on it's GitHub repository.
 
 **Describe alternatives you've considered**

OpenCL/inc_common.cl

@@ -2627,34 +2627,15 @@ DECLSPEC int hc_find_keyboard_layout_map (const u32 search, const int search_len
   return -1;
 }
 
-DECLSPEC int hc_execute_keyboard_layout_mapping (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const int pw_len, LOCAL_AS keyboard_layout_mapping_t *s_keyboard_layout_mapping_buf, const int keyboard_layout_mapping_cnt)
+DECLSPEC int hc_execute_keyboard_layout_mapping (u32 *w, const int pw_len, LOCAL_AS keyboard_layout_mapping_t *s_keyboard_layout_mapping_buf, const int keyboard_layout_mapping_cnt)
 {
-  u32 out_buf[16] = { 0 };
+  u32 out_buf[32] = { 0 };
 
   u8 *out_ptr = (u8 *) out_buf;
 
   int out_len = 0;
 
-  // TC/VC passwords are limited to 64
-
-  u32 w[16];
-
-  w[ 0] = w0[0];
-  w[ 1] = w0[1];
-  w[ 2] = w0[2];
-  w[ 3] = w0[3];
-  w[ 4] = w1[0];
-  w[ 5] = w1[1];
-  w[ 6] = w1[2];
-  w[ 7] = w1[3];
-  w[ 8] = w2[0];
-  w[ 9] = w2[1];
-  w[10] = w2[2];
-  w[11] = w2[3];
-  w[12] = w3[0];
-  w[13] = w3[1];
-  w[14] = w3[2];
-  w[15] = w3[3];
+  // TC/VC passwords are limited to 128
 
   u8 *w_ptr = (u8 *) w;
 
@@ -2733,22 +2714,38 @@ DECLSPEC int hc_execute_keyboard_layout_mapping (u32 *w0, u32 *w1, u32 *w2, u32
     }
   }
 
-  w0[0] = out_buf[ 0];
-  w0[1] = out_buf[ 1];
-  w0[2] = out_buf[ 2];
-  w0[3] = out_buf[ 3];
-  w1[0] = out_buf[ 4];
-  w1[1] = out_buf[ 5];
-  w1[2] = out_buf[ 6];
-  w1[3] = out_buf[ 7];
-  w2[0] = out_buf[ 8];
-  w2[1] = out_buf[ 9];
-  w2[2] = out_buf[10];
-  w2[3] = out_buf[11];
-  w3[0] = out_buf[12];
-  w3[1] = out_buf[13];
-  w3[2] = out_buf[14];
-  w3[3] = out_buf[15];
+  w[ 0] = out_buf[ 0];
+  w[ 1] = out_buf[ 1];
+  w[ 2] = out_buf[ 2];
+  w[ 3] = out_buf[ 3];
+  w[ 4] = out_buf[ 4];
+  w[ 5] = out_buf[ 5];
+  w[ 6] = out_buf[ 6];
+  w[ 7] = out_buf[ 7];
+  w[ 8] = out_buf[ 8];
+  w[ 9] = out_buf[ 9];
+  w[10] = out_buf[10];
+  w[11] = out_buf[11];
+  w[12] = out_buf[12];
+  w[13] = out_buf[13];
+  w[14] = out_buf[14];
+  w[15] = out_buf[15];
+  w[16] = out_buf[16];
+  w[17] = out_buf[17];
+  w[18] = out_buf[18];
+  w[19] = out_buf[19];
+  w[20] = out_buf[20];
+  w[21] = out_buf[21];
+  w[22] = out_buf[22];
+  w[23] = out_buf[23];
+  w[24] = out_buf[24];
+  w[25] = out_buf[25];
+  w[26] = out_buf[26];
+  w[27] = out_buf[27];
+  w[28] = out_buf[28];
+  w[29] = out_buf[29];
+  w[30] = out_buf[30];
+  w[31] = out_buf[31];
 
   return out_len;
 }

OpenCL/inc_common.h

@@ -256,7 +256,7 @@ DECLSPEC int is_valid_hex_32 (const u32 v);
 DECLSPEC int is_valid_base58_8 (const u8 v);
 DECLSPEC int is_valid_base58_32 (const u32 v);
 DECLSPEC int hc_find_keyboard_layout_map (const u32 search, const int search_len, LOCAL_AS keyboard_layout_mapping_t *s_keyboard_layout_mapping_buf, const int keyboard_layout_mapping_cnt);
-DECLSPEC int hc_execute_keyboard_layout_mapping (u32 *w0, u32 *w1, u32 *w2, u32 *w3, const int pw_len, LOCAL_AS keyboard_layout_mapping_t *s_keyboard_layout_mapping_buf, const int keyboard_layout_mapping_cnt);
+DECLSPEC int hc_execute_keyboard_layout_mapping (u32 *w, const int pw_len, LOCAL_AS keyboard_layout_mapping_t *s_keyboard_layout_mapping_buf, const int keyboard_layout_mapping_cnt);
 DECLSPEC void make_utf16be (const u32x *in, u32x *out1, u32x *out2);
 DECLSPEC void make_utf16beN (const u32x *in, u32x *out1, u32x *out2);
 DECLSPEC void make_utf16le (const u32x *in, u32x *out1, u32x *out2);

OpenCL/inc_rp.cl

@@ -688,7 +688,9 @@ DECLSPEC int mangle_dupeblock_last (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const
 
 DECLSPEC int mangle_title_sep (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p1, u32 *buf, const int len)
 {
-  if ((len + 4) >= RP_PASSWORD_SIZE) return (len); // cheap way to not need to check for overflow of i + 1
+  if (len >= RP_PASSWORD_SIZE) return (len);
+
+  u32 rem = 0xff;
 
   for (int i = 0, idx = 0; i < len; i += 4, idx += 1)
   {
@@ -696,22 +698,18 @@ DECLSPEC int mangle_title_sep (MAYBE_UNUSED const u8 p0, MAYBE_UNUSED const u8 p
 
     buf[idx] = t | generate_cmask (t);
 
-    u32 out0 = 0;
-    u32 out1 = 0;
+    u32 out = rem;
 
-    if (((t >>  0) & 0xff) == p0) out0 |= 0x0000ff00;
-    if (((t >>  8) & 0xff) == p0) out0 |= 0x00ff0000;
-    if (((t >> 16) & 0xff) == p0) out0 |= 0xff000000;
-    if (((t >> 24) & 0xff) == p0) out1 |= 0x000000ff;
+    rem = 0;
 
-    buf[idx + 0] &= ~(generate_cmask (buf[idx + 0]) & out0);
-    buf[idx + 1] &= ~(generate_cmask (buf[idx + 1]) & out1);
+    if (((t >>  0) & 0xff) == p0) out |= 0x0000ff00;
+    if (((t >>  8) & 0xff) == p0) out |= 0x00ff0000;
+    if (((t >> 16) & 0xff) == p0) out |= 0xff000000;
+    if (((t >> 24) & 0xff) == p0) rem |= 0x000000ff;
+
+    buf[idx] &= ~(generate_cmask (buf[idx]) & out);
   }
 
-  const u32 t = buf[0];
-
-  buf[0] = t & ~(0x00000020 & generate_cmask (t));
-
   return (len);
 }
 

OpenCL/inc_truecrypt_keyfile.cl

@@ -33,3 +33,67 @@ DECLSPEC u32 u8add (const u32 a, const u32 b)
 
   return r;
 }
+
+DECLSPEC u32 hc_apply_keyfile_tc (u32 *w, const int pw_len, const GLOBAL_AS tc_t *tc)
+{
+  if (tc->keyfile_enabled == 0) return pw_len;
+
+  if (pw_len > 64)
+  {
+    w[ 0] = u8add (w[ 0], tc->keyfile_buf32[ 0]);
+    w[ 1] = u8add (w[ 1], tc->keyfile_buf32[ 1]);
+    w[ 2] = u8add (w[ 2], tc->keyfile_buf32[ 2]);
+    w[ 3] = u8add (w[ 3], tc->keyfile_buf32[ 3]);
+    w[ 4] = u8add (w[ 4], tc->keyfile_buf32[ 4]);
+    w[ 5] = u8add (w[ 5], tc->keyfile_buf32[ 5]);
+    w[ 6] = u8add (w[ 6], tc->keyfile_buf32[ 6]);
+    w[ 7] = u8add (w[ 7], tc->keyfile_buf32[ 7]);
+    w[ 8] = u8add (w[ 8], tc->keyfile_buf32[ 8]);
+    w[ 9] = u8add (w[ 9], tc->keyfile_buf32[ 9]);
+    w[10] = u8add (w[10], tc->keyfile_buf32[10]);
+    w[11] = u8add (w[11], tc->keyfile_buf32[11]);
+    w[12] = u8add (w[12], tc->keyfile_buf32[12]);
+    w[13] = u8add (w[13], tc->keyfile_buf32[13]);
+    w[14] = u8add (w[14], tc->keyfile_buf32[14]);
+    w[15] = u8add (w[15], tc->keyfile_buf32[15]);
+    w[16] = u8add (w[16], tc->keyfile_buf32[16]);
+    w[17] = u8add (w[17], tc->keyfile_buf32[17]);
+    w[18] = u8add (w[18], tc->keyfile_buf32[18]);
+    w[19] = u8add (w[19], tc->keyfile_buf32[19]);
+    w[20] = u8add (w[20], tc->keyfile_buf32[20]);
+    w[21] = u8add (w[21], tc->keyfile_buf32[21]);
+    w[22] = u8add (w[22], tc->keyfile_buf32[22]);
+    w[23] = u8add (w[23], tc->keyfile_buf32[23]);
+    w[24] = u8add (w[24], tc->keyfile_buf32[24]);
+    w[25] = u8add (w[25], tc->keyfile_buf32[25]);
+    w[26] = u8add (w[26], tc->keyfile_buf32[26]);
+    w[27] = u8add (w[27], tc->keyfile_buf32[27]);
+    w[28] = u8add (w[28], tc->keyfile_buf32[28]);
+    w[29] = u8add (w[29], tc->keyfile_buf32[29]);
+    w[30] = u8add (w[30], tc->keyfile_buf32[30]);
+    w[31] = u8add (w[31], tc->keyfile_buf32[31]);
+
+    return 128;
+  }
+  else
+  {
+    w[ 0] = u8add (w[ 0], tc->keyfile_buf16[ 0]);
+    w[ 1] = u8add (w[ 1], tc->keyfile_buf16[ 1]);
+    w[ 2] = u8add (w[ 2], tc->keyfile_buf16[ 2]);
+    w[ 3] = u8add (w[ 3], tc->keyfile_buf16[ 3]);
+    w[ 4] = u8add (w[ 4], tc->keyfile_buf16[ 4]);
+    w[ 5] = u8add (w[ 5], tc->keyfile_buf16[ 5]);
+    w[ 6] = u8add (w[ 6], tc->keyfile_buf16[ 6]);
+    w[ 7] = u8add (w[ 7], tc->keyfile_buf16[ 7]);
+    w[ 8] = u8add (w[ 8], tc->keyfile_buf16[ 8]);
+    w[ 9] = u8add (w[ 9], tc->keyfile_buf16[ 9]);
+    w[10] = u8add (w[10], tc->keyfile_buf16[10]);
+    w[11] = u8add (w[11], tc->keyfile_buf16[11]);
+    w[12] = u8add (w[12], tc->keyfile_buf16[12]);
+    w[13] = u8add (w[13], tc->keyfile_buf16[13]);
+    w[14] = u8add (w[14], tc->keyfile_buf16[14]);
+    w[15] = u8add (w[15], tc->keyfile_buf16[15]);
+
+    return 64;
+  }
+}

OpenCL/inc_truecrypt_keyfile.h

@@ -7,5 +7,6 @@
 #define _INC_TRUECRYPT_KEYFILE_H
 
 DECLSPEC u32 u8add (const u32 a, const u32 b);
+DECLSPEC u32 hc_apply_keyfile_tc (u32 *w, const int pw_len, const GLOBAL_AS tc_t *tc);
 
 #endif // _INC_TRUECRYPT_KEYFILE_H

OpenCL/inc_veracrypt_keyfile.cl

@@ -0,0 +1,99 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "inc_vendor.h"
+#include "inc_types.h"
+#include "inc_platform.h"
+#include "inc_common.h"
+#include "inc_veracrypt_keyfile.h"
+
+DECLSPEC u32 u8add (const u32 a, const u32 b)
+{
+  const u32 a1 = (a >>  0) & 0xff;
+  const u32 a2 = (a >>  8) & 0xff;
+  const u32 a3 = (a >> 16) & 0xff;
+  const u32 a4 = (a >> 24) & 0xff;
+
+  const u32 b1 = (b >>  0) & 0xff;
+  const u32 b2 = (b >>  8) & 0xff;
+  const u32 b3 = (b >> 16) & 0xff;
+  const u32 b4 = (b >> 24) & 0xff;
+
+  const u32 r1 = (a1 + b1) & 0xff;
+  const u32 r2 = (a2 + b2) & 0xff;
+  const u32 r3 = (a3 + b3) & 0xff;
+  const u32 r4 = (a4 + b4) & 0xff;
+
+  const u32 r = r1 <<  0
+              | r2 <<  8
+              | r3 << 16
+              | r4 << 24;
+
+  return r;
+}
+
+DECLSPEC u32 hc_apply_keyfile_vc (u32 *w, const int pw_len, const GLOBAL_AS vc_t *vc)
+{
+  if (vc->keyfile_enabled == 0) return pw_len;
+
+  if (pw_len > 64)
+  {
+    w[ 0] = u8add (w[ 0], vc->keyfile_buf32[ 0]);
+    w[ 1] = u8add (w[ 1], vc->keyfile_buf32[ 1]);
+    w[ 2] = u8add (w[ 2], vc->keyfile_buf32[ 2]);
+    w[ 3] = u8add (w[ 3], vc->keyfile_buf32[ 3]);
+    w[ 4] = u8add (w[ 4], vc->keyfile_buf32[ 4]);
+    w[ 5] = u8add (w[ 5], vc->keyfile_buf32[ 5]);
+    w[ 6] = u8add (w[ 6], vc->keyfile_buf32[ 6]);
+    w[ 7] = u8add (w[ 7], vc->keyfile_buf32[ 7]);
+    w[ 8] = u8add (w[ 8], vc->keyfile_buf32[ 8]);
+    w[ 9] = u8add (w[ 9], vc->keyfile_buf32[ 9]);
+    w[10] = u8add (w[10], vc->keyfile_buf32[10]);
+    w[11] = u8add (w[11], vc->keyfile_buf32[11]);
+    w[12] = u8add (w[12], vc->keyfile_buf32[12]);
+    w[13] = u8add (w[13], vc->keyfile_buf32[13]);
+    w[14] = u8add (w[14], vc->keyfile_buf32[14]);
+    w[15] = u8add (w[15], vc->keyfile_buf32[15]);
+    w[16] = u8add (w[16], vc->keyfile_buf32[16]);
+    w[17] = u8add (w[17], vc->keyfile_buf32[17]);
+    w[18] = u8add (w[18], vc->keyfile_buf32[18]);
+    w[19] = u8add (w[19], vc->keyfile_buf32[19]);
+    w[20] = u8add (w[20], vc->keyfile_buf32[20]);
+    w[21] = u8add (w[21], vc->keyfile_buf32[21]);
+    w[22] = u8add (w[22], vc->keyfile_buf32[22]);
+    w[23] = u8add (w[23], vc->keyfile_buf32[23]);
+    w[24] = u8add (w[24], vc->keyfile_buf32[24]);
+    w[25] = u8add (w[25], vc->keyfile_buf32[25]);
+    w[26] = u8add (w[26], vc->keyfile_buf32[26]);
+    w[27] = u8add (w[27], vc->keyfile_buf32[27]);
+    w[28] = u8add (w[28], vc->keyfile_buf32[28]);
+    w[29] = u8add (w[29], vc->keyfile_buf32[29]);
+    w[30] = u8add (w[30], vc->keyfile_buf32[30]);
+    w[31] = u8add (w[31], vc->keyfile_buf32[31]);
+
+    return 128;
+  }
+  else
+  {
+    w[ 0] = u8add (w[ 0], vc->keyfile_buf16[ 0]);
+    w[ 1] = u8add (w[ 1], vc->keyfile_buf16[ 1]);
+    w[ 2] = u8add (w[ 2], vc->keyfile_buf16[ 2]);
+    w[ 3] = u8add (w[ 3], vc->keyfile_buf16[ 3]);
+    w[ 4] = u8add (w[ 4], vc->keyfile_buf16[ 4]);
+    w[ 5] = u8add (w[ 5], vc->keyfile_buf16[ 5]);
+    w[ 6] = u8add (w[ 6], vc->keyfile_buf16[ 6]);
+    w[ 7] = u8add (w[ 7], vc->keyfile_buf16[ 7]);
+    w[ 8] = u8add (w[ 8], vc->keyfile_buf16[ 8]);
+    w[ 9] = u8add (w[ 9], vc->keyfile_buf16[ 9]);
+    w[10] = u8add (w[10], vc->keyfile_buf16[10]);
+    w[11] = u8add (w[11], vc->keyfile_buf16[11]);
+    w[12] = u8add (w[12], vc->keyfile_buf16[12]);
+    w[13] = u8add (w[13], vc->keyfile_buf16[13]);
+    w[14] = u8add (w[14], vc->keyfile_buf16[14]);
+    w[15] = u8add (w[15], vc->keyfile_buf16[15]);
+
+    return 64;
+  }
+}

OpenCL/inc_veracrypt_keyfile.h

@@ -0,0 +1,12 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#ifndef _INC_VERACRYPT_KEYFILE_H
+#define _INC_VERACRYPT_KEYFILE_H
+
+DECLSPEC u32 u8add (const u32 a, const u32 b);
+DECLSPEC u32 hc_apply_keyfile_vc (u32 *w, const int pw_len, const GLOBAL_AS vc_t *vc);
+
+#endif // _INC_VERACRYPT_KEYFILE_H

OpenCL/m01500_a3-pure.cl

@@ -2229,9 +2229,7 @@ KERNEL_FQ void m01500_mxx (KERN_ATTR_BITSLICE ())
 
         u32 tmpResult = 0;
 
-        #ifdef _unroll
         #pragma unroll
-        #endif
         for (int i = 0; i < 32; i++)
         {
           const u32 b0 = -((search[0] >> i) & 1);
@@ -2247,12 +2245,12 @@ KERNEL_FQ void m01500_mxx (KERN_ATTR_BITSLICE ())
 
         const u32 r0 = search[0];
         const u32 r1 = search[1];
+        #ifdef KERNEL_STATIC
         const u32 r2 = 0;
         const u32 r3 = 0;
-
-        #ifdef KERNEL_STATIC
-        #include COMPARE_M
         #endif
+
+        #include COMPARE_M
       }
     }
     else
@@ -2260,9 +2258,7 @@ KERNEL_FQ void m01500_mxx (KERN_ATTR_BITSLICE ())
       u32 out0[32];
       u32 out1[32];
 
-      #ifdef _unroll
       #pragma unroll
-      #endif
       for (int i = 0; i < 32; i++)
       {
         out0[i] = out[ 0 + 31 - i];
@@ -2272,15 +2268,13 @@ KERNEL_FQ void m01500_mxx (KERN_ATTR_BITSLICE ())
       transpose32c (out0);
       transpose32c (out1);
 
-      #ifdef _unroll
       #pragma unroll
-      #endif
       for (int slice = 0; slice < 32; slice++)
       {
         const u32 r0 = out0[31 - slice];
         const u32 r1 = out1[31 - slice];
-        const u32 r2 = 0;
         #ifdef KERNEL_STATIC
+        const u32 r2 = 0;
         const u32 r3 = 0;
         #endif
 

OpenCL/m03000_a3-pure.cl

@@ -2070,9 +2070,7 @@ KERNEL_FQ void m03000_mxx (KERN_ATTR_BITSLICE ())
 
         u32 tmpResult = 0;
 
-        #ifdef _unroll
         #pragma unroll
-        #endif
         for (int i = 0; i < 32; i++)
         {
           const u32 b0 = -((search[0] >> i) & 1);
@@ -2088,8 +2086,8 @@ KERNEL_FQ void m03000_mxx (KERN_ATTR_BITSLICE ())
 
         const u32 r0 = search[0];
         const u32 r1 = search[1];
-        const u32 r2 = 0;
         #ifdef KERNEL_STATIC
+        const u32 r2 = 0;
         const u32 r3 = 0;
         #endif
 
@@ -2101,9 +2099,7 @@ KERNEL_FQ void m03000_mxx (KERN_ATTR_BITSLICE ())
       u32 out0[32];
       u32 out1[32];
 
-      #ifdef _unroll
       #pragma unroll
-      #endif
       for (int i = 0; i < 32; i++)
       {
         out0[i] = out[ 0 + 31 - i];
@@ -2113,15 +2109,13 @@ KERNEL_FQ void m03000_mxx (KERN_ATTR_BITSLICE ())
       transpose32c (out0);
       transpose32c (out1);
 
-      #ifdef _unroll
       #pragma unroll
-      #endif
       for (int slice = 0; slice < 32; slice++)
       {
         const u32 r0 = out0[31 - slice];
         const u32 r1 = out1[31 - slice];
-        const u32 r2 = 0;
         #ifdef KERNEL_STATIC
+        const u32 r2 = 0;
         const u32 r3 = 0;
         #endif
 

OpenCL/m06211-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc_tmp
@@ -108,52 +110,50 @@ KERNEL_FQ void m06211_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   ripemd160_hmac_ctx_t ripemd160_hmac_ctx;
 
-  ripemd160_hmac_init_64 (&ripemd160_hmac_ctx, w0, w1, w2, w3);
+  ripemd160_hmac_init (&ripemd160_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = ripemd160_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = ripemd160_hmac_ctx.ipad.h[1];
@@ -173,6 +173,11 @@ KERNEL_FQ void m06211_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
   {
     ripemd160_hmac_ctx_t ripemd160_hmac_ctx2 = ripemd160_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j << 24;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m06212-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc_tmp
@@ -108,52 +110,50 @@ KERNEL_FQ void m06212_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   ripemd160_hmac_ctx_t ripemd160_hmac_ctx;
 
-  ripemd160_hmac_init_64 (&ripemd160_hmac_ctx, w0, w1, w2, w3);
+  ripemd160_hmac_init (&ripemd160_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = ripemd160_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = ripemd160_hmac_ctx.ipad.h[1];
@@ -173,6 +173,11 @@ KERNEL_FQ void m06212_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
   {
     ripemd160_hmac_ctx_t ripemd160_hmac_ctx2 = ripemd160_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j << 24;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m06213-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc_tmp
@@ -108,52 +110,50 @@ KERNEL_FQ void m06213_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   ripemd160_hmac_ctx_t ripemd160_hmac_ctx;
 
-  ripemd160_hmac_init_64 (&ripemd160_hmac_ctx, w0, w1, w2, w3);
+  ripemd160_hmac_init (&ripemd160_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = ripemd160_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = ripemd160_hmac_ctx.ipad.h[1];
@@ -173,6 +173,11 @@ KERNEL_FQ void m06213_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
   {
     ripemd160_hmac_ctx_t ripemd160_hmac_ctx2 = ripemd160_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j << 24;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m06221-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc64_tmp
@@ -130,105 +132,50 @@ KERNEL_FQ void m06221_init (KERN_ATTR_TMPS_ESALT (tc64_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
-  u32 w4[4];
-  u32 w5[4];
-  u32 w6[4];
-  u32 w7[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
-  w4[0] = pws[gid].i[16];
-  w4[1] = pws[gid].i[17];
-  w4[2] = pws[gid].i[18];
-  w4[3] = pws[gid].i[19];
-  w5[0] = pws[gid].i[20];
-  w5[1] = pws[gid].i[21];
-  w5[2] = pws[gid].i[22];
-  w5[3] = pws[gid].i[23];
-  w6[0] = pws[gid].i[24];
-  w6[1] = pws[gid].i[25];
-  w6[2] = pws[gid].i[26];
-  w6[3] = pws[gid].i[27];
-  w7[0] = pws[gid].i[28];
-  w7[1] = pws[gid].i[29];
-  w7[2] = pws[gid].i[30];
-  w7[3] = pws[gid].i[31];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
-  w4[0] = hc_swap32_S (w4[0]);
-  w4[1] = hc_swap32_S (w4[1]);
-  w4[2] = hc_swap32_S (w4[2]);
-  w4[3] = hc_swap32_S (w4[3]);
-  w5[0] = hc_swap32_S (w5[0]);
-  w5[1] = hc_swap32_S (w5[1]);
-  w5[2] = hc_swap32_S (w5[2]);
-  w5[3] = hc_swap32_S (w5[3]);
-  w6[0] = hc_swap32_S (w6[0]);
-  w6[1] = hc_swap32_S (w6[1]);
-  w6[2] = hc_swap32_S (w6[2]);
-  w6[3] = hc_swap32_S (w6[3]);
-  w7[0] = hc_swap32_S (w7[0]);
-  w7[1] = hc_swap32_S (w7[1]);
-  w7[2] = hc_swap32_S (w7[2]);
-  w7[3] = hc_swap32_S (w7[3]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha512_hmac_ctx_t sha512_hmac_ctx;
 
-  sha512_hmac_init_128 (&sha512_hmac_ctx, w0, w1, w2, w3, w5, w5, w6, w7);
+  sha512_hmac_init_swap (&sha512_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha512_hmac_ctx.ipad.h[1];
@@ -254,6 +201,15 @@ KERNEL_FQ void m06221_init (KERN_ATTR_TMPS_ESALT (tc64_tmp_t, tc_t))
   {
     sha512_hmac_ctx_t sha512_hmac_ctx2 = sha512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m06222-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc64_tmp
@@ -130,105 +132,50 @@ KERNEL_FQ void m06222_init (KERN_ATTR_TMPS_ESALT (tc64_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
-  u32 w4[4];
-  u32 w5[4];
-  u32 w6[4];
-  u32 w7[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
-  w4[0] = pws[gid].i[16];
-  w4[1] = pws[gid].i[17];
-  w4[2] = pws[gid].i[18];
-  w4[3] = pws[gid].i[19];
-  w5[0] = pws[gid].i[20];
-  w5[1] = pws[gid].i[21];
-  w5[2] = pws[gid].i[22];
-  w5[3] = pws[gid].i[23];
-  w6[0] = pws[gid].i[24];
-  w6[1] = pws[gid].i[25];
-  w6[2] = pws[gid].i[26];
-  w6[3] = pws[gid].i[27];
-  w7[0] = pws[gid].i[28];
-  w7[1] = pws[gid].i[29];
-  w7[2] = pws[gid].i[30];
-  w7[3] = pws[gid].i[31];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
-  w4[0] = hc_swap32_S (w4[0]);
-  w4[1] = hc_swap32_S (w4[1]);
-  w4[2] = hc_swap32_S (w4[2]);
-  w4[3] = hc_swap32_S (w4[3]);
-  w5[0] = hc_swap32_S (w5[0]);
-  w5[1] = hc_swap32_S (w5[1]);
-  w5[2] = hc_swap32_S (w5[2]);
-  w5[3] = hc_swap32_S (w5[3]);
-  w6[0] = hc_swap32_S (w6[0]);
-  w6[1] = hc_swap32_S (w6[1]);
-  w6[2] = hc_swap32_S (w6[2]);
-  w6[3] = hc_swap32_S (w6[3]);
-  w7[0] = hc_swap32_S (w7[0]);
-  w7[1] = hc_swap32_S (w7[1]);
-  w7[2] = hc_swap32_S (w7[2]);
-  w7[3] = hc_swap32_S (w7[3]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha512_hmac_ctx_t sha512_hmac_ctx;
 
-  sha512_hmac_init_128 (&sha512_hmac_ctx, w0, w1, w2, w3, w5, w5, w6, w7);
+  sha512_hmac_init_swap (&sha512_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha512_hmac_ctx.ipad.h[1];
@@ -254,6 +201,15 @@ KERNEL_FQ void m06222_init (KERN_ATTR_TMPS_ESALT (tc64_tmp_t, tc_t))
   {
     sha512_hmac_ctx_t sha512_hmac_ctx2 = sha512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m06223-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc64_tmp
@@ -130,105 +132,50 @@ KERNEL_FQ void m06223_init (KERN_ATTR_TMPS_ESALT (tc64_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
-  u32 w4[4];
-  u32 w5[4];
-  u32 w6[4];
-  u32 w7[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
-  w4[0] = pws[gid].i[16];
-  w4[1] = pws[gid].i[17];
-  w4[2] = pws[gid].i[18];
-  w4[3] = pws[gid].i[19];
-  w5[0] = pws[gid].i[20];
-  w5[1] = pws[gid].i[21];
-  w5[2] = pws[gid].i[22];
-  w5[3] = pws[gid].i[23];
-  w6[0] = pws[gid].i[24];
-  w6[1] = pws[gid].i[25];
-  w6[2] = pws[gid].i[26];
-  w6[3] = pws[gid].i[27];
-  w7[0] = pws[gid].i[28];
-  w7[1] = pws[gid].i[29];
-  w7[2] = pws[gid].i[30];
-  w7[3] = pws[gid].i[31];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
-  w4[0] = hc_swap32_S (w4[0]);
-  w4[1] = hc_swap32_S (w4[1]);
-  w4[2] = hc_swap32_S (w4[2]);
-  w4[3] = hc_swap32_S (w4[3]);
-  w5[0] = hc_swap32_S (w5[0]);
-  w5[1] = hc_swap32_S (w5[1]);
-  w5[2] = hc_swap32_S (w5[2]);
-  w5[3] = hc_swap32_S (w5[3]);
-  w6[0] = hc_swap32_S (w6[0]);
-  w6[1] = hc_swap32_S (w6[1]);
-  w6[2] = hc_swap32_S (w6[2]);
-  w6[3] = hc_swap32_S (w6[3]);
-  w7[0] = hc_swap32_S (w7[0]);
-  w7[1] = hc_swap32_S (w7[1]);
-  w7[2] = hc_swap32_S (w7[2]);
-  w7[3] = hc_swap32_S (w7[3]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha512_hmac_ctx_t sha512_hmac_ctx;
 
-  sha512_hmac_init_128 (&sha512_hmac_ctx, w0, w1, w2, w3, w5, w5, w6, w7);
+  sha512_hmac_init_swap (&sha512_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha512_hmac_ctx.ipad.h[1];
@@ -254,6 +201,15 @@ KERNEL_FQ void m06223_init (KERN_ATTR_TMPS_ESALT (tc64_tmp_t, tc_t))
   {
     sha512_hmac_ctx_t sha512_hmac_ctx2 = sha512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m06231-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc_tmp
@@ -210,69 +212,50 @@ KERNEL_FQ void m06231_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   whirlpool_hmac_ctx_t whirlpool_hmac_ctx;
 
-  whirlpool_hmac_init_64 (&whirlpool_hmac_ctx, w0, w1, w2, w3, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
+  whirlpool_hmac_init_swap (&whirlpool_hmac_ctx, w, pw_len, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
 
   tmps[gid].ipad[ 0] = whirlpool_hmac_ctx.ipad.h[ 0];
   tmps[gid].ipad[ 1] = whirlpool_hmac_ctx.ipad.h[ 1];
@@ -314,6 +297,11 @@ KERNEL_FQ void m06231_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
   {
     whirlpool_hmac_ctx_t whirlpool_hmac_ctx2 = whirlpool_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m06232-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc_tmp
@@ -210,69 +212,50 @@ KERNEL_FQ void m06232_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   whirlpool_hmac_ctx_t whirlpool_hmac_ctx;
 
-  whirlpool_hmac_init_64 (&whirlpool_hmac_ctx, w0, w1, w2, w3, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
+  whirlpool_hmac_init_swap (&whirlpool_hmac_ctx, w, pw_len, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
 
   tmps[gid].ipad[ 0] = whirlpool_hmac_ctx.ipad.h[ 0];
   tmps[gid].ipad[ 1] = whirlpool_hmac_ctx.ipad.h[ 1];
@@ -314,6 +297,11 @@ KERNEL_FQ void m06232_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
   {
     whirlpool_hmac_ctx_t whirlpool_hmac_ctx2 = whirlpool_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m06233-pure.cl

@@ -21,7 +21,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -30,9 +32,9 @@ typedef struct tc
 } tc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
+#include "inc_truecrypt_keyfile.cl"
 #endif
 
 typedef struct tc_tmp
@@ -210,69 +212,50 @@ KERNEL_FQ void m06233_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_tc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   whirlpool_hmac_ctx_t whirlpool_hmac_ctx;
 
-  whirlpool_hmac_init_64 (&whirlpool_hmac_ctx, w0, w1, w2, w3, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
+  whirlpool_hmac_init_swap (&whirlpool_hmac_ctx, w, pw_len, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
 
   tmps[gid].ipad[ 0] = whirlpool_hmac_ctx.ipad.h[ 0];
   tmps[gid].ipad[ 1] = whirlpool_hmac_ctx.ipad.h[ 1];
@@ -314,6 +297,11 @@ KERNEL_FQ void m06233_init (KERN_ATTR_TMPS_ESALT (tc_tmp_t, tc_t))
   {
     whirlpool_hmac_ctx_t whirlpool_hmac_ctx2 = whirlpool_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13711-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -150,52 +152,50 @@ KERNEL_FQ void m13711_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   ripemd160_hmac_ctx_t ripemd160_hmac_ctx;
 
-  ripemd160_hmac_init_64 (&ripemd160_hmac_ctx, w0, w1, w2, w3);
+  ripemd160_hmac_init (&ripemd160_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = ripemd160_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = ripemd160_hmac_ctx.ipad.h[1];
@@ -215,6 +215,11 @@ KERNEL_FQ void m13711_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     ripemd160_hmac_ctx_t ripemd160_hmac_ctx2 = ripemd160_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j << 24;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13712-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -201,52 +203,50 @@ KERNEL_FQ void m13712_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   ripemd160_hmac_ctx_t ripemd160_hmac_ctx;
 
-  ripemd160_hmac_init_64 (&ripemd160_hmac_ctx, w0, w1, w2, w3);
+  ripemd160_hmac_init (&ripemd160_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = ripemd160_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = ripemd160_hmac_ctx.ipad.h[1];
@@ -266,6 +266,11 @@ KERNEL_FQ void m13712_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     ripemd160_hmac_ctx_t ripemd160_hmac_ctx2 = ripemd160_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j << 24;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13713-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -266,52 +268,50 @@ KERNEL_FQ void m13713_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   ripemd160_hmac_ctx_t ripemd160_hmac_ctx;
 
-  ripemd160_hmac_init_64 (&ripemd160_hmac_ctx, w0, w1, w2, w3);
+  ripemd160_hmac_init (&ripemd160_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = ripemd160_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = ripemd160_hmac_ctx.ipad.h[1];
@@ -331,6 +331,11 @@ KERNEL_FQ void m13713_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     ripemd160_hmac_ctx_t ripemd160_hmac_ctx2 = ripemd160_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j << 24;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13721-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc64_tmp
@@ -172,105 +174,50 @@ KERNEL_FQ void m13721_init (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
-  u32 w4[4];
-  u32 w5[4];
-  u32 w6[4];
-  u32 w7[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
-  w4[0] = pws[gid].i[16];
-  w4[1] = pws[gid].i[17];
-  w4[2] = pws[gid].i[18];
-  w4[3] = pws[gid].i[19];
-  w5[0] = pws[gid].i[20];
-  w5[1] = pws[gid].i[21];
-  w5[2] = pws[gid].i[22];
-  w5[3] = pws[gid].i[23];
-  w6[0] = pws[gid].i[24];
-  w6[1] = pws[gid].i[25];
-  w6[2] = pws[gid].i[26];
-  w6[3] = pws[gid].i[27];
-  w7[0] = pws[gid].i[28];
-  w7[1] = pws[gid].i[29];
-  w7[2] = pws[gid].i[30];
-  w7[3] = pws[gid].i[31];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
-  w4[0] = hc_swap32_S (w4[0]);
-  w4[1] = hc_swap32_S (w4[1]);
-  w4[2] = hc_swap32_S (w4[2]);
-  w4[3] = hc_swap32_S (w4[3]);
-  w5[0] = hc_swap32_S (w5[0]);
-  w5[1] = hc_swap32_S (w5[1]);
-  w5[2] = hc_swap32_S (w5[2]);
-  w5[3] = hc_swap32_S (w5[3]);
-  w6[0] = hc_swap32_S (w6[0]);
-  w6[1] = hc_swap32_S (w6[1]);
-  w6[2] = hc_swap32_S (w6[2]);
-  w6[3] = hc_swap32_S (w6[3]);
-  w7[0] = hc_swap32_S (w7[0]);
-  w7[1] = hc_swap32_S (w7[1]);
-  w7[2] = hc_swap32_S (w7[2]);
-  w7[3] = hc_swap32_S (w7[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha512_hmac_ctx_t sha512_hmac_ctx;
 
-  sha512_hmac_init_128 (&sha512_hmac_ctx, w0, w1, w2, w3, w5, w5, w6, w7);
+  sha512_hmac_init_swap (&sha512_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha512_hmac_ctx.ipad.h[1];
@@ -296,6 +243,15 @@ KERNEL_FQ void m13721_init (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
   {
     sha512_hmac_ctx_t sha512_hmac_ctx2 = sha512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13722-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc64_tmp
@@ -223,105 +225,50 @@ KERNEL_FQ void m13722_init (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
-  u32 w4[4];
-  u32 w5[4];
-  u32 w6[4];
-  u32 w7[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
-  w4[0] = pws[gid].i[16];
-  w4[1] = pws[gid].i[17];
-  w4[2] = pws[gid].i[18];
-  w4[3] = pws[gid].i[19];
-  w5[0] = pws[gid].i[20];
-  w5[1] = pws[gid].i[21];
-  w5[2] = pws[gid].i[22];
-  w5[3] = pws[gid].i[23];
-  w6[0] = pws[gid].i[24];
-  w6[1] = pws[gid].i[25];
-  w6[2] = pws[gid].i[26];
-  w6[3] = pws[gid].i[27];
-  w7[0] = pws[gid].i[28];
-  w7[1] = pws[gid].i[29];
-  w7[2] = pws[gid].i[30];
-  w7[3] = pws[gid].i[31];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
-  w4[0] = hc_swap32_S (w4[0]);
-  w4[1] = hc_swap32_S (w4[1]);
-  w4[2] = hc_swap32_S (w4[2]);
-  w4[3] = hc_swap32_S (w4[3]);
-  w5[0] = hc_swap32_S (w5[0]);
-  w5[1] = hc_swap32_S (w5[1]);
-  w5[2] = hc_swap32_S (w5[2]);
-  w5[3] = hc_swap32_S (w5[3]);
-  w6[0] = hc_swap32_S (w6[0]);
-  w6[1] = hc_swap32_S (w6[1]);
-  w6[2] = hc_swap32_S (w6[2]);
-  w6[3] = hc_swap32_S (w6[3]);
-  w7[0] = hc_swap32_S (w7[0]);
-  w7[1] = hc_swap32_S (w7[1]);
-  w7[2] = hc_swap32_S (w7[2]);
-  w7[3] = hc_swap32_S (w7[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha512_hmac_ctx_t sha512_hmac_ctx;
 
-  sha512_hmac_init_128 (&sha512_hmac_ctx, w0, w1, w2, w3, w5, w5, w6, w7);
+  sha512_hmac_init_swap (&sha512_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha512_hmac_ctx.ipad.h[1];
@@ -347,6 +294,15 @@ KERNEL_FQ void m13722_init (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
   {
     sha512_hmac_ctx_t sha512_hmac_ctx2 = sha512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13723-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc64_tmp
@@ -288,105 +290,50 @@ KERNEL_FQ void m13723_init (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
-  u32 w4[4];
-  u32 w5[4];
-  u32 w6[4];
-  u32 w7[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
-  w4[0] = pws[gid].i[16];
-  w4[1] = pws[gid].i[17];
-  w4[2] = pws[gid].i[18];
-  w4[3] = pws[gid].i[19];
-  w5[0] = pws[gid].i[20];
-  w5[1] = pws[gid].i[21];
-  w5[2] = pws[gid].i[22];
-  w5[3] = pws[gid].i[23];
-  w6[0] = pws[gid].i[24];
-  w6[1] = pws[gid].i[25];
-  w6[2] = pws[gid].i[26];
-  w6[3] = pws[gid].i[27];
-  w7[0] = pws[gid].i[28];
-  w7[1] = pws[gid].i[29];
-  w7[2] = pws[gid].i[30];
-  w7[3] = pws[gid].i[31];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
-  w4[0] = hc_swap32_S (w4[0]);
-  w4[1] = hc_swap32_S (w4[1]);
-  w4[2] = hc_swap32_S (w4[2]);
-  w4[3] = hc_swap32_S (w4[3]);
-  w5[0] = hc_swap32_S (w5[0]);
-  w5[1] = hc_swap32_S (w5[1]);
-  w5[2] = hc_swap32_S (w5[2]);
-  w5[3] = hc_swap32_S (w5[3]);
-  w6[0] = hc_swap32_S (w6[0]);
-  w6[1] = hc_swap32_S (w6[1]);
-  w6[2] = hc_swap32_S (w6[2]);
-  w6[3] = hc_swap32_S (w6[3]);
-  w7[0] = hc_swap32_S (w7[0]);
-  w7[1] = hc_swap32_S (w7[1]);
-  w7[2] = hc_swap32_S (w7[2]);
-  w7[3] = hc_swap32_S (w7[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha512_hmac_ctx_t sha512_hmac_ctx;
 
-  sha512_hmac_init_128 (&sha512_hmac_ctx, w0, w1, w2, w3, w5, w5, w6, w7);
+  sha512_hmac_init_swap (&sha512_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha512_hmac_ctx.ipad.h[1];
@@ -412,6 +359,15 @@ KERNEL_FQ void m13723_init (KERN_ATTR_TMPS_ESALT (vc64_tmp_t, vc_t))
   {
     sha512_hmac_ctx_t sha512_hmac_ctx2 = sha512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13731-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -252,69 +254,50 @@ KERNEL_FQ void m13731_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   whirlpool_hmac_ctx_t whirlpool_hmac_ctx;
 
-  whirlpool_hmac_init_64 (&whirlpool_hmac_ctx, w0, w1, w2, w3, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
+  whirlpool_hmac_init_swap (&whirlpool_hmac_ctx, w, pw_len, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
 
   tmps[gid].ipad[ 0] = whirlpool_hmac_ctx.ipad.h[ 0];
   tmps[gid].ipad[ 1] = whirlpool_hmac_ctx.ipad.h[ 1];
@@ -356,6 +339,11 @@ KERNEL_FQ void m13731_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     whirlpool_hmac_ctx_t whirlpool_hmac_ctx2 = whirlpool_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13732-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -303,69 +305,50 @@ KERNEL_FQ void m13732_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   whirlpool_hmac_ctx_t whirlpool_hmac_ctx;
 
-  whirlpool_hmac_init_64 (&whirlpool_hmac_ctx, w0, w1, w2, w3, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
+  whirlpool_hmac_init_swap (&whirlpool_hmac_ctx, w, pw_len, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
 
   tmps[gid].ipad[ 0] = whirlpool_hmac_ctx.ipad.h[ 0];
   tmps[gid].ipad[ 1] = whirlpool_hmac_ctx.ipad.h[ 1];
@@ -407,6 +390,11 @@ KERNEL_FQ void m13732_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     whirlpool_hmac_ctx_t whirlpool_hmac_ctx2 = whirlpool_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13733-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -368,69 +370,50 @@ KERNEL_FQ void m13733_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   whirlpool_hmac_ctx_t whirlpool_hmac_ctx;
 
-  whirlpool_hmac_init_64 (&whirlpool_hmac_ctx, w0, w1, w2, w3, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
+  whirlpool_hmac_init_swap (&whirlpool_hmac_ctx, w, pw_len, s_MT0, s_MT1, s_MT2, s_MT3, s_MT4, s_MT5, s_MT6, s_MT7);
 
   tmps[gid].ipad[ 0] = whirlpool_hmac_ctx.ipad.h[ 0];
   tmps[gid].ipad[ 1] = whirlpool_hmac_ctx.ipad.h[ 1];
@@ -472,6 +455,11 @@ KERNEL_FQ void m13733_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     whirlpool_hmac_ctx_t whirlpool_hmac_ctx2 = whirlpool_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13751-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -156,69 +158,50 @@ KERNEL_FQ void m13751_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha256_hmac_ctx_t sha256_hmac_ctx;
 
-  sha256_hmac_init_64 (&sha256_hmac_ctx, w0, w1, w2, w3);
+  sha256_hmac_init_swap (&sha256_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha256_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha256_hmac_ctx.ipad.h[1];
@@ -244,6 +227,15 @@ KERNEL_FQ void m13751_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     sha256_hmac_ctx_t sha256_hmac_ctx2 = sha256_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13752-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -207,69 +209,50 @@ KERNEL_FQ void m13752_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha256_hmac_ctx_t sha256_hmac_ctx;
 
-  sha256_hmac_init_64 (&sha256_hmac_ctx, w0, w1, w2, w3);
+  sha256_hmac_init_swap (&sha256_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha256_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha256_hmac_ctx.ipad.h[1];
@@ -295,6 +278,15 @@ KERNEL_FQ void m13752_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     sha256_hmac_ctx_t sha256_hmac_ctx2 = sha256_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13753-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc_tmp
@@ -272,69 +274,50 @@ KERNEL_FQ void m13753_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   sha256_hmac_ctx_t sha256_hmac_ctx;
 
-  sha256_hmac_init_64 (&sha256_hmac_ctx, w0, w1, w2, w3);
+  sha256_hmac_init_swap (&sha256_hmac_ctx, w, pw_len);
 
   tmps[gid].ipad[0] = sha256_hmac_ctx.ipad.h[0];
   tmps[gid].ipad[1] = sha256_hmac_ctx.ipad.h[1];
@@ -360,6 +343,15 @@ KERNEL_FQ void m13753_init (KERN_ATTR_TMPS_ESALT (vc_tmp_t, vc_t))
   {
     sha256_hmac_ctx_t sha256_hmac_ctx2 = sha256_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+    u32 w4[4];
+    u32 w5[4];
+    u32 w6[4];
+    u32 w7[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13771-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc64_sbog_tmp
@@ -222,69 +224,50 @@ KERNEL_FQ void m13771_init (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   streebog512_hmac_ctx_t streebog512_hmac_ctx;
 
-  streebog512_hmac_init_64 (&streebog512_hmac_ctx, w0, w1, w2, w3, s_sbob_sl64);
+  streebog512_hmac_init_swap (&streebog512_hmac_ctx, w, pw_len, s_sbob_sl64);
 
   tmps[gid].ipad_hash[0] = streebog512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad_hash[1] = streebog512_hmac_ctx.ipad.h[1];
@@ -328,6 +311,11 @@ KERNEL_FQ void m13771_init (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
   {
     streebog512_hmac_ctx_t streebog512_hmac_ctx2 = streebog512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13772-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc64_sbog_tmp
@@ -273,69 +275,50 @@ KERNEL_FQ void m13772_init (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   streebog512_hmac_ctx_t streebog512_hmac_ctx;
 
-  streebog512_hmac_init_64 (&streebog512_hmac_ctx, w0, w1, w2, w3, s_sbob_sl64);
+  streebog512_hmac_init_swap (&streebog512_hmac_ctx, w, pw_len, s_sbob_sl64);
 
   tmps[gid].ipad_hash[0] = streebog512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad_hash[1] = streebog512_hmac_ctx.ipad.h[1];
@@ -379,6 +362,11 @@ KERNEL_FQ void m13772_init (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
   {
     streebog512_hmac_ctx_t streebog512_hmac_ctx2 = streebog512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m13773-pure.cl

@@ -23,7 +23,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -36,10 +38,10 @@ typedef struct vc
 } vc_t;
 
 #ifdef KERNEL_STATIC
-#include "inc_truecrypt_keyfile.cl"
 #include "inc_truecrypt_crc32.cl"
 #include "inc_truecrypt_xts.cl"
 #include "inc_veracrypt_xts.cl"
+#include "inc_veracrypt_keyfile.cl"
 #endif
 
 typedef struct vc64_sbog_tmp
@@ -338,69 +340,50 @@ KERNEL_FQ void m13773_init (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
    * base
    */
 
-  u32 w0[4];
-  u32 w1[4];
-  u32 w2[4];
-  u32 w3[4];
+  u32 w[32];
 
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
-  w2[2] = pws[gid].i[10];
-  w2[3] = pws[gid].i[11];
-  w3[0] = pws[gid].i[12];
-  w3[1] = pws[gid].i[13];
-  w3[2] = pws[gid].i[14];
-  w3[3] = pws[gid].i[15];
+  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];
+  w[16] = pws[gid].i[16];
+  w[17] = pws[gid].i[17];
+  w[18] = pws[gid].i[18];
+  w[19] = pws[gid].i[19];
+  w[20] = pws[gid].i[20];
+  w[21] = pws[gid].i[21];
+  w[22] = pws[gid].i[22];
+  w[23] = pws[gid].i[23];
+  w[24] = pws[gid].i[24];
+  w[25] = pws[gid].i[25];
+  w[26] = pws[gid].i[26];
+  w[27] = pws[gid].i[27];
+  w[28] = pws[gid].i[28];
+  w[29] = pws[gid].i[29];
+  w[30] = pws[gid].i[30];
+  w[31] = pws[gid].i[31];
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32 pw_len = pws[gid].pw_len;
 
-  hc_execute_keyboard_layout_mapping (w0, w1, w2, w3, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
+  hc_execute_keyboard_layout_mapping (w, pw_len, s_keyboard_layout_mapping_buf, keyboard_layout_mapping_cnt);
 
-  w0[0] = u8add (w0[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 0]);
-  w0[1] = u8add (w0[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 1]);
-  w0[2] = u8add (w0[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 2]);
-  w0[3] = u8add (w0[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 3]);
-  w1[0] = u8add (w1[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 4]);
-  w1[1] = u8add (w1[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 5]);
-  w1[2] = u8add (w1[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 6]);
-  w1[3] = u8add (w1[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 7]);
-  w2[0] = u8add (w2[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 8]);
-  w2[1] = u8add (w2[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[ 9]);
-  w2[2] = u8add (w2[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[10]);
-  w2[3] = u8add (w2[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[11]);
-  w3[0] = u8add (w3[0], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[12]);
-  w3[1] = u8add (w3[1], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[13]);
-  w3[2] = u8add (w3[2], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[14]);
-  w3[3] = u8add (w3[3], esalt_bufs[DIGESTS_OFFSET].keyfile_buf[15]);
-
-  w0[0] = hc_swap32_S (w0[0]);
-  w0[1] = hc_swap32_S (w0[1]);
-  w0[2] = hc_swap32_S (w0[2]);
-  w0[3] = hc_swap32_S (w0[3]);
-  w1[0] = hc_swap32_S (w1[0]);
-  w1[1] = hc_swap32_S (w1[1]);
-  w1[2] = hc_swap32_S (w1[2]);
-  w1[3] = hc_swap32_S (w1[3]);
-  w2[0] = hc_swap32_S (w2[0]);
-  w2[1] = hc_swap32_S (w2[1]);
-  w2[2] = hc_swap32_S (w2[2]);
-  w2[3] = hc_swap32_S (w2[3]);
-  w3[0] = hc_swap32_S (w3[0]);
-  w3[1] = hc_swap32_S (w3[1]);
-  w3[2] = hc_swap32_S (w3[2]);
-  w3[3] = hc_swap32_S (w3[3]);
+  pw_len = hc_apply_keyfile_vc (w, pw_len, &esalt_bufs[DIGESTS_OFFSET]);
 
   streebog512_hmac_ctx_t streebog512_hmac_ctx;
 
-  streebog512_hmac_init_64 (&streebog512_hmac_ctx, w0, w1, w2, w3, s_sbob_sl64);
+  streebog512_hmac_init_swap (&streebog512_hmac_ctx, w, pw_len, s_sbob_sl64);
 
   tmps[gid].ipad_hash[0] = streebog512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad_hash[1] = streebog512_hmac_ctx.ipad.h[1];
@@ -444,6 +427,11 @@ KERNEL_FQ void m13773_init (KERN_ATTR_TMPS_ESALT (vc64_sbog_tmp_t, vc_t))
   {
     streebog512_hmac_ctx_t streebog512_hmac_ctx2 = streebog512_hmac_ctx;
 
+    u32 w0[4];
+    u32 w1[4];
+    u32 w2[4];
+    u32 w3[4];
+
     w0[0] = j;
     w0[1] = 0;
     w0[2] = 0;

OpenCL/m15900-pure.cl

@@ -255,7 +255,7 @@ KERNEL_FQ void m15900_init (KERN_ATTR_TMPS_ESALT (dpapimk_tmp_v2_t, dpapimk_t))
 
   sha512_hmac_ctx_t sha512_hmac_ctx;
 
-  sha512_hmac_init_128 (&sha512_hmac_ctx, w0, w1, w2, w3, w5, w5, w6, w7);
+  sha512_hmac_init_128 (&sha512_hmac_ctx, w0, w1, w2, w3, w4, w5, w6, w7);
 
   tmps[gid].ipad64[0] = sha512_hmac_ctx.ipad.h[0];
   tmps[gid].ipad64[1] = sha512_hmac_ctx.ipad.h[1];

OpenCL/m25500-optimized.cl

@@ -1,455 +0,0 @@
-/**
- * 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_hash_sha256.cl"
-#include "inc_cipher_aes.cl"
-#include "inc_cipher_aes-gcm.cl"
-#endif
-
-#define COMPARE_S "inc_comp_single.cl"
-#define COMPARE_M "inc_comp_multi.cl"
-
-typedef struct pbkdf2_sha256_tmp
-{
-  u32  ipad[8];
-  u32  opad[8];
-
-  u32  dgst[32];
-  u32  out[32];
-
-} pbkdf2_sha256_tmp_t;
-
-typedef struct pbkdf2_sha256_aes_gcm
-{
-  u32 salt_buf[64];
-  u32 iv_buf[4];
-  u32 iv_len;
-  u32 ct_buf[16];
-  u32 ct_len;
-
-} pbkdf2_sha256_aes_gcm_t;
-
-DECLSPEC void hmac_sha256_run_V (u32x *w0, u32x *w1, u32x *w2, u32x *w3, u32x *ipad, u32x *opad, u32x *digest)
-{
-  digest[0] = ipad[0];
-  digest[1] = ipad[1];
-  digest[2] = ipad[2];
-  digest[3] = ipad[3];
-  digest[4] = ipad[4];
-  digest[5] = ipad[5];
-  digest[6] = ipad[6];
-  digest[7] = ipad[7];
-
-  sha256_transform_vector (w0, w1, w2, w3, digest);
-
-  w0[0] = digest[0];
-  w0[1] = digest[1];
-  w0[2] = digest[2];
-  w0[3] = digest[3];
-  w1[0] = digest[4];
-  w1[1] = digest[5];
-  w1[2] = digest[6];
-  w1[3] = digest[7];
-  w2[0] = 0x80000000;
-  w2[1] = 0;
-  w2[2] = 0;
-  w2[3] = 0;
-  w3[0] = 0;
-  w3[1] = 0;
-  w3[2] = 0;
-  w3[3] = (64 + 32) * 8;
-
-  digest[0] = opad[0];
-  digest[1] = opad[1];
-  digest[2] = opad[2];
-  digest[3] = opad[3];
-  digest[4] = opad[4];
-  digest[5] = opad[5];
-  digest[6] = opad[6];
-  digest[7] = opad[7];
-
-  sha256_transform_vector (w0, w1, w2, w3, digest);
-}
-
-KERNEL_FQ void m25500_init (KERN_ATTR_TMPS_ESALT (pbkdf2_sha256_tmp_t, pbkdf2_sha256_aes_gcm_t))
-{
-  /**
-   * base
-   */
-
-  const u64 gid = get_global_id (0);
-
-  if (gid >= gid_max) return;
-
-  sha256_hmac_ctx_t sha256_hmac_ctx;
-
-  sha256_hmac_init_global_swap (&sha256_hmac_ctx, pws[gid].i, pws[gid].pw_len);
-
-  tmps[gid].ipad[0] = sha256_hmac_ctx.ipad.h[0];
-  tmps[gid].ipad[1] = sha256_hmac_ctx.ipad.h[1];
-  tmps[gid].ipad[2] = sha256_hmac_ctx.ipad.h[2];
-  tmps[gid].ipad[3] = sha256_hmac_ctx.ipad.h[3];
-  tmps[gid].ipad[4] = sha256_hmac_ctx.ipad.h[4];
-  tmps[gid].ipad[5] = sha256_hmac_ctx.ipad.h[5];
-  tmps[gid].ipad[6] = sha256_hmac_ctx.ipad.h[6];
-  tmps[gid].ipad[7] = sha256_hmac_ctx.ipad.h[7];
-
-  tmps[gid].opad[0] = sha256_hmac_ctx.opad.h[0];
-  tmps[gid].opad[1] = sha256_hmac_ctx.opad.h[1];
-  tmps[gid].opad[2] = sha256_hmac_ctx.opad.h[2];
-  tmps[gid].opad[3] = sha256_hmac_ctx.opad.h[3];
-  tmps[gid].opad[4] = sha256_hmac_ctx.opad.h[4];
-  tmps[gid].opad[5] = sha256_hmac_ctx.opad.h[5];
-  tmps[gid].opad[6] = sha256_hmac_ctx.opad.h[6];
-  tmps[gid].opad[7] = sha256_hmac_ctx.opad.h[7];
-
-  sha256_hmac_update_global_swap (&sha256_hmac_ctx, esalt_bufs[DIGESTS_OFFSET].salt_buf, salt_bufs[SALT_POS].salt_len);
-
-  for (u32 i = 0, j = 1; i < 8; i += 8, j += 1)
-  {
-    sha256_hmac_ctx_t sha256_hmac_ctx2 = sha256_hmac_ctx;
-
-    u32 w0[4];
-    u32 w1[4];
-    u32 w2[4];
-    u32 w3[4];
-
-    w0[0] = j;
-    w0[1] = 0;
-    w0[2] = 0;
-    w0[3] = 0;
-    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;
-
-    sha256_hmac_update_64 (&sha256_hmac_ctx2, w0, w1, w2, w3, 4);
-
-    sha256_hmac_final (&sha256_hmac_ctx2);
-
-    tmps[gid].dgst[i + 0] = sha256_hmac_ctx2.opad.h[0];
-    tmps[gid].dgst[i + 1] = sha256_hmac_ctx2.opad.h[1];
-    tmps[gid].dgst[i + 2] = sha256_hmac_ctx2.opad.h[2];
-    tmps[gid].dgst[i + 3] = sha256_hmac_ctx2.opad.h[3];
-    tmps[gid].dgst[i + 4] = sha256_hmac_ctx2.opad.h[4];
-    tmps[gid].dgst[i + 5] = sha256_hmac_ctx2.opad.h[5];
-    tmps[gid].dgst[i + 6] = sha256_hmac_ctx2.opad.h[6];
-    tmps[gid].dgst[i + 7] = sha256_hmac_ctx2.opad.h[7];
-
-    tmps[gid].out[i + 0] = tmps[gid].dgst[i + 0];
-    tmps[gid].out[i + 1] = tmps[gid].dgst[i + 1];
-    tmps[gid].out[i + 2] = tmps[gid].dgst[i + 2];
-    tmps[gid].out[i + 3] = tmps[gid].dgst[i + 3];
-    tmps[gid].out[i + 4] = tmps[gid].dgst[i + 4];
-    tmps[gid].out[i + 5] = tmps[gid].dgst[i + 5];
-    tmps[gid].out[i + 6] = tmps[gid].dgst[i + 6];
-    tmps[gid].out[i + 7] = tmps[gid].dgst[i + 7];
-  }
-}
-
-KERNEL_FQ void m25500_loop (KERN_ATTR_TMPS_ESALT (pbkdf2_sha256_tmp_t, pbkdf2_sha256_aes_gcm_t))
-{
-  const u64 gid = get_global_id (0);
-
-  if ((gid * VECT_SIZE) >= gid_max) return;
-
-  u32x ipad[8];
-  u32x opad[8];
-
-  ipad[0] = packv (tmps, ipad, gid, 0);
-  ipad[1] = packv (tmps, ipad, gid, 1);
-  ipad[2] = packv (tmps, ipad, gid, 2);
-  ipad[3] = packv (tmps, ipad, gid, 3);
-  ipad[4] = packv (tmps, ipad, gid, 4);
-  ipad[5] = packv (tmps, ipad, gid, 5);
-  ipad[6] = packv (tmps, ipad, gid, 6);
-  ipad[7] = packv (tmps, ipad, gid, 7);
-
-  opad[0] = packv (tmps, opad, gid, 0);
-  opad[1] = packv (tmps, opad, gid, 1);
-  opad[2] = packv (tmps, opad, gid, 2);
-  opad[3] = packv (tmps, opad, gid, 3);
-  opad[4] = packv (tmps, opad, gid, 4);
-  opad[5] = packv (tmps, opad, gid, 5);
-  opad[6] = packv (tmps, opad, gid, 6);
-  opad[7] = packv (tmps, opad, gid, 7);
-
-  for (u32 i = 0; i < 8; i += 8)
-  {
-    u32x dgst[8];
-    u32x out[8];
-
-    dgst[0] = packv (tmps, dgst, gid, i + 0);
-    dgst[1] = packv (tmps, dgst, gid, i + 1);
-    dgst[2] = packv (tmps, dgst, gid, i + 2);
-    dgst[3] = packv (tmps, dgst, gid, i + 3);
-    dgst[4] = packv (tmps, dgst, gid, i + 4);
-    dgst[5] = packv (tmps, dgst, gid, i + 5);
-    dgst[6] = packv (tmps, dgst, gid, i + 6);
-    dgst[7] = packv (tmps, dgst, gid, i + 7);
-
-    out[0] = packv (tmps, out, gid, i + 0);
-    out[1] = packv (tmps, out, gid, i + 1);
-    out[2] = packv (tmps, out, gid, i + 2);
-    out[3] = packv (tmps, out, gid, i + 3);
-    out[4] = packv (tmps, out, gid, i + 4);
-    out[5] = packv (tmps, out, gid, i + 5);
-    out[6] = packv (tmps, out, gid, i + 6);
-    out[7] = packv (tmps, out, gid, i + 7);
-
-    for (u32 j = 0; j < loop_cnt; j++)
-    {
-      u32x w0[4];
-      u32x w1[4];
-      u32x w2[4];
-      u32x w3[4];
-
-      w0[0] = dgst[0];
-      w0[1] = dgst[1];
-      w0[2] = dgst[2];
-      w0[3] = dgst[3];
-      w1[0] = dgst[4];
-      w1[1] = dgst[5];
-      w1[2] = dgst[6];
-      w1[3] = dgst[7];
-      w2[0] = 0x80000000;
-      w2[1] = 0;
-      w2[2] = 0;
-      w2[3] = 0;
-      w3[0] = 0;
-      w3[1] = 0;
-      w3[2] = 0;
-      w3[3] = (64 + 32) * 8;
-
-      hmac_sha256_run_V (w0, w1, w2, w3, ipad, opad, dgst);
-
-      out[0] ^= dgst[0];
-      out[1] ^= dgst[1];
-      out[2] ^= dgst[2];
-      out[3] ^= dgst[3];
-      out[4] ^= dgst[4];
-      out[5] ^= dgst[5];
-      out[6] ^= dgst[6];
-      out[7] ^= dgst[7];
-    }
-
-    unpackv (tmps, dgst, gid, i + 0, dgst[0]);
-    unpackv (tmps, dgst, gid, i + 1, dgst[1]);
-    unpackv (tmps, dgst, gid, i + 2, dgst[2]);
-    unpackv (tmps, dgst, gid, i + 3, dgst[3]);
-    unpackv (tmps, dgst, gid, i + 4, dgst[4]);
-    unpackv (tmps, dgst, gid, i + 5, dgst[5]);
-    unpackv (tmps, dgst, gid, i + 6, dgst[6]);
-    unpackv (tmps, dgst, gid, i + 7, dgst[7]);
-
-    unpackv (tmps, out, gid, i + 0, out[0]);
-    unpackv (tmps, out, gid, i + 1, out[1]);
-    unpackv (tmps, out, gid, i + 2, out[2]);
-    unpackv (tmps, out, gid, i + 3, out[3]);
-    unpackv (tmps, out, gid, i + 4, out[4]);
-    unpackv (tmps, out, gid, i + 5, out[5]);
-    unpackv (tmps, out, gid, i + 6, out[6]);
-    unpackv (tmps, out, gid, i + 7, out[7]);
-  }
-}
-
-KERNEL_FQ void m25500_comp (KERN_ATTR_TMPS_ESALT (pbkdf2_sha256_tmp_t, pbkdf2_sha256_aes_gcm_t))
-{
-  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_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_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_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;
-
-  // keys
-
-  u32 ukey[8];
-
-  ukey[0] = tmps[gid].out[0];
-  ukey[1] = tmps[gid].out[1];
-  ukey[2] = tmps[gid].out[2];
-  ukey[3] = tmps[gid].out[3];
-  ukey[4] = tmps[gid].out[4];
-  ukey[5] = tmps[gid].out[5];
-  ukey[6] = tmps[gid].out[6];
-  ukey[7] = tmps[gid].out[7];
-
-  u32 key[60] = { 0 };
-
-  u32 subKey[4] = { 0 };
-
-  AES256_set_encrypt_key (key, ukey, s_te0, s_te1, s_te2, s_te3);
-
-  AES256_encrypt (key, subKey, subKey, s_te0, s_te1, s_te2, s_te3, s_te4);
-
-  // iv
-
-  const u32 iv[4] = {
-    esalt_bufs[DIGESTS_OFFSET].iv_buf[0],
-    esalt_bufs[DIGESTS_OFFSET].iv_buf[1],
-    esalt_bufs[DIGESTS_OFFSET].iv_buf[2],
-    esalt_bufs[DIGESTS_OFFSET].iv_buf[3]
-  };
-
-  u32 J0[4] = {
-    iv[0],
-    iv[1],
-    iv[2],
-    0x00000001
-  };
-
-  // ct
-
-  u32 enc[14] = { 0 };
-
-  enc[ 0] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 0];
-  enc[ 1] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 1];
-  enc[ 2] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 2];
-  enc[ 3] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 3];
-  enc[ 4] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 4];
-  enc[ 5] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 5];
-  enc[ 6] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 6];
-  enc[ 7] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 7];
-  enc[ 8] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 8];
-  enc[ 9] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 9];
-  enc[10] = esalt_bufs[DIGESTS_OFFSET].ct_buf[10];
-  enc[11] = esalt_bufs[DIGESTS_OFFSET].ct_buf[11];
-  enc[12] = esalt_bufs[DIGESTS_OFFSET].ct_buf[12];
-  enc[13] = esalt_bufs[DIGESTS_OFFSET].ct_buf[13];
-
-  u32 enc_len = esalt_bufs[DIGESTS_OFFSET].ct_len;
-
-  u32 S[4] = { 0 };
-
-  u32 t[4] = { 0 };
-
-  S[0] ^= enc[0];
-  S[1] ^= enc[1];
-  S[2] ^= enc[2];
-  S[3] ^= enc[3];
-
-  AES_GCM_gf_mult (S, subKey, t);
-
-  S[0] = t[0] ^ enc[4];
-  S[1] = t[1] ^ enc[5];
-  S[2] = t[2] ^ enc[6];
-  S[3] = t[3] ^ enc[7];
-
-  AES_GCM_gf_mult (S, subKey, t);
-
-  S[0] = t[0] ^ enc[8];
-  S[1] = t[1] ^ enc[9];
-  S[2] = t[2] ^ enc[10];
-  S[3] = t[3] ^ enc[11];
-
-  AES_GCM_gf_mult (S, subKey, t);
-
-  S[0] = t[0];
-  S[1] = t[1];
-  S[2] = t[2];
-  S[3] = t[3];
-
-  t[0] = enc[12];
-  t[1] = enc[13];
-  t[2] = 0;
-  t[3] = 0;
-
-  S[0] ^= t[0];
-  S[1] ^= t[1];
-  S[2] ^= t[2];
-  S[3] ^= t[3];
-
-  AES_GCM_gf_mult (S, subKey, t);
-
-  S[0] = t[0];
-  S[1] = t[1];
-  S[2] = t[2];
-  S[3] = t[3];
-
-  u32 len_buf[4] = { 0 };
-
-  len_buf[0] = 0;
-  len_buf[3] = enc_len * 8;
-
-  S[0] ^= len_buf[0];
-  S[1] ^= len_buf[1];
-  S[2] ^= len_buf[2];
-  S[3] ^= len_buf[3];
-
-  AES_GCM_gf_mult (S, subKey, t);
-
-  S[0] = t[0];
-  S[1] = t[1];
-  S[2] = t[2];
-  S[3] = t[3];
-
-  J0[3] = 0x00000001;
-
-  u32 T[4] = { 0 };
-
-  AES256_encrypt (key, J0, T, s_te0, s_te1, s_te2, s_te3, s_te4);
-
-  /* compare tag */
-
-  const u32 r0 = T[0] ^ S[0];
-  const u32 r1 = T[1] ^ S[1];
-  const u32 r2 = T[2] ^ S[2];
-  const u32 r3 = T[3] ^ S[3];
-
-  #define il_pos 0
-
-  #ifdef KERNEL_STATIC
-  #include COMPARE_M
-  #endif
-}

OpenCL/m25500-pure.cl

@@ -403,3 +403,185 @@ KERNEL_FQ void m25500_comp (KERN_ATTR_TMPS_ESALT (pbkdf2_sha256_tmp_t, pbkdf2_sh
   #include COMPARE_M
   #endif
 }
+
+/*
+Optimized GCM: No real speed benefit. For documentation purpose
+
+KERNEL_FQ void m25500_comp (KERN_ATTR_TMPS_ESALT (pbkdf2_sha256_tmp_t, pbkdf2_sha256_aes_gcm_t))
+{
+  const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
+  const u64 lsz = get_local_size (0);
+
+  #ifdef REAL_SHM
+
+  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_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_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;
+
+  // keys
+
+  u32 ukey[8];
+
+  ukey[0] = tmps[gid].out[0];
+  ukey[1] = tmps[gid].out[1];
+  ukey[2] = tmps[gid].out[2];
+  ukey[3] = tmps[gid].out[3];
+  ukey[4] = tmps[gid].out[4];
+  ukey[5] = tmps[gid].out[5];
+  ukey[6] = tmps[gid].out[6];
+  ukey[7] = tmps[gid].out[7];
+
+  u32 key[60] = { 0 };
+
+  u32 subKey[4] = { 0 };
+
+  AES256_set_encrypt_key (key, ukey, s_te0, s_te1, s_te2, s_te3);
+
+  AES256_encrypt (key, subKey, subKey, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+  // iv
+
+  const u32 iv[4] = {
+    esalt_bufs[DIGESTS_OFFSET].iv_buf[0],
+    esalt_bufs[DIGESTS_OFFSET].iv_buf[1],
+    esalt_bufs[DIGESTS_OFFSET].iv_buf[2],
+    esalt_bufs[DIGESTS_OFFSET].iv_buf[3]
+  };
+
+  u32 J0[4] = {
+    iv[0],
+    iv[1],
+    iv[2],
+    0x00000001
+  };
+
+  // ct
+
+  u32 enc[14] = { 0 };
+
+  enc[ 0] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 0];
+  enc[ 1] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 1];
+  enc[ 2] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 2];
+  enc[ 3] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 3];
+  enc[ 4] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 4];
+  enc[ 5] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 5];
+  enc[ 6] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 6];
+  enc[ 7] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 7];
+  enc[ 8] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 8];
+  enc[ 9] = esalt_bufs[DIGESTS_OFFSET].ct_buf[ 9];
+  enc[10] = esalt_bufs[DIGESTS_OFFSET].ct_buf[10];
+  enc[11] = esalt_bufs[DIGESTS_OFFSET].ct_buf[11];
+  enc[12] = esalt_bufs[DIGESTS_OFFSET].ct_buf[12];
+  enc[13] = esalt_bufs[DIGESTS_OFFSET].ct_buf[13];
+
+  u32 enc_len = esalt_bufs[DIGESTS_OFFSET].ct_len;
+
+  u32 S[4] = { 0 };
+
+  u32 t[4] = { 0 };
+
+  S[0] ^= enc[0];
+  S[1] ^= enc[1];
+  S[2] ^= enc[2];
+  S[3] ^= enc[3];
+
+  AES_GCM_gf_mult (S, subKey, t);
+
+  S[0] = t[0] ^ enc[4];
+  S[1] = t[1] ^ enc[5];
+  S[2] = t[2] ^ enc[6];
+  S[3] = t[3] ^ enc[7];
+
+  AES_GCM_gf_mult (S, subKey, t);
+
+  S[0] = t[0] ^ enc[8];
+  S[1] = t[1] ^ enc[9];
+  S[2] = t[2] ^ enc[10];
+  S[3] = t[3] ^ enc[11];
+
+  AES_GCM_gf_mult (S, subKey, t);
+
+  S[0] = t[0];
+  S[1] = t[1];
+  S[2] = t[2];
+  S[3] = t[3];
+
+  t[0] = enc[12];
+  t[1] = enc[13];
+  t[2] = 0;
+  t[3] = 0;
+
+  S[0] ^= t[0];
+  S[1] ^= t[1];
+  S[2] ^= t[2];
+  S[3] ^= t[3];
+
+  AES_GCM_gf_mult (S, subKey, t);
+
+  S[0] = t[0];
+  S[1] = t[1];
+  S[2] = t[2];
+  S[3] = t[3];
+
+  u32 len_buf[4] = { 0 };
+
+  len_buf[0] = 0;
+  len_buf[3] = enc_len * 8;
+
+  S[0] ^= len_buf[0];
+  S[1] ^= len_buf[1];
+  S[2] ^= len_buf[2];
+  S[3] ^= len_buf[3];
+
+  AES_GCM_gf_mult (S, subKey, t);
+
+  S[0] = t[0];
+  S[1] = t[1];
+  S[2] = t[2];
+  S[3] = t[3];
+
+  J0[3] = 0x00000001;
+
+  u32 T[4] = { 0 };
+
+  AES256_encrypt (key, J0, T, s_te0, s_te1, s_te2, s_te3, s_te4);
+
+  const u32 r0 = T[0] ^ S[0];
+  const u32 r1 = T[1] ^ S[1];
+  const u32 r2 = T[2] ^ S[2];
+  const u32 r3 = T[3] ^ S[3];
+
+  #define il_pos 0
+
+  #ifdef KERNEL_STATIC
+  #include COMPARE_M
+  #endif
+}
+
+*/

OpenCL/m26200_a0-pure.cl

@@ -0,0 +1,234 @@
+/**
+ * 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.h"
+#include "inc_rp.cl"
+#include "inc_scalar.cl"
+#endif
+
+CONSTANT_VK u32a PE_CONST[256] = 
+{
+      0, 49345, 49537,   320, 49921,   960,   640, 49729, 50689,  1728,  1920, 51009,  1280, 50625, 50305,  1088,
+  52225,  3264,  3456, 52545,  3840, 53185, 52865,  3648,  2560, 51905, 52097,  2880, 51457,  2496,  2176, 51265,
+  55297,  6336,  6528, 55617,  6912, 56257, 55937,  6720,  7680, 57025, 57217,  8000, 56577,  7616,  7296, 56385,
+   5120, 54465, 54657,  5440, 55041,  6080,  5760, 54849, 53761,  4800,  4992, 54081,  4352, 53697, 53377,  4160,
+  61441, 12480, 12672, 61761, 13056, 62401, 62081, 12864, 13824, 63169, 63361, 14144, 62721, 13760, 13440, 62529,
+  15360, 64705, 64897, 15680, 65281, 16320, 16000, 65089, 64001, 15040, 15232, 64321, 14592, 63937, 63617, 14400,
+  10240, 59585, 59777, 10560, 60161, 11200, 10880, 59969, 60929, 11968, 12160, 61249, 11520, 60865, 60545, 11328,
+  58369,  9408,  9600, 58689,  9984, 59329, 59009,  9792,  8704, 58049, 58241,  9024, 57601,  8640,  8320, 57409,
+  40961, 24768, 24960, 41281, 25344, 41921, 41601, 25152, 26112, 42689, 42881, 26432, 42241, 26048, 25728, 42049,
+  27648, 44225, 44417, 27968, 44801, 28608, 28288, 44609, 43521, 27328, 27520, 43841, 26880, 43457, 43137, 26688,
+  30720, 47297, 47489, 31040, 47873, 31680, 31360, 47681, 48641, 32448, 32640, 48961, 32000, 48577, 48257, 31808,
+  46081, 29888, 30080, 46401, 30464, 47041, 46721, 30272, 29184, 45761, 45953, 29504, 45313, 29120, 28800, 45121,
+  20480, 37057, 37249, 20800, 37633, 21440, 21120, 37441, 38401, 22208, 22400, 38721, 21760, 38337, 38017, 21568,
+  39937, 23744, 23936, 40257, 24320, 40897, 40577, 24128, 23040, 39617, 39809, 23360, 39169, 22976, 22656, 38977,
+  34817, 18624, 18816, 35137, 19200, 35777, 35457, 19008, 19968, 36545, 36737, 20288, 36097, 19904, 19584, 35905,
+  17408, 33985, 34177, 17728, 34561, 18368, 18048, 34369, 33281, 17088, 17280, 33601, 16640, 33217, 32897, 16448
+};
+
+KERNEL_FQ void m26200_mxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    pw_t tmp = PASTE_PW;
+
+    tmp.pw_len = apply_rules (rules_buf[il_pos].cmds, tmp.i, tmp.pw_len);
+
+    u32 hash = 17;
+
+    u8  scratch[16] = { 0 };
+
+    u8 *input = (u8 *) tmp.i;
+
+    for (u32 i = 0; i < 5; i++)
+    {
+      for (u32 j = 0; j < tmp.pw_len; j++)
+      {
+        int idx = 15 - (j & 15);
+
+        scratch[idx] ^= input[j];
+      }
+
+      for (u32 j = 0; j < 16; j += 2)
+      {
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[15]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[14]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[13]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[12]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[11]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[10]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 9]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 8]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 7]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 6]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 5]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 4]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 3]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 2]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 1]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 0]]);
+
+        scratch[j]     = (unsigned char)( hash       & 0xff);
+        scratch[j + 1] = (unsigned char)((hash >> 8) & 0xff);
+      }
+    }
+
+    u8 target[16] = { 0 };
+
+    for (u32 i = 0; i < 16; i++)
+    {
+      u8 lower = (scratch[i] & 0x7f);
+
+      if ((lower >= 'A' && lower <= 'Z') || (lower >= 'a' && lower <= 'z'))
+      {
+        target[i] = lower;
+      }
+      else
+      {
+        target[i] = (u8)((scratch[i] >> 4) + 0x61);
+      }
+    }
+
+    u32 *digest = (u32 *) target;
+
+    const u32 r0 = digest[DGST_R0];
+    const u32 r1 = digest[DGST_R1];
+    const u32 r2 = digest[DGST_R2];
+    const u32 r3 = digest[DGST_R3];
+
+    COMPARE_M_SCALAR (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26200_sxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * 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]
+  };
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    pw_t tmp = PASTE_PW;
+
+    tmp.pw_len = apply_rules (rules_buf[il_pos].cmds, tmp.i, tmp.pw_len);
+
+    u32 hash = 17;
+
+    u8  scratch[16] = { 0 };
+
+    u8 *input = (u8 *) tmp.i;
+
+    for (u32 i = 0; i < 5; i++)
+    {
+      for (u32 j = 0; j < tmp.pw_len; j++)
+      {
+        int idx = 15 - (j & 15);
+
+        scratch[idx] ^= input[j];
+      }
+
+      for (u32 j = 0; j < 16; j += 2)
+      {
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[15]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[14]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[13]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[12]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[11]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[10]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 9]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 8]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 7]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 6]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 5]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 4]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 3]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 2]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 1]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 0]]);
+
+        scratch[j]     = (unsigned char)( hash       & 0xff);
+        scratch[j + 1] = (unsigned char)((hash >> 8) & 0xff);
+      }
+    }
+
+    u8 target[16] = { 0 };
+
+    for (u32 i = 0; i < 16; i++)
+    {
+      u8 lower = (scratch[i] & 0x7f);
+
+      if ((lower >= 'A' && lower <= 'Z') || (lower >= 'a' && lower <= 'z'))
+      {
+        target[i] = lower;
+      }
+      else
+      {
+        target[i] = (u8)((scratch[i] >> 4) + 0x61);
+      }
+    }
+
+    u32 *digest = (u32 *) target;
+
+    const u32 r0 = digest[DGST_R0];
+    const u32 r1 = digest[DGST_R1];
+    const u32 r2 = digest[DGST_R2];
+    const u32 r3 = digest[DGST_R3];
+
+    COMPARE_S_SCALAR (r0, r1, r2, r3);
+  }
+}

OpenCL/m26200_a1-pure.cl

@@ -0,0 +1,367 @@
+/**
+ * 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"
+#endif
+
+CONSTANT_VK u32a PE_CONST[256] =
+{
+      0, 49345, 49537,   320, 49921,   960,   640, 49729, 50689,  1728,  1920, 51009,  1280, 50625, 50305,  1088,
+  52225,  3264,  3456, 52545,  3840, 53185, 52865,  3648,  2560, 51905, 52097,  2880, 51457,  2496,  2176, 51265,
+  55297,  6336,  6528, 55617,  6912, 56257, 55937,  6720,  7680, 57025, 57217,  8000, 56577,  7616,  7296, 56385,
+   5120, 54465, 54657,  5440, 55041,  6080,  5760, 54849, 53761,  4800,  4992, 54081,  4352, 53697, 53377,  4160,
+  61441, 12480, 12672, 61761, 13056, 62401, 62081, 12864, 13824, 63169, 63361, 14144, 62721, 13760, 13440, 62529,
+  15360, 64705, 64897, 15680, 65281, 16320, 16000, 65089, 64001, 15040, 15232, 64321, 14592, 63937, 63617, 14400,
+  10240, 59585, 59777, 10560, 60161, 11200, 10880, 59969, 60929, 11968, 12160, 61249, 11520, 60865, 60545, 11328,
+  58369,  9408,  9600, 58689,  9984, 59329, 59009,  9792,  8704, 58049, 58241,  9024, 57601,  8640,  8320, 57409,
+  40961, 24768, 24960, 41281, 25344, 41921, 41601, 25152, 26112, 42689, 42881, 26432, 42241, 26048, 25728, 42049,
+  27648, 44225, 44417, 27968, 44801, 28608, 28288, 44609, 43521, 27328, 27520, 43841, 26880, 43457, 43137, 26688,
+  30720, 47297, 47489, 31040, 47873, 31680, 31360, 47681, 48641, 32448, 32640, 48961, 32000, 48577, 48257, 31808,
+  46081, 29888, 30080, 46401, 30464, 47041, 46721, 30272, 29184, 45761, 45953, 29504, 45313, 29120, 28800, 45121,
+  20480, 37057, 37249, 20800, 37633, 21440, 21120, 37441, 38401, 22208, 22400, 38721, 21760, 38337, 38017, 21568,
+  39937, 23744, 23936, 40257, 24320, 40897, 40577, 24128, 23040, 39617, 39809, 23360, 39169, 22976, 22656, 38977,
+  34817, 18624, 18816, 35137, 19200, 35777, 35457, 19008, 19968, 36545, 36737, 20288, 36097, 19904, 19584, 35905,
+  17408, 33985, 34177, 17728, 34561, 18368, 18048, 34369, 33281, 17088, 17280, 33601, 16640, 33217, 32897, 16448
+};
+
+KERNEL_FQ void m26200_mxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  u32 pw_buf0[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];
+
+  const u32 pw_l_len = pws[gid].pw_len & 63;
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    const u32 pw_r_len = pwlenx_create_combt (combs_buf, il_pos) & 63;
+
+    const u32 pw_len = (pw_l_len + pw_r_len) & 63;
+
+    /**
+     * concat password candidate
+     */
+
+    u32 wordl0[4] = { 0 };
+    u32 wordl1[4] = { 0 };
+    u32 wordl2[4] = { 0 };
+    u32 wordl3[4] = { 0 };
+
+    wordl0[0] = pw_buf0[0];
+    wordl0[1] = pw_buf0[1];
+    wordl0[2] = pw_buf0[2];
+    wordl0[3] = pw_buf0[3];
+
+    u32 wordr0[4] = { 0 };
+    u32 wordr1[4] = { 0 };
+    u32 wordr2[4] = { 0 };
+    u32 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);
+    }
+
+    u32 w0[4];
+    u32 w1[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];
+
+    u32 w[8] = { 0 };
+
+    w[ 0] = w0[0];
+    w[ 1] = w0[1];
+    w[ 2] = w0[2];
+    w[ 3] = w0[3];
+    w[ 4] = w1[0];
+    w[ 5] = w1[1];
+    w[ 6] = w1[2];
+    w[ 7] = w1[3];
+
+    u32 hash = 17;
+
+    u8  scratch[16] = { 0 };
+
+    u8 *input = (u8 *) w;
+
+    for (u32 i = 0; i < 5; i++)
+    {
+      for (u32 j = 0; j < pw_len; j++)
+      {
+        int idx = 15 - (j & 15);
+
+        scratch[idx] ^= input[j];
+      }
+
+      for (u32 j = 0; j < 16; j += 2)
+      {
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[15]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[14]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[13]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[12]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[11]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[10]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 9]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 8]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 7]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 6]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 5]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 4]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 3]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 2]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 1]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 0]]);
+
+        scratch[j]     = (unsigned char)( hash       & 0xff);
+        scratch[j + 1] = (unsigned char)((hash >> 8) & 0xff);
+      }
+    }
+
+    u8 target[16] = { 0 };
+
+    for (u32 i = 0; i < 16; i++)
+    {
+      u8 lower = (scratch[i] & 0x7f);
+
+      if ((lower >= 'A' && lower <= 'Z') || (lower >= 'a' && lower <= 'z'))
+      {
+        target[i] = lower;
+      }
+      else
+      {
+        target[i] = (u8)((scratch[i] >> 4) + 0x61);
+      }
+    }
+
+    u32 *digest = (u32 *) target;
+
+    const u32 r0 = digest[DGST_R0];
+    const u32 r1 = digest[DGST_R1];
+    const u32 r2 = digest[DGST_R2];
+    const u32 r3 = digest[DGST_R3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26200_sxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * 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]
+  };
+
+  /**
+   * base
+   */
+
+  u32 pw_buf0[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];
+
+  const u32 pw_l_len = pws[gid].pw_len & 63;
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+
+    const u32 pw_r_len = pwlenx_create_combt (combs_buf, il_pos) & 63;
+
+    const u32 pw_len = (pw_l_len + pw_r_len) & 63;
+
+    /**
+     * concat password candidate
+     */
+
+    u32 wordl0[4] = { 0 };
+    u32 wordl1[4] = { 0 };
+    u32 wordl2[4] = { 0 };
+    u32 wordl3[4] = { 0 };
+
+    wordl0[0] = pw_buf0[0];
+    wordl0[1] = pw_buf0[1];
+    wordl0[2] = pw_buf0[2];
+    wordl0[3] = pw_buf0[3];
+
+    u32 wordr0[4] = { 0 };
+    u32 wordr1[4] = { 0 };
+    u32 wordr2[4] = { 0 };
+    u32 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);
+    }
+
+    u32 w0[4];
+    u32 w1[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];
+
+    u32 w[8] = { 0 };
+
+    w[ 0] = w0[0];
+    w[ 1] = w0[1];
+    w[ 2] = w0[2];
+    w[ 3] = w0[3];
+    w[ 4] = w1[0];
+    w[ 5] = w1[1];
+    w[ 6] = w1[2];
+    w[ 7] = w1[3];
+
+    u32 hash = 17;
+
+    u8  scratch[16] = { 0 };
+
+    u8 *input = (u8 *) w;
+
+    for (u32 i = 0; i < 5; i++)
+    {
+      for (u32 j = 0; j < pw_len; j++)
+      {
+        int idx = 15 - (j & 15);
+
+        scratch[idx] ^= input[j];
+      }
+
+      for (u32 j = 0; j < 16; j += 2)
+      {
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[15]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[14]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[13]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[12]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[11]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[10]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 9]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 8]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 7]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 6]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 5]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 4]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 3]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 2]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 1]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 0]]);
+
+        scratch[j]     = (unsigned char)( hash       & 0xff);
+        scratch[j + 1] = (unsigned char)((hash >> 8) & 0xff);
+      }
+    }
+
+    u8 target[16] = { 0 };
+
+    for (u32 i = 0; i < 16; i++)
+    {
+      u8 lower = (scratch[i] & 0x7f);
+
+      if ((lower >= 'A' && lower <= 'Z') || (lower >= 'a' && lower <= 'z'))
+      {
+        target[i] = lower;
+      }
+      else
+      {
+        target[i] = (u8)((scratch[i] >> 4) + 0x61);
+      }
+    }
+
+    u32 *digest = (u32 *) target;
+
+    const u32 r0 = digest[DGST_R0];
+    const u32 r1 = digest[DGST_R1];
+    const u32 r2 = digest[DGST_R2];
+    const u32 r3 = digest[DGST_R3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}

OpenCL/m26200_a3-pure.cl

@@ -0,0 +1,254 @@
+/**
+ * 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"
+#endif
+
+CONSTANT_VK u32a PE_CONST[256] =
+{
+      0, 49345, 49537,   320, 49921,   960,   640, 49729, 50689,  1728,  1920, 51009,  1280, 50625, 50305,  1088,
+  52225,  3264,  3456, 52545,  3840, 53185, 52865,  3648,  2560, 51905, 52097,  2880, 51457,  2496,  2176, 51265,
+  55297,  6336,  6528, 55617,  6912, 56257, 55937,  6720,  7680, 57025, 57217,  8000, 56577,  7616,  7296, 56385,
+   5120, 54465, 54657,  5440, 55041,  6080,  5760, 54849, 53761,  4800,  4992, 54081,  4352, 53697, 53377,  4160,
+  61441, 12480, 12672, 61761, 13056, 62401, 62081, 12864, 13824, 63169, 63361, 14144, 62721, 13760, 13440, 62529,
+  15360, 64705, 64897, 15680, 65281, 16320, 16000, 65089, 64001, 15040, 15232, 64321, 14592, 63937, 63617, 14400,
+  10240, 59585, 59777, 10560, 60161, 11200, 10880, 59969, 60929, 11968, 12160, 61249, 11520, 60865, 60545, 11328,
+  58369,  9408,  9600, 58689,  9984, 59329, 59009,  9792,  8704, 58049, 58241,  9024, 57601,  8640,  8320, 57409,
+  40961, 24768, 24960, 41281, 25344, 41921, 41601, 25152, 26112, 42689, 42881, 26432, 42241, 26048, 25728, 42049,
+  27648, 44225, 44417, 27968, 44801, 28608, 28288, 44609, 43521, 27328, 27520, 43841, 26880, 43457, 43137, 26688,
+  30720, 47297, 47489, 31040, 47873, 31680, 31360, 47681, 48641, 32448, 32640, 48961, 32000, 48577, 48257, 31808,
+  46081, 29888, 30080, 46401, 30464, 47041, 46721, 30272, 29184, 45761, 45953, 29504, 45313, 29120, 28800, 45121,
+  20480, 37057, 37249, 20800, 37633, 21440, 21120, 37441, 38401, 22208, 22400, 38721, 21760, 38337, 38017, 21568,
+  39937, 23744, 23936, 40257, 24320, 40897, 40577, 24128, 23040, 39617, 39809, 23360, 39169, 22976, 22656, 38977,
+  34817, 18624, 18816, 35137, 19200, 35777, 35457, 19008, 19968, 36545, 36737, 20288, 36097, 19904, 19584, 35905,
+  17408, 33985, 34177, 17728, 34561, 18368, 18048, 34369, 33281, 17088, 17280, 33601, 16640, 33217, 32897, 16448
+};
+
+KERNEL_FQ void m26200_mxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  /**
+   * loop
+   */
+
+  u32 w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32 w0r = ix_create_bft (bfs_buf, il_pos);
+
+    const u32 w0 = w0l | w0r;
+
+    w[0] = w0;
+
+    u32 hash = 17;
+
+    u8  scratch[16] = { 0 };
+
+    u8 *input = (u8 *) w;
+
+    for (u32 i = 0; i < 5; i++)
+    {
+      for (u32 j = 0; j < pw_len; j++)
+      {
+        int idx = 15 - (j & 15);
+
+        scratch[idx] ^= input[j];
+      }
+
+      for (u32 j = 0; j < 16; j += 2)
+      {
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[15]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[14]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[13]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[12]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[11]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[10]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 9]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 8]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 7]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 6]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 5]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 4]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 3]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 2]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 1]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 0]]);
+
+        scratch[j]     = (unsigned char)( hash       & 0xff);
+        scratch[j + 1] = (unsigned char)((hash >> 8) & 0xff);
+      }
+    }
+
+    u8 target[16] = { 0 };
+
+    for (u32 i = 0; i < 16; i++)
+    {
+      u8 lower = (scratch[i] & 0x7f);
+
+      if ((lower >= 'A' && lower <= 'Z') || (lower >= 'a' && lower <= 'z'))
+      {
+        target[i] = lower;
+      }
+      else
+      {
+        target[i] = (u8)((scratch[i] >> 4) + 0x61);
+      }
+    }
+
+    u32 *digest = (u32 *) target;
+
+    const u32 r0 = digest[DGST_R0];
+    const u32 r1 = digest[DGST_R1];
+    const u32 r2 = digest[DGST_R2];
+    const u32 r3 = digest[DGST_R3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26200_sxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * 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]
+  };
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32 w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  /**
+   * loop
+   */
+
+  u32 w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
+  {
+    const u32 w0r = ix_create_bft (bfs_buf, il_pos);
+
+    const u32 w0 = w0l | w0r;
+
+    w[0] = w0;
+
+    u32 hash = 17;
+
+    u8  scratch[16] = { 0 };
+
+    u8 *input = (u8 *) w;
+
+    for (u32 i = 0; i < 5; i++)
+    {
+      for (u32 j = 0; j < pw_len; j++)
+      {
+        int idx = 15 - (j & 15);
+
+        scratch[idx] ^= input[j];
+      }
+
+      for (u32 j = 0; j < 16; j += 2)
+      {
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[15]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[14]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[13]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[12]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[11]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[10]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 9]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 8]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 7]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 6]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 5]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 4]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 3]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 2]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 1]]);
+        hash = (hash >> 8 ^ PE_CONST[hash & 0xff] ^ PE_CONST[scratch[ 0]]);
+
+        scratch[j]     = (unsigned char)( hash       & 0xff);
+        scratch[j + 1] = (unsigned char)((hash >> 8) & 0xff);
+      }
+    }
+
+    u8 target[16] = { 0 };
+
+    for (u32 i = 0; i < 16; i++)
+    {
+      u8 lower = (scratch[i] & 0x7f);
+
+      if ((lower >= 'A' && lower <= 'Z') || (lower >= 'a' && lower <= 'z'))
+      {
+        target[i] = lower;
+      }
+      else
+      {
+        target[i] = (u8)((scratch[i] >> 4) + 0x61);
+      }
+    }
+
+    u32 *digest = (u32 *) target;
+
+    const u32 r0 = digest[DGST_R0];
+    const u32 r1 = digest[DGST_R1];
+    const u32 r2 = digest[DGST_R2];
+    const u32 r3 = digest[DGST_R3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}

OpenCL/m26300_a0-pure.cl

@@ -0,0 +1,181 @@
+/**
+ * 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.h"
+#include "inc_rp.cl"
+#include "inc_scalar.cl"
+#include "inc_hash_sha256.cl"
+#endif
+
+KERNEL_FQ void m26300_mxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  sha256_ctx_t ctx0;
+
+  sha256_init (&ctx0);
+
+  sha256_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    pw_t tmp = PASTE_PW;
+
+    tmp.pw_len = apply_rules (rules_buf[il_pos].cmds, tmp.i, tmp.pw_len);
+
+    sha256_ctx_t ctx = ctx0;
+
+    sha256_update_swap (&ctx, tmp.i, tmp.pw_len);
+
+    /**
+     * pepper
+     */
+
+    u32 p0[4];
+    u32 p1[4];
+    u32 p2[4];
+    u32 p3[4];
+
+    p0[0] = hc_swap32_S (FORTIGATE_A);
+    p0[1] = hc_swap32_S (FORTIGATE_B);
+    p0[2] = hc_swap32_S (FORTIGATE_C);
+    p0[3] = hc_swap32_S (FORTIGATE_D);
+    p1[0] = hc_swap32_S (FORTIGATE_E);
+    p1[1] = hc_swap32_S (FORTIGATE_F);
+    p1[2] = 0;
+    p1[3] = 0;
+    p2[0] = 0;
+    p2[1] = 0;
+    p2[2] = 0;
+    p2[3] = 0;
+    p3[0] = 0;
+    p3[1] = 0;
+    p3[2] = 0;
+    p3[3] = 0;
+
+    sha256_update_64 (&ctx, p0, p1, p2, p3, 24);
+
+    sha256_final (&ctx);
+
+    const u32 r0 = ctx.h[DGST_R0];
+    const u32 r1 = ctx.h[DGST_R1];
+    const u32 r2 = ctx.h[DGST_R2];
+    const u32 r3 = ctx.h[DGST_R3];
+
+    COMPARE_M_SCALAR (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26300_sxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * 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]
+  };
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  sha256_ctx_t ctx0;
+
+  sha256_init (&ctx0);
+
+  sha256_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    pw_t tmp = PASTE_PW;
+
+    tmp.pw_len = apply_rules (rules_buf[il_pos].cmds, tmp.i, tmp.pw_len);
+
+    sha256_ctx_t ctx = ctx0;
+
+    sha256_update_swap (&ctx, tmp.i, tmp.pw_len);
+
+    /**
+     * pepper
+     */
+
+    u32 p0[4];
+    u32 p1[4];
+    u32 p2[4];
+    u32 p3[4];
+
+    p0[0] = hc_swap32_S (FORTIGATE_A);
+    p0[1] = hc_swap32_S (FORTIGATE_B);
+    p0[2] = hc_swap32_S (FORTIGATE_C);
+    p0[3] = hc_swap32_S (FORTIGATE_D);
+    p1[0] = hc_swap32_S (FORTIGATE_E);
+    p1[1] = hc_swap32_S (FORTIGATE_F);
+    p1[2] = 0;
+    p1[3] = 0;
+    p2[0] = 0;
+    p2[1] = 0;
+    p2[2] = 0;
+    p2[3] = 0;
+    p3[0] = 0;
+    p3[1] = 0;
+    p3[2] = 0;
+    p3[3] = 0;
+
+    sha256_update_64 (&ctx, p0, p1, p2, p3, 24);
+
+    sha256_final (&ctx);
+
+    const u32 r0 = ctx.h[DGST_R0];
+    const u32 r1 = ctx.h[DGST_R1];
+    const u32 r2 = ctx.h[DGST_R2];
+    const u32 r3 = ctx.h[DGST_R3];
+
+    COMPARE_S_SCALAR (r0, r1, r2, r3);
+  }
+}

OpenCL/m26300_a1-pure.cl

@@ -0,0 +1,171 @@
+/**
+ * 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_hash_sha256.cl"
+#endif
+
+KERNEL_FQ void m26300_mxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  sha256_ctx_t ctx0;
+
+  sha256_init (&ctx0);
+
+  sha256_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+
+  sha256_update_global_swap (&ctx0, pws[gid].i, pws[gid].pw_len);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    sha256_ctx_t ctx = ctx0;
+
+    sha256_update_global_swap (&ctx, combs_buf[il_pos].i, combs_buf[il_pos].pw_len);
+
+    /**
+     * pepper
+     */
+
+    u32 p0[4];
+    u32 p1[4];
+    u32 p2[4];
+    u32 p3[4];
+
+    p0[0] = hc_swap32_S (FORTIGATE_A);
+    p0[1] = hc_swap32_S (FORTIGATE_B);
+    p0[2] = hc_swap32_S (FORTIGATE_C);
+    p0[3] = hc_swap32_S (FORTIGATE_D);
+    p1[0] = hc_swap32_S (FORTIGATE_E);
+    p1[1] = hc_swap32_S (FORTIGATE_F);
+    p1[2] = 0;
+    p1[3] = 0;
+    p2[0] = 0;
+    p2[1] = 0;
+    p2[2] = 0;
+    p2[3] = 0;
+    p3[0] = 0;
+    p3[1] = 0;
+    p3[2] = 0;
+    p3[3] = 0;
+
+    sha256_update_64 (&ctx, p0, p1, p2, p3, 24);
+
+    sha256_final (&ctx);
+
+    const u32 r0 = ctx.h[DGST_R0];
+    const u32 r1 = ctx.h[DGST_R1];
+    const u32 r2 = ctx.h[DGST_R2];
+    const u32 r3 = ctx.h[DGST_R3];
+
+    COMPARE_M_SCALAR (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26300_sxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * 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]
+  };
+
+  /**
+   * base
+   */
+
+  sha256_ctx_t ctx0;
+
+  sha256_init (&ctx0);
+
+  sha256_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+
+  sha256_update_global_swap (&ctx0, pws[gid].i, pws[gid].pw_len);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  {
+    sha256_ctx_t ctx = ctx0;
+
+    sha256_update_global_swap (&ctx, combs_buf[il_pos].i, combs_buf[il_pos].pw_len);
+
+    /**
+     * pepper
+     */
+
+    u32 p0[4];
+    u32 p1[4];
+    u32 p2[4];
+    u32 p3[4];
+
+    p0[0] = hc_swap32_S (FORTIGATE_A);
+    p0[1] = hc_swap32_S (FORTIGATE_B);
+    p0[2] = hc_swap32_S (FORTIGATE_C);
+    p0[3] = hc_swap32_S (FORTIGATE_D);
+    p1[0] = hc_swap32_S (FORTIGATE_E);
+    p1[1] = hc_swap32_S (FORTIGATE_F);
+    p1[2] = 0;
+    p1[3] = 0;
+    p2[0] = 0;
+    p2[1] = 0;
+    p2[2] = 0;
+    p2[3] = 0;
+    p3[0] = 0;
+    p3[1] = 0;
+    p3[2] = 0;
+    p3[3] = 0;
+
+    sha256_update_64 (&ctx, p0, p1, p2, p3, 24);
+
+    sha256_final (&ctx);
+
+    const u32 r0 = ctx.h[DGST_R0];
+    const u32 r1 = ctx.h[DGST_R1];
+    const u32 r2 = ctx.h[DGST_R2];
+    const u32 r3 = ctx.h[DGST_R3];
+
+    COMPARE_S_SCALAR (r0, r1, r2, r3);
+  }
+}

OpenCL/m26300_a3-pure.cl

@@ -0,0 +1,205 @@
+/**
+ * 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_hash_sha256.cl"
+#endif
+
+KERNEL_FQ void m26300_mxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32x w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  sha256_ctx_t ctx0;
+
+  sha256_init (&ctx0);
+
+  sha256_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+
+  /**
+   * loop
+   */
+
+  u32x 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;
+
+    w[0] = w0;
+
+    sha256_ctx_vector_t ctx;
+
+    sha256_init_vector_from_scalar (&ctx, &ctx0);
+
+    sha256_update_vector (&ctx, w, pw_len);
+
+    /**
+     * pepper
+     */
+
+    u32x p0[4];
+    u32x p1[4];
+    u32x p2[4];
+    u32x p3[4];
+
+    p0[0] = hc_swap32 (FORTIGATE_A);
+    p0[1] = hc_swap32 (FORTIGATE_B);
+    p0[2] = hc_swap32 (FORTIGATE_C);
+    p0[3] = hc_swap32 (FORTIGATE_D);
+    p1[0] = hc_swap32 (FORTIGATE_E);
+    p1[1] = hc_swap32 (FORTIGATE_F);
+    p1[2] = 0;
+    p1[3] = 0;
+    p2[0] = 0;
+    p2[1] = 0;
+    p2[2] = 0;
+    p2[3] = 0;
+    p3[0] = 0;
+    p3[1] = 0;
+    p3[2] = 0;
+    p3[3] = 0;
+
+    sha256_update_vector_64 (&ctx, p0, p1, p2, p3, 24);
+
+    sha256_final_vector (&ctx);
+
+    const u32x r0 = ctx.h[DGST_R0];
+    const u32x r1 = ctx.h[DGST_R1];
+    const u32x r2 = ctx.h[DGST_R2];
+    const u32x r3 = ctx.h[DGST_R3];
+
+    COMPARE_M_SIMD (r0, r1, r2, r3);
+  }
+}
+
+KERNEL_FQ void m26300_sxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= gid_max) return;
+
+  /**
+   * 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]
+  };
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32x w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  sha256_ctx_t ctx0;
+
+  sha256_init (&ctx0);
+
+  sha256_update_global_swap (&ctx0, salt_bufs[SALT_POS].salt_buf, salt_bufs[SALT_POS].salt_len);
+
+  /**
+   * loop
+   */
+
+  u32x 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;
+
+    w[0] = w0;
+
+    sha256_ctx_vector_t ctx;
+
+    sha256_init_vector_from_scalar (&ctx, &ctx0);
+
+    sha256_update_vector (&ctx, w, pw_len);
+
+    /**
+     * pepper
+     */
+
+    u32x p0[4];
+    u32x p1[4];
+    u32x p2[4];
+    u32x p3[4];
+
+    p0[0] = hc_swap32 (FORTIGATE_A);
+    p0[1] = hc_swap32 (FORTIGATE_B);
+    p0[2] = hc_swap32 (FORTIGATE_C);
+    p0[3] = hc_swap32 (FORTIGATE_D);
+    p1[0] = hc_swap32 (FORTIGATE_E);
+    p1[1] = hc_swap32 (FORTIGATE_F);
+    p1[2] = 0;
+    p1[3] = 0;
+    p2[0] = 0;
+    p2[1] = 0;
+    p2[2] = 0;
+    p2[3] = 0;
+    p3[0] = 0;
+    p3[1] = 0;
+    p3[2] = 0;
+    p3[3] = 0;
+
+    sha256_update_vector_64 (&ctx, p0, p1, p2, p3, 24);
+
+    sha256_final_vector (&ctx);
+
+    const u32x r0 = ctx.h[DGST_R0];
+    const u32x r1 = ctx.h[DGST_R1];
+    const u32x r2 = ctx.h[DGST_R2];
+    const u32x r3 = ctx.h[DGST_R3];
+
+    COMPARE_S_SIMD (r0, r1, r2, r3);
+  }
+}

charsets/special/Russian/ru_cp866-special.hcchr

@@ -0,0 +1 @@
+€亗儎厗噲墛媽崕彁憭摂晼棙櫄洔潪煚、￥ウЖ┆<EFBFBD><EFBFBD><EFBFBD>徕沅彐玷殛腱眍镳顸

charsets/standard/Russian/ru_cp1251.hcchr

@@ -1 +1 @@
-ǜ估谅媚牌侨墒颂臀闲岩釉罩棕仝圮蒉哙徕沅彐玷殛腱眍镳耱篝貊鼬<EFBFBD><EFBFBD><EFBFBD><EFBFBD>
+ǜ懒旅呐魄壬仕掏蜗醒矣哉肿刭谯茌捱噌忏溴骁栝觌祉铒瘃蝮趱鲼<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>

charsets/standard/Russian/ru_cp866.hcchr

@@ -0,0 +1 @@
+€亗儎厗噲墛媽崕彁憭摂晼棙櫄洔潪煚、￥ウЖ┆<EFBFBD><EFBFBD><EFBFBD>徕沅彐玷殛腱眍镳<EFBFBD>

docs/changes.txt

@@ -1,4 +1,44 @@
-* changes v6.2.1 -> v6.x.x
+* changes v6.2.2 -> v6.2.x
+
+##
+## Algorithms
+##
+
+- 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
+
+##
+## Bugs
+##
+
+- Fixed 'E' rule in pure kernel mode which was ignoring letters that are in positions that are multiples of 4
+- Fixed false negative in hash-mode 15900 (DPAPI masterkey file v2) if password was longer than 64 characters
+- Fixed integer overflow in Recovered/Time status view column caused by division > 0 but < 1
+
+##
+## Features
+##
+
+- Added temperature watchdog and fanspeed readings for CPU and GPU on macOS using iokit
+
+##
+## Improvements
+##
+
+- Alias Devices: Show a warning in case the user specifically listed a device to use which in a later step is skipped because it is an alias of another active device
+- Backend Information: Added new column showing the PCI Address per CUDA/OpenCL device to easier identify broken cards
+- VeraCrypt: Increase password length support for non-boot volumes from 64 to 128
+
+##
+## Technical
+##
+
+- Autodetect: Limit the number of errors per hash-mode try to 100 to avoid long startup time
+- LM: Workaround JiT compiler bug in -m 3000 on NV leading to false negatives with large amount of hashes
+- Tests: Changed tests for VeraCrypt from -a 0 to -a 3, because password extension is not available to all shells
+
+* changes v6.2.1 -> v6.2.2
 
 ##
 ## Algorithms
@@ -6,8 +46,10 @@
 
 - Added hash-mode: bcrypt(md5($pass)) / bcryptmd5
 - Added hash-mode: bcrypt(sha1($pass)) / bcryptsha1
+- Added hash-mode: FortiGate256 (FortiOS256)
 - Added hash-mode: Linux Kernel Crypto API (2.4)
 - Added hash-mode: MurmurHash
+- Added hash-mode: OpenEdge Progress Encode
 - Added hash-mode: md5(utf16le($pass))
 - Added hash-mode: sha1(utf16le($pass))
 - Added hash-mode: sha256(utf16le($pass))
@@ -25,29 +67,33 @@
 ## Features
 ##
 
-- Autodetect hash-type: performs an automatic analysis of the input hash(es), associating compatible algorithms, or executing the attack if only one compatible format is found.
+- Autodetect hash-type: performs an automatic analysis of input hash(es), either listing compatible algorithms, or executing the attack (if only one compatible format is found)
+- Autodetect hash-type: added option --identify to only perform autodetection of hash-type, without back-end device initialization
 
 ##
 ## Bugs
 ##
 
-- Fixed error message in -a 9 mode with rules in case number of words from wordlist are not in sync with number of unique salts
-- Fixed false negatives with TrueCrypt/VeraCrypt in case zip or gzip compressed files were used as keyfiles
-- Fixed error handling logic in monitor thread to not return in case of error (disc full, permission error, ...) but to retry instead
+- Fixed error message in -a 9 mode with rules when number of words from wordlist is not in sync with number of unique salts
+- Fixed error-handling logic in monitor thread to not return in case of error (disk full, permission error, ...) but to retry instead
+- Fixed false negatives with TrueCrypt/VeraCrypt when zip- or gzip-compressed files were used as keyfiles
+- Fixed free memory-size output for skipped GPU (both automatic and manual) of --backend-info information screen
 
 ##
 ## Improvements
 ##
 
 - AES Crypt Plugin: Reduced max password length from 256 to 128 which improved performance by 22%
-- CUDA Backend: Do not warn about missing CUDA SDK installation if --stdout is used
-- Folder Management: Add support for XDG Base Directory specification if hashcat was installed using make install
-- OpenCL Backend: Use CL_DEVICE_BOARD_NAME_AMD instead of CL_DEVICE_NAME for device name in case OpenCL runtime supports this query
-- Performance Monitor: Add -S as a user suggestion to improve cracking performance in specific attack configurations
+- CUDA Backend: If --stdout is used, do not warn about missing CUDA SDK
+- Folder Management: Add support for XDG Base Directory specification when hashcat is installed using 'make install'
+- Hardware Monitor: Add support for GPU device utilization readings from sysfs (AMD on Linux)
+- OpenCL Backend: Use CL_DEVICE_BOARD_NAME_AMD instead of CL_DEVICE_NAME for device name (when supported by OpenCL runtime)
+- Performance Monitor: Suggest -S to improve cracking performance in specific attack configurations
 - RAR3-p (Compressed): Fix workaround in unrar library in AES constant table generation to enable multi-threading support
-- RC4 Kernels: Improved performance by 20%+ for hash-modes Kerberos 5 (etype 23), MS Office (<= 2003) and PDF (<= 1.6) by using new RC4 code
-- Status Screen: Show currently running kernel type (pure, optimized) and generator type (host, device)
-- UTF8-to-UTF16: Replaced naive UTF8 to UTF16 conversion with true conversion for RAR3, AES Crypt, MultiBit HD (scrypt) and Umbraco HMAC-SHA1
+- RC4 Kernels: Improved performance by 20%+ with new RC4 code in hash-modes Kerberos 5 (etype 23), MS Office (<= 2003) and PDF (<= 1.6)
+- Scrypt Kernels: Re-enable use of kernel cache in scrypt-based kernels
+- Status Screen: Show currently-running kernel type (pure, optimized) and generator type (host, device)
+- UTF8-to-UTF16: Replaced naive UTF8-to-UTF16 conversion with true conversion for RAR3, AES Crypt, MultiBit HD (scrypt) and Umbraco HMAC-SHA1
 
 ##
 ## Technical
@@ -56,10 +102,15 @@
 - Dependencies: Updated LZMA SDK from 19.00 to 21.02 alpha
 - Dependencies: Updated xxHash from 0.1.0 to v0.8.0 - Stable XXH3
 - Documentation: Update missing documentation in plugin developer guide for OPTS_TYPE_MP_MULTI_DISABLE and OPTS_TYPE_NATIVE_THREADS
+- Hashrate: Update inner-loop hashrate prediction to handle new salt_repeats feature and also respect _loop2 kernel runtime
 - Kernels: Add RC4 cipher to crypto library with optimized shared memory access pattern which will not cause any bank conflicts if -u <= 32
-- Kernels: Add standalone true UTF8 to UTF16 converter kernel that runs after amplifier. Use OPTS_TYPE_POST_AMP_UTF16LE from plugin
+- Kernels: Add standalone true UTF8-to-UTF16 converter kernel that runs after amplifier. Use OPTS_TYPE_POST_AMP_UTF16LE from plugin
+- Kernel Cache: Add module_jit_build_options() string from modules to kernel cache checksum calculation
 - Modules: Recategorized HASH_CATEGORY option in various modules
 - Modules: Added hash categories HASH_CATEGORY_IMS and HASH_CATEGORY_CRYPTOCURRENCY_WALLET
+- Modules: Changed hash category of Python passlib from HASH_CATEGORY_GENERIC_KDF to HASH_CATEGORY_FRAMEWORK
+- Unit-Tests: Added missing support for OPTS_TYPE_LOOP_PREPARE, OPTS_TYPE_LOOP_PREPARE2 and salt_repeats in self-test
+- Unit-Tests: Added missing unit-test for Stargazer Stellar Wallet XLM
 
 * changes v6.2.0 -> v6.2.1
 

docs/credits.txt

@@ -22,7 +22,7 @@ Gabriele "matrix" Gristina <matrix@hashcat.net> (@gm4tr1x)
 * Compressed wordlist feature
 * OpenCL Info feature
 * Apple macOS port
-* Hardware monitor initial code base
+* Hardware monitor initial code base and maintenance
 * Test suite initial code base
 * Makefile initial code base
 * Multithreading initial code base

docs/limits.txt

@@ -26,6 +26,8 @@ It only means the filename itself.
 ## Hashing algorithms that internally use UTF-16 characters could in special cases lead to false negatives
 ##
 
+For optimized kernels only (-O): Since hashcat v6.2.1 there's true UTF16 support for pure kernels:
+
 The UTF-16 conversion implementation used within the kernel code is very elementary and for performance
 reasons does not respect all complicated encoding rules required to correctly convert, for instance, ASCII
 or UTF-8 to UTF-16LE (or UTF-16BE).
@@ -33,6 +35,8 @@ or UTF-8 to UTF-16LE (or UTF-16BE).
 The implementation most likely fails with multi-byte characters, because we basically add a zero byte every
 second byte within the kernel conversion code.
 
+Since hashcat v6.2.1 there's true UTF16 support for pure kernels.
+
 ##
 ## The use of --keep-guessing eventually skips reporting duplicate passwords
 ##

docs/readme.txt

@@ -7,7 +7,7 @@
            |_____|     /___________/     |_____|  /_____/        /_______|
                  :                             :                         :
 
-hashcat v6.2.1
+hashcat v6.2.2
 ==============
 
 AMD GPUs on Linux require "RadeonOpenCompute (ROCm)" Software Platform (3.1 or later)
@@ -146,9 +146,6 @@ NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or
 - PBKDF2-HMAC-SHA1
 - PBKDF2-HMAC-SHA256
 - PBKDF2-HMAC-SHA512
-- Python passlib pbkdf2-sha512
-- Python passlib pbkdf2-sha256
-- Python passlib pbkdf2-sha1
 - scrypt
 - phpass
 - TACACS+
@@ -202,6 +199,7 @@ NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or
 - Domain Cached Credentials (DCC), MS Cache
 - Domain Cached Credentials 2 (DCC2), MS Cache 2
 - FortiGate (FortiOS)
+- FortiGate256 (FortiOS256)
 - ArubaOS
 - Juniper IVE
 - Juniper NetScreen/SSG (ScreenOS)
@@ -258,6 +256,7 @@ NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or
 - Lotus Notes/Domino 5
 - Lotus Notes/Domino 6
 - Lotus Notes/Domino 8
+- OpenEdge Progress Encode
 - Oracle Transportation Management (SHA256)
 - Huawei sha1(md5($pass).$salt)
 - AuthMe sha256
@@ -355,6 +354,9 @@ NVIDIA GPUs require "NVIDIA Driver" (440.64 or later) and "CUDA Toolkit" (9.0 or
 - Django (SHA-1)
 - Atlassian (PBKDF2-HMAC-SHA1)
 - Ruby on Rails Restful-Authentication
+- Python passlib pbkdf2-sha512
+- Python passlib pbkdf2-sha256
+- Python passlib pbkdf2-sha1
 - PKCS#8 Private Keys (PBKDF2-HMAC-SHA1 + 3DES/AES)
 - PKCS#8 Private Keys (PBKDF2-HMAC-SHA256 + 3DES/AES)
 - JKS Java Key Store Private Keys (SHA1)

extra/tab_completion/hashcat.sh

@@ -411,7 +411,7 @@ _hashcat_contains ()
 
 _hashcat ()
 {
-  local VERSION=6.2.1
+  local VERSION=6.2.2
 
   local ATTACK_MODES="0 1 3 6 7 9"
   local HCCAPX_MESSAGE_PAIRS="0 1 2 3 4 5"
@@ -426,7 +426,7 @@ _hashcat ()
   local BUILD_IN_CHARSETS='?l ?u ?d ?a ?b ?s ?h ?H'
 
   local SHORT_OPTS="-m -a -V -h -b -t -T -o -p -c -d -D -w -n -u -j -k -r -g -1 -2 -3 -4 -i -I -s -l -O -S -z"
-  local LONG_OPTS="--hash-type --attack-mode --version --help --quiet --benchmark --benchmark-all --hex-salt --hex-wordlist --hex-charset --force --status --status-json --status-timer --stdin-timeout-abort --machine-readable --loopback --markov-hcstat2 --markov-disable --markov-classic --markov-threshold --runtime --session --speed-only --progress-only --restore --restore-file-path --restore-disable --outfile --outfile-format --outfile-autohex-disable --outfile-check-timer --outfile-check-dir --wordlist-autohex-disable --separator --show --left --username --remove --remove-timer --potfile-disable --potfile-path --debug-mode --debug-file --induction-dir --segment-size --bitmap-min --bitmap-max --cpu-affinity --example-hashes --hash-info --backend-ignore-cuda --backend-ignore-opencl --backend-info --backend-devices --opencl-device-types --backend-vector-width --workload-profile --kernel-accel --kernel-loops --kernel-threads --spin-damp --hwmon-disable --hwmon-temp-abort --skip --limit --keyspace --rule-left --rule-right --rules-file --generate-rules --generate-rules-func-min --generate-rules-func-max --generate-rules-seed --custom-charset1 --custom-charset2 --custom-charset3 --custom-charset4 --hook-threads --increment --increment-min --increment-max --logfile-disable --scrypt-tmto --keyboard-layout-mapping --truecrypt-keyfiles --veracrypt-keyfiles --veracrypt-pim-start --veracrypt-pim-stop --stdout --keep-guessing --hccapx-message-pair --nonce-error-corrections --encoding-from --encoding-to --optimized-kernel-enable --self-test-disable  --slow-candidates --brain-server --brain-server-timer --brain-client --brain-client-features --brain-host --brain-port --brain-session --brain-session-whitelist --brain-password"
+  local LONG_OPTS="--hash-type --attack-mode --version --help --quiet --benchmark --benchmark-all --hex-salt --hex-wordlist --hex-charset --force --status --status-json --status-timer --stdin-timeout-abort --machine-readable --loopback --markov-hcstat2 --markov-disable --markov-classic --markov-threshold --runtime --session --speed-only --progress-only --restore --restore-file-path --restore-disable --outfile --outfile-format --outfile-autohex-disable --outfile-check-timer --outfile-check-dir --wordlist-autohex-disable --separator --show --left --username --remove --remove-timer --potfile-disable --potfile-path --debug-mode --debug-file --induction-dir --segment-size --bitmap-min --bitmap-max --cpu-affinity --example-hashes --hash-info --backend-ignore-cuda --backend-ignore-opencl --backend-info --backend-devices --opencl-device-types --backend-vector-width --workload-profile --kernel-accel --kernel-loops --kernel-threads --spin-damp --hwmon-disable --hwmon-temp-abort --skip --limit --keyspace --rule-left --rule-right --rules-file --generate-rules --generate-rules-func-min --generate-rules-func-max --generate-rules-seed --custom-charset1 --custom-charset2 --custom-charset3 --custom-charset4 --hook-threads --increment --increment-min --increment-max --logfile-disable --scrypt-tmto --keyboard-layout-mapping --truecrypt-keyfiles --veracrypt-keyfiles --veracrypt-pim-start --veracrypt-pim-stop --stdout --keep-guessing --hccapx-message-pair --nonce-error-corrections --encoding-from --encoding-to --optimized-kernel-enable --self-test-disable  --slow-candidates --brain-server --brain-server-timer --brain-client --brain-client-features --brain-host --brain-port --brain-session --brain-session-whitelist --brain-password --identify"
   local OPTIONS="-m -a -t -o -p -c -d -w -n -u -j -k -r -g -1 -2 -3 -4 -s -l --hash-type --attack-mode --status-timer --stdin-timeout-abort --markov-hcstat2 --markov-threshold --runtime --session --timer --outfile --outfile-format --outfile-check-timer --outfile-check-dir --separator --remove-timer --potfile-path --restore-file-path --debug-mode --debug-file --induction-dir --segment-size --bitmap-min --bitmap-max --cpu-affinity --backend-devices --opencl-device-types --backend-vector-width --workload-profile --kernel-accel --kernel-loops --kernel-threads --spin-damp --hwmon-temp-abort --skip --limit --rule-left --rule-right --rules-file --generate-rules --generate-rules-func-min --generate-rules-func-max --generate-rules-seed --custom-charset1 --custom-charset2 --custom-charset3 --custom-charset4 --hook-threads --increment-min --increment-max --scrypt-tmto --keyboard-layout-mapping --truecrypt-keyfiles --veracrypt-keyfiles --veracrypt-pim-start --veracrypt-pim-stop --hccapx-message-pair --nonce-error-corrections --encoding-from --encoding-to --brain-server-timer --brain-client-features --brain-host --brain-password --brain-port --brain-session --brain-session-whitelist"
 
   COMPREPLY=()

include/cpu_crc32.h

@@ -9,7 +9,7 @@
 #include <stdio.h>
 #include <errno.h>
 
-int cpu_crc32        (const char *filename, u8 keytab[64]);
+int cpu_crc32 (const char *filename, u8 *keytab, const size_t keytabsz);
 u32 cpu_crc32_buffer (const u8 *buf, const size_t length);
 
 #endif // _CPU_CRC32_H

include/ext_ADL.h

@@ -314,4 +314,18 @@ typedef hm_adl_lib_t ADL_PTR;
 
 void *HC_API_CALL ADL_Main_Memory_Alloc (const int iSize);
 
+int adl_init (void *hashcat_ctx);
+void adl_close (void *hashcat_ctx);
+int hm_ADL_Main_Control_Destroy (void *hashcat_ctx);
+int hm_ADL_Main_Control_Create (void *hashcat_ctx, ADL_MAIN_MALLOC_CALLBACK callback, int iEnumConnectedAdapters);
+int hm_ADL_Adapter_NumberOfAdapters_Get (void *hashcat_ctx, int *lpNumAdapters);
+int hm_ADL_Adapter_AdapterInfo_Get (void *hashcat_ctx, LPAdapterInfo lpInfo, int iInputSize);
+int hm_ADL_Overdrive5_Temperature_Get (void *hashcat_ctx, int iAdapterIndex, int iThermalControllerIndex, ADLTemperature *lpTemperature);
+int hm_ADL_Overdrive6_Temperature_Get (void *hashcat_ctx, int iAdapterIndex, int *iTemperature);
+int hm_ADL_Overdrive_CurrentActivity_Get (void *hashcat_ctx, int iAdapterIndex, ADLPMActivity *lpActivity);
+int hm_ADL_Overdrive5_FanSpeed_Get (void *hashcat_ctx, int iAdapterIndex, int iThermalControllerIndex, ADLFanSpeedValue *lpFanSpeedValue);
+int hm_ADL_Overdrive6_FanSpeed_Get (void *hashcat_ctx, int iAdapterIndex, ADLOD6FanSpeedInfo *lpFanSpeedInfo);
+int hm_ADL_Overdrive_Caps (void *hashcat_ctx, int iAdapterIndex, int *od_supported, int *od_enabled, int *od_version);
+int hm_ADL_Overdrive6_TargetTemperatureData_Get (void *hashcat_ctx, int iAdapterIndex, int *cur_temp, int *default_temp);
+
 #endif // _EXT_ADL_H

include/ext_iokit.h

@@ -0,0 +1,128 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#ifndef _EXT_IOKIT_H
+#define _EXT_IOKIT_H
+
+#if defined(__APPLE__)
+#include <IOKit/IOKitLib.h>
+
+// Apple SMC Keys
+#define HM_IOKIT_SMC_SENSOR_GRAPHICS_HOT "SGHT"
+#define HM_IOKIT_SMC_CPU_PROXIMITY       "TC0P"
+#define HM_IOKIT_SMC_GPU_PROXIMITY       "TG0P"
+#define HM_IOKIT_SMC_PECI_GPU            "TCGC"
+
+#define KERNEL_INDEX_SMC 2
+
+#define DATATYPE_FPE2   "fpe2"
+#define DATATYPE_FLT    "flt "
+#define DATATYPE_UINT8  "ui8 "
+#define DATATYPE_UINT16 "ui16"
+#define DATATYPE_UINT32 "ui32"
+#define DATATYPE_SP78   "sp78"
+
+typedef enum
+{
+  SMC_CMD_READ_BYTES   =  5,
+  SMC_CMD_WRITE_BYTES  =  6,
+  SMC_CMD_READ_INDEX   =  8,
+  SMC_CMD_READ_KEYINFO =  9,
+  SMC_CMD_READ_PLIMIT  = 11,
+  SMC_CMD_READ_VERS    = 12
+
+} SMCCommands_t;
+
+typedef struct
+{
+  char   major;
+  char   minor;
+  char   build;
+  char   reserved[1];
+  UInt16 release;
+
+} SMCKeyData_vers_t;
+
+typedef struct
+{
+  UInt16 version;
+  UInt16 length;
+  UInt32 cpuPLimit;
+  UInt32 gpuPLimit;
+  UInt32 memPLimit;
+
+} SMCKeyData_pLimitData_t;
+
+typedef struct
+{
+  UInt32 dataSize;
+  UInt32 dataType;
+
+  char   dataAttributes;
+
+} SMCKeyData_keyInfo_t;
+
+typedef char SMCBytes_t[32];
+
+typedef struct
+{
+  UInt32 key;
+
+  SMCKeyData_vers_t vers;
+  SMCKeyData_pLimitData_t pLimitData;
+  SMCKeyData_keyInfo_t keyInfo;
+
+  char   result;
+  char   status;
+  char   data8;
+
+  UInt32       data32;
+  SMCBytes_t   bytes;
+
+} SMCKeyData_t;
+
+typedef char UInt32Char_t[5];
+
+typedef struct
+{
+  UInt32Char_t key;
+  UInt32       dataSize;
+  UInt32Char_t dataType;
+  SMCBytes_t   bytes;
+
+} SMCVal_t;
+
+#endif // __APPLE__
+
+typedef int HM_ADAPTER_IOKIT;
+
+typedef void *IOKIT_LIB;
+
+typedef struct hm_iokit_lib
+{
+  #if defined(__APPLE__)
+  io_connect_t conn;
+  #endif // __APPLE__
+
+} hm_iokit_lib_t;
+
+typedef hm_iokit_lib_t IOKIT_PTR;
+
+#if defined(__APPLE__)
+UInt32 hm_IOKIT_strtoul (char *str, int size, int base);
+void hm_IOKIT_ultostr (char *str, UInt32 val);
+kern_return_t hm_IOKIT_SMCOpen (void *hashcat_ctx, io_connect_t *conn);
+kern_return_t hm_IOKIT_SMCClose (io_connect_t conn);
+kern_return_t hm_IOKIT_SMCCall (int index, SMCKeyData_t *inData, SMCKeyData_t *outData, io_connect_t conn);
+kern_return_t hm_IOKIT_SMCReadKey (UInt32Char_t key, SMCVal_t *val, io_connect_t conn);
+int hm_IOKIT_SMCGetSensorGraphicHot (void *hashcat_ctx);
+int hm_IOKIT_SMCGetTemperature (void *hashcat_ctx, char *key, double *temp);
+bool hm_IOKIT_SMCGetFanRPM (char *key, io_connect_t conn, float *ret);
+int hm_IOKIT_get_fan_speed_current (void *hashcat_ctx, char *fan_speed_buf);
+bool iokit_init (void *hashcat_ctx);
+bool iokit_close (void *hashcat_ctx);
+#endif // __APPLE__
+
+#endif // _EXT_IOKIT_H

include/ext_nvapi.h

@@ -239,4 +239,15 @@ typedef struct hm_nvapi_lib
 
 typedef hm_nvapi_lib_t NVAPI_PTR;
 
+int nvapi_init (void *hashcat_ctx);
+void nvapi_close (void *hashcat_ctx);
+void hm_NvAPI_GetErrorMessage (NVAPI_PTR *nvapi, const NvAPI_Status NvAPI_rc, NvAPI_ShortString string);
+int hm_NvAPI_Initialize (void *hashcat_ctx);
+int hm_NvAPI_Unload (void *hashcat_ctx);
+int hm_NvAPI_EnumPhysicalGPUs (void *hashcat_ctx, NvPhysicalGpuHandle nvGPUHandle[NVAPI_MAX_PHYSICAL_GPUS], NvU32 *pGpuCount);
+int hm_NvAPI_GPU_GetPerfPoliciesInfo (void *hashcat_ctx, NvPhysicalGpuHandle hPhysicalGpu, NV_GPU_PERF_POLICIES_INFO_PARAMS_V1 *perfPolicies_info);
+int hm_NvAPI_GPU_GetPerfPoliciesStatus (void *hashcat_ctx, NvPhysicalGpuHandle hPhysicalGpu, NV_GPU_PERF_POLICIES_STATUS_PARAMS_V1 *perfPolicies_status);
+int hm_NvAPI_GPU_GetBusId (void *hashcat_ctx, NvPhysicalGpuHandle hPhysicalGpu, NvU32 *pBusId);
+int hm_NvAPI_GPU_GetBusSlotId (void *hashcat_ctx, NvPhysicalGpuHandle hPhysicalGpu, NvU32 *pBusSlotId);
+
 #endif // _EXT_NVAPI_H

include/ext_nvml.h

@@ -217,4 +217,20 @@ typedef struct hm_nvml_lib
 
 typedef hm_nvml_lib_t NVML_PTR;
 
+int nvml_init (void *hashcat_ctx);
+void nvml_close (void *hashcat_ctx);
+const char *hm_NVML_nvmlErrorString (NVML_PTR *nvml, const nvmlReturn_t nvml_rc);
+
+int hm_NVML_nvmlInit (void *hashcat_ctx);
+int hm_NVML_nvmlShutdown (void *hashcat_ctx);
+int hm_NVML_nvmlDeviceGetCount (void *hashcat_ctx, unsigned int *deviceCount);
+int hm_NVML_nvmlDeviceGetHandleByIndex (void *hashcat_ctx, unsigned int device_index, nvmlDevice_t *device);
+int hm_NVML_nvmlDeviceGetTemperature (void *hashcat_ctx, nvmlDevice_t device, nvmlTemperatureSensors_t sensorType, unsigned int *temp);
+int hm_NVML_nvmlDeviceGetFanSpeed (void *hashcat_ctx, nvmlDevice_t device, unsigned int *speed);
+int hm_NVML_nvmlDeviceGetUtilizationRates (void *hashcat_ctx, nvmlDevice_t device, nvmlUtilization_t *utilization);
+int hm_NVML_nvmlDeviceGetClockInfo (void *hashcat_ctx, nvmlDevice_t device, nvmlClockType_t type, unsigned int *clockfreq);
+int hm_NVML_nvmlDeviceGetTemperatureThreshold (void *hashcat_ctx, nvmlDevice_t device, nvmlTemperatureThresholds_t thresholdType, unsigned int *temp);
+int hm_NVML_nvmlDeviceGetCurrPcieLinkWidth (void *hashcat_ctx, nvmlDevice_t device, unsigned int *currLinkWidth);
+int hm_NVML_nvmlDeviceGetPciInfo (void *hashcat_ctx, nvmlDevice_t device, nvmlPciInfo_t *pci);
+
 #endif // _NVML_H

include/ext_sysfs.h

@@ -6,6 +6,8 @@
 #ifndef _EXT_SYSFS_H
 #define _EXT_SYSFS_H
 
+#include <stdbool.h>
+
 static const char SYS_BUS_PCI_DEVICES[] = "/sys/bus/pci/devices";
 
 typedef int HM_ADAPTER_SYSFS;
@@ -22,4 +24,15 @@ typedef struct hm_sysfs_lib
 
 typedef hm_sysfs_lib_t SYSFS_PTR;
 
+bool sysfs_init (void *hashcat_ctx);
+void sysfs_close (void *hashcat_ctx);
+char *hm_SYSFS_get_syspath_device (void *hashcat_ctx, const int backend_device_idx);
+char *hm_SYSFS_get_syspath_hwmon (void *hashcat_ctx, const int backend_device_idx);
+int hm_SYSFS_get_fan_speed_current (void *hashcat_ctx, const int backend_device_idx, int *val);
+int hm_SYSFS_get_temperature_current (void *hashcat_ctx, const int backend_device_idx, int *val);
+int hm_SYSFS_get_pp_dpm_sclk (void *hashcat_ctx, const int backend_device_idx, int *val);
+int hm_SYSFS_get_pp_dpm_mclk (void *hashcat_ctx, const int backend_device_idx, int *val);
+int hm_SYSFS_get_pp_dpm_pcie (void *hashcat_ctx, const int backend_device_idx, int *val);
+int hm_SYSFS_get_gpu_busy_percent (void *hashcat_ctx, const int backend_device_idx, int *val);
+
 #endif // _EXT_SYSFS_H

include/hwmon.h

@@ -16,6 +16,9 @@ int hm_get_threshold_shutdown_with_devices_idx (hashcat_ctx_t *hashcat_ctx, cons
 int hm_get_temperature_with_devices_idx        (hashcat_ctx_t *hashcat_ctx, const int backend_device_idx);
 int hm_get_fanpolicy_with_devices_idx          (hashcat_ctx_t *hashcat_ctx, const int backend_device_idx);
 int hm_get_fanspeed_with_devices_idx           (hashcat_ctx_t *hashcat_ctx, const int backend_device_idx);
+#if defined(__APPLE__)
+int hm_get_fanspeed_apple                      (hashcat_ctx_t *hashcat_ctx, char *fan_speed_buf);
+#endif
 int hm_get_buslanes_with_devices_idx           (hashcat_ctx_t *hashcat_ctx, const int backend_device_idx);
 int hm_get_utilization_with_devices_idx        (hashcat_ctx_t *hashcat_ctx, const int backend_device_idx);
 int hm_get_memoryspeed_with_devices_idx        (hashcat_ctx_t *hashcat_ctx, const int backend_device_idx);

include/status.h

@@ -77,9 +77,9 @@ char       *status_get_speed_sec_dev                  (const hashcat_ctx_t *hash
 int         status_get_cpt_cur_min                    (const hashcat_ctx_t *hashcat_ctx);
 int         status_get_cpt_cur_hour                   (const hashcat_ctx_t *hashcat_ctx);
 int         status_get_cpt_cur_day                    (const hashcat_ctx_t *hashcat_ctx);
-int         status_get_cpt_avg_min                    (const hashcat_ctx_t *hashcat_ctx);
-int         status_get_cpt_avg_hour                   (const hashcat_ctx_t *hashcat_ctx);
-int         status_get_cpt_avg_day                    (const hashcat_ctx_t *hashcat_ctx);
+double      status_get_cpt_avg_min                    (const hashcat_ctx_t *hashcat_ctx);
+double      status_get_cpt_avg_hour                   (const hashcat_ctx_t *hashcat_ctx);
+double      status_get_cpt_avg_day                    (const hashcat_ctx_t *hashcat_ctx);
 char       *status_get_cpt                            (const hashcat_ctx_t *hashcat_ctx);
 int         status_get_salt_pos_dev                   (const hashcat_ctx_t *hashcat_ctx, const int backend_devices_idx);
 int         status_get_innerloop_pos_dev              (const hashcat_ctx_t *hashcat_ctx, const int backend_devices_idx);
@@ -98,6 +98,9 @@ char       *status_get_brain_link_send_bytes_sec_dev  (const hashcat_ctx_t *hash
 char       *status_get_brain_rx_all                   (const hashcat_ctx_t *hashcat_ctx);
 char       *status_get_brain_tx_all                   (const hashcat_ctx_t *hashcat_ctx);
 #endif
+#if defined(__APPLE__)
+char       *status_get_hwmon_fan_dev                  (const hashcat_ctx_t *hashcat_ctx);
+#endif
 char       *status_get_hwmon_dev                      (const hashcat_ctx_t *hashcat_ctx, const int backend_devices_idx);
 int         status_get_corespeed_dev                  (const hashcat_ctx_t *hashcat_ctx, const int backend_devices_idx);
 int         status_get_memoryspeed_dev                (const hashcat_ctx_t *hashcat_ctx, const int backend_devices_idx);

include/types.h

@@ -611,7 +611,11 @@ typedef enum user_options_defaults
   DEBUG_MODE               = 0,
   FORCE                    = false,
   HWMON_DISABLE            = false,
+  #if defined (__APPLE__)
+  HWMON_TEMP_ABORT         = 100,
+  #else
   HWMON_TEMP_ABORT         = 90,
+  #endif
   HASH_INFO                = false,
   HASH_MODE                = 0,
   HCCAPX_MESSAGE_PAIR      = 0,
@@ -619,6 +623,7 @@ typedef enum user_options_defaults
   HEX_SALT                 = false,
   HEX_WORDLIST             = false,
   HOOK_THREADS             = 0,
+  IDENTIFY                 = false,
   INCREMENT                = false,
   INCREMENT_MAX            = PW_MAX,
   INCREMENT_MIN            = 1,
@@ -725,73 +730,74 @@ typedef enum user_options_map
   IDX_HEX_SALT                  = 0xff1a,
   IDX_HEX_WORDLIST              = 0xff1b,
   IDX_HOOK_THREADS              = 0xff1c,
+  IDX_IDENTIFY                  = 0xff1d,
   IDX_INCREMENT                 = 'i',
-  IDX_INCREMENT_MAX             = 0xff1d,
-  IDX_INCREMENT_MIN             = 0xff1e,
-  IDX_INDUCTION_DIR             = 0xff1f,
-  IDX_KEEP_GUESSING             = 0xff20,
+  IDX_INCREMENT_MAX             = 0xff1e,
+  IDX_INCREMENT_MIN             = 0xff1f,
+  IDX_INDUCTION_DIR             = 0xff20,
+  IDX_KEEP_GUESSING             = 0xff21,
   IDX_KERNEL_ACCEL              = 'n',
   IDX_KERNEL_LOOPS              = 'u',
   IDX_KERNEL_THREADS            = 'T',
-  IDX_KEYBOARD_LAYOUT_MAPPING   = 0xff21,
-  IDX_KEYSPACE                  = 0xff22,
-  IDX_LEFT                      = 0xff23,
+  IDX_KEYBOARD_LAYOUT_MAPPING   = 0xff22,
+  IDX_KEYSPACE                  = 0xff23,
+  IDX_LEFT                      = 0xff24,
   IDX_LIMIT                     = 'l',
-  IDX_LOGFILE_DISABLE           = 0xff24,
-  IDX_LOOPBACK                  = 0xff25,
-  IDX_MACHINE_READABLE          = 0xff26,
-  IDX_MARKOV_CLASSIC            = 0xff27,
-  IDX_MARKOV_DISABLE            = 0xff28,
-  IDX_MARKOV_HCSTAT2            = 0xff29,
+  IDX_LOGFILE_DISABLE           = 0xff25,
+  IDX_LOOPBACK                  = 0xff26,
+  IDX_MACHINE_READABLE          = 0xff27,
+  IDX_MARKOV_CLASSIC            = 0xff28,
+  IDX_MARKOV_DISABLE            = 0xff29,
+  IDX_MARKOV_HCSTAT2            = 0xff2a,
   IDX_MARKOV_THRESHOLD          = 't',
-  IDX_NONCE_ERROR_CORRECTIONS   = 0xff2a,
+  IDX_NONCE_ERROR_CORRECTIONS   = 0xff2b,
   IDX_OPENCL_DEVICE_TYPES       = 'D',
   IDX_OPTIMIZED_KERNEL_ENABLE   = 'O',
-  IDX_OUTFILE_AUTOHEX_DISABLE   = 0xff2b,
-  IDX_OUTFILE_CHECK_DIR         = 0xff2c,
-  IDX_OUTFILE_CHECK_TIMER       = 0xff2d,
-  IDX_OUTFILE_FORMAT            = 0xff2e,
+  IDX_OUTFILE_AUTOHEX_DISABLE   = 0xff2c,
+  IDX_OUTFILE_CHECK_DIR         = 0xff2d,
+  IDX_OUTFILE_CHECK_TIMER       = 0xff2e,
+  IDX_OUTFILE_FORMAT            = 0xff2f,
   IDX_OUTFILE                   = 'o',
-  IDX_POTFILE_DISABLE           = 0xff2f,
-  IDX_POTFILE_PATH              = 0xff30,
-  IDX_PROGRESS_ONLY             = 0xff31,
-  IDX_QUIET                     = 0xff32,
-  IDX_REMOVE                    = 0xff33,
-  IDX_REMOVE_TIMER              = 0xff34,
-  IDX_RESTORE                   = 0xff35,
-  IDX_RESTORE_DISABLE           = 0xff36,
-  IDX_RESTORE_FILE_PATH         = 0xff37,
+  IDX_POTFILE_DISABLE           = 0xff30,
+  IDX_POTFILE_PATH              = 0xff31,
+  IDX_PROGRESS_ONLY             = 0xff32,
+  IDX_QUIET                     = 0xff33,
+  IDX_REMOVE                    = 0xff34,
+  IDX_REMOVE_TIMER              = 0xff35,
+  IDX_RESTORE                   = 0xff36,
+  IDX_RESTORE_DISABLE           = 0xff37,
+  IDX_RESTORE_FILE_PATH         = 0xff38,
   IDX_RP_FILE                   = 'r',
-  IDX_RP_GEN_FUNC_MAX           = 0xff38,
-  IDX_RP_GEN_FUNC_MIN           = 0xff39,
+  IDX_RP_GEN_FUNC_MAX           = 0xff39,
+  IDX_RP_GEN_FUNC_MIN           = 0xff3a,
   IDX_RP_GEN                    = 'g',
-  IDX_RP_GEN_SEED               = 0xff3a,
+  IDX_RP_GEN_SEED               = 0xff3b,
   IDX_RULE_BUF_L                = 'j',
   IDX_RULE_BUF_R                = 'k',
-  IDX_RUNTIME                   = 0xff3b,
-  IDX_SCRYPT_TMTO               = 0xff3c,
+  IDX_RUNTIME                   = 0xff3c,
+  IDX_SCRYPT_TMTO               = 0xff3d,
   IDX_SEGMENT_SIZE              = 'c',
-  IDX_SELF_TEST_DISABLE         = 0xff3d,
+  IDX_SELF_TEST_DISABLE         = 0xff3e,
   IDX_SEPARATOR                 = 'p',
-  IDX_SESSION                   = 0xff3e,
-  IDX_SHOW                      = 0xff3f,
+  IDX_SESSION                   = 0xff3f,
+  IDX_SHOW                      = 0xff40,
   IDX_SKIP                      = 's',
   IDX_SLOW_CANDIDATES           = 'S',
-  IDX_SPEED_ONLY                = 0xff40,
-  IDX_SPIN_DAMP                 = 0xff41,
-  IDX_STATUS                    = 0xff42,
-  IDX_STATUS_JSON               = 0xff43,
-  IDX_STATUS_TIMER              = 0xff44,
-  IDX_STDOUT_FLAG               = 0xff45,
-  IDX_STDIN_TIMEOUT_ABORT       = 0xff46,
-  IDX_TRUECRYPT_KEYFILES        = 0xff47,
-  IDX_USERNAME                  = 0xff48,
-  IDX_VERACRYPT_KEYFILES        = 0xff49,
-  IDX_VERACRYPT_PIM_START       = 0xff4a,
-  IDX_VERACRYPT_PIM_STOP        = 0xff4b,
+  IDX_SPEED_ONLY                = 0xff41,
+  IDX_SPIN_DAMP                 = 0xff42,
+  IDX_STATUS                    = 0xff43,
+  IDX_STATUS_JSON               = 0xff44,
+  IDX_STATUS_TIMER              = 0xff45,
+  IDX_STDOUT_FLAG               = 0xff46,
+  IDX_STDIN_TIMEOUT_ABORT       = 0xff47,
+  IDX_TRUECRYPT_KEYFILES        = 0xff48,
+  IDX_USERNAME                  = 0xff49,
+  IDX_VERACRYPT_KEYFILES        = 0xff4a,
+  IDX_VERACRYPT_PIM_START       = 0xff4b,
+  IDX_VERACRYPT_PIM_STOP        = 0xff4c,
   IDX_VERSION_LOWER             = 'v',
   IDX_VERSION                   = 'V',
-  IDX_WORDLIST_AUTOHEX_DISABLE  = 0xff4c,
+  IDX_WORDLIST_AUTOHEX_DISABLE  = 0xff4d,
   IDX_WORKLOAD_PROFILE          = 'w',
 
 } user_options_map_t;
@@ -1587,6 +1593,7 @@ typedef struct backend_ctx
   bool                need_nvml;
   bool                need_nvapi;
   bool                need_sysfs;
+  bool                need_iokit;
 
   int                 comptime;
 
@@ -1630,6 +1637,7 @@ typedef enum kernel_workload
 #include "ext_nvapi.h"
 #include "ext_nvml.h"
 #include "ext_sysfs.h"
+#include "ext_iokit.h"
 
 typedef struct hm_attrs
 {
@@ -1637,6 +1645,7 @@ typedef struct hm_attrs
   HM_ADAPTER_NVML    nvml;
   HM_ADAPTER_NVAPI   nvapi;
   HM_ADAPTER_SYSFS   sysfs;
+  HM_ADAPTER_IOKIT   iokit;
 
   int od_version;
 
@@ -1661,6 +1670,7 @@ typedef struct hwmon_ctx
   void *hm_nvml;
   void *hm_nvapi;
   void *hm_sysfs;
+  void *hm_iokit;
 
   hm_attrs_t *hm_device;
 
@@ -1951,6 +1961,7 @@ typedef struct user_options
   #endif
   bool         hash_mode_chgd;
   bool         hccapx_message_pair_chgd;
+  bool         identify;
   bool         increment_max_chgd;
   bool         increment_min_chgd;
   bool         kernel_accel_chgd;
@@ -2240,6 +2251,9 @@ typedef struct device_info
   double  exec_msec_dev;
   char   *speed_sec_dev;
   char   *guess_candidates_dev;
+  #if defined(__APPLE__)
+  char   *hwmon_fan_dev;
+  #endif
   char   *hwmon_dev;
   int     corespeed_dev;
   int     memoryspeed_dev;
@@ -2682,7 +2696,7 @@ typedef enum hash_category
   HASH_CATEGORY_PRIVATE_KEY             = 20,
   HASH_CATEGORY_IMS                     = 21,
   HASH_CATEGORY_CRYPTOCURRENCY_WALLET   = 22,
-
+  HASH_CATEGORY_FBE                     = 23
 } hash_category_t;
 
 // hash specific

src/Makefile

@@ -6,7 +6,7 @@
 SHARED                  ?= 0
 DEBUG                   := 0
 PRODUCTION              := 0
-PRODUCTION_VERSION      := v6.2.1
+PRODUCTION_VERSION      := v6.2.2
 ENABLE_CUBIN            ?= 1
 ENABLE_BRAIN            ?= 1
 ENABLE_UNRAR            ?= 1
@@ -189,6 +189,7 @@ endif
 ## because LZMA SDK
 ifeq ($(CC),clang)
 CFLAGS                  += -Wno-enum-conversion
+CFLAGS                  += -Wno-typedef-redefinition
 endif
 
 ## because ZLIB
@@ -314,6 +315,7 @@ endif # FreeBSD
 ifeq ($(UNAME),Darwin)
 export MACOSX_DEPLOYMENT_TARGET=10.9
 CFLAGS_NATIVE           := $(CFLAGS)
+CFLAGS_NATIVE           += -DWITH_HWMON
 
 ifeq ($(shell test $(DARWIN_VERSION) -le 15; echo $$?), 0)
 CFLAGS_NATIVE           += -DMISSING_CLOCK_GETTIME
@@ -321,6 +323,7 @@ endif
 
 LFLAGS_NATIVE           := $(LFLAGS)
 LFLAGS_NATIVE           += -framework OpenCL
+LFLAGS_NATIVE           += -framework IOKit
 LFLAGS_NATIVE           += -lpthread
 LFLAGS_NATIVE           += -liconv
 endif # Darwin
@@ -359,11 +362,10 @@ CXXFLAGS                :=
 
 EMU_OBJS_ALL            := emu_general emu_inc_common emu_inc_platform emu_inc_scalar emu_inc_simd
 EMU_OBJS_ALL            += emu_inc_rp emu_inc_rp_optimized
-EMU_OBJS_ALL            += emu_inc_truecrypt_crc32 emu_inc_truecrypt_keyfile emu_inc_truecrypt_xts emu_inc_veracrypt_xts
 EMU_OBJS_ALL            += emu_inc_hash_md4 emu_inc_hash_md5 emu_inc_hash_ripemd160 emu_inc_hash_sha1 emu_inc_hash_sha256 emu_inc_hash_sha384 emu_inc_hash_sha512 emu_inc_hash_streebog256 emu_inc_hash_streebog512 emu_inc_ecc_secp256k1
 EMU_OBJS_ALL            += emu_inc_cipher_aes emu_inc_cipher_camellia emu_inc_cipher_des emu_inc_cipher_kuznyechik emu_inc_cipher_serpent emu_inc_cipher_twofish
 
-OBJS_ALL                := affinity autotune backend benchmark bitmap bitops combinator common convert cpt cpu_crc32 debugfile dictstat dispatch dynloader event ext_ADL ext_cuda ext_nvapi ext_nvml ext_nvrtc ext_OpenCL ext_sysfs ext_lzma filehandling folder hashcat hashes hlfmt hwmon induct interface keyboard_layout locking logfile loopback memory monitor mpsp outfile_check outfile pidfile potfile restore rp rp_cpu selftest slow_candidates shared status stdout straight terminal thread timer tuningdb usage user_options wordlist $(EMU_OBJS_ALL)
+OBJS_ALL                := affinity autotune backend benchmark bitmap bitops combinator common convert cpt cpu_crc32 debugfile dictstat dispatch dynloader event ext_ADL ext_cuda ext_nvapi ext_nvml ext_nvrtc ext_OpenCL ext_sysfs ext_iokit ext_lzma filehandling folder hashcat hashes hlfmt hwmon induct interface keyboard_layout locking logfile loopback memory monitor mpsp outfile_check outfile pidfile potfile restore rp rp_cpu selftest slow_candidates shared status stdout straight terminal thread timer tuningdb usage user_options wordlist $(EMU_OBJS_ALL)
 
 ifeq ($(ENABLE_BRAIN),1)
 OBJS_ALL                += brain

src/backend.c

@@ -131,6 +131,17 @@ static int backend_ctx_find_alias_devices (hashcat_ctx_t *hashcat_ctx)
       backend_ctx->opencl_devices_active--;
 
       backend_ctx->backend_devices_active--;
+
+      // show a warning for specifically listed devices if they are an alias
+
+      if (backend_ctx->backend_devices_filter != (u64) -1)
+      {
+        if (backend_ctx->backend_devices_filter & (1ULL << alias_device->device_id))
+        {
+          event_log_warning (hashcat_ctx, "The device #%d specifically listed was skipped because it is an alias of device #%d", alias_device->device_id + 1, backend_device->device_id + 1);
+          event_log_warning (hashcat_ctx, NULL);
+        }
+      }
     }
   }
 
@@ -3017,6 +3028,28 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
   }
   else
   {
+    // innerloop prediction to get a speed estimation is hard, because we don't know in advance how much
+    // time the different kernels take and if their weightnings are equally distributed.
+    // - for instance, a regular _loop kernel is likely to be the slowest, but _loop2 kernel can also be slow.
+    //   in fact, _loop2 can be even slower (see iTunes backup >= 10.0).
+    // - hooks can have a large influence depending on the OS.
+    //   spawning threads and memory allocations take a lot of time on windows (compared to linux).
+    // - the kernel execution can take shortcuts based on intermediate values
+    //   while these intermediate valus depend on input values.
+    // - if we meassure runtimes of different kernels to find out about their weightning
+    //   we need to call them with real input values otherwise we miss the shortcuts inside the kernel.
+    // - the problem is that these real input values could crack the hash which makes the chaos perfect.
+    //
+    // so the innerloop prediction is not perfectly accurate, because we:
+    //
+    // 1. completely ignore hooks and the time they take.
+    // 2. assume that the code in _loop and _loop2 is similar,
+    //    but we respect the different iteration counts in _loop and _loop2.
+    // 3. ignore _comp kernel runtimes (probably irrelevant).
+    //
+    // as soon as the first restore checkpoint is reached the prediction is accurate.
+    // also the closer it gets to that point.
+
     if (true)
     {
       if (device_param->is_cuda == true)
@@ -3160,7 +3193,10 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
              * speed
              */
 
-            const float iter_part = (float) (loop_pos + loop_left) / iter;
+            const u32 iter1r = hashes->salts_buf[salt_pos].salt_iter  * (salt_repeats + 1);
+            const u32 iter2r = hashes->salts_buf[salt_pos].salt_iter2 * (salt_repeats + 1);
+
+            const double iter_part = (double) ((iter * salt_repeat) + loop_pos + loop_left) / (double) (iter1r + iter2r);
 
             const u64 perf_sum_all = (u64) (pws_cnt * iter_part);
 
@@ -3176,7 +3212,7 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
             {
               if (speed_msec > 4000)
               {
-                device_param->outerloop_multi *= (double) iter / (double) (loop_pos + loop_left);
+                device_param->outerloop_multi *= 1 / iter_part;
 
                 device_param->speed_pos = 1;
 
@@ -3295,6 +3331,25 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
             //bug?
             //while (status_ctx->run_thread_level2 == false) break;
             if (status_ctx->run_thread_level2 == false) break;
+
+            /**
+             * speed
+             */
+
+            const u32 iter1r = hashes->salts_buf[salt_pos].salt_iter  * (salt_repeats + 1);
+            const u32 iter2r = hashes->salts_buf[salt_pos].salt_iter2 * (salt_repeats + 1);
+
+            const double iter_part = (double) (iter1r + (iter * salt_repeat) + loop_pos + loop_left) / (double) (iter1r + iter2r);
+
+            const u64 perf_sum_all = (u64) (pws_cnt * iter_part);
+
+            double speed_msec = hc_timer_get (device_param->timer_speed);
+
+            const u32 speed_pos = device_param->speed_pos;
+
+            device_param->speed_cnt[speed_pos] = perf_sum_all;
+
+            device_param->speed_msec[speed_pos] = speed_msec;
           }
         }
       }
@@ -5119,6 +5174,7 @@ int backend_ctx_init (hashcat_ctx_t *hashcat_ctx)
   if (user_options->show      == true) return 0;
   if (user_options->usage     == true) return 0;
   if (user_options->version   == true) return 0;
+  if (user_options->identify  == true) return 0;
 
   hc_device_param_t *devices_param = (hc_device_param_t *) hccalloc (DEVICES_MAX, sizeof (hc_device_param_t));
 
@@ -5608,6 +5664,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
   bool need_nvml    = false;
   bool need_nvapi   = false;
   bool need_sysfs   = false;
+  bool need_iokit   = false;
 
   int backend_devices_idx = 0;
 
@@ -5856,10 +5913,12 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
         device_param->skipped = true;
       }
 
+      #if !defined (__APPLE__)
       if ((backend_ctx->opencl_device_types_filter & CL_DEVICE_TYPE_GPU) == 0)
       {
         device_param->skipped = true;
       }
+      #endif
 
       if ((device_param->opencl_platform_vendor_id == VENDOR_ID_NV) && (device_param->opencl_device_vendor_id == VENDOR_ID_NV))
       {
@@ -6614,6 +6673,13 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
         // vendor specific
 
+        #if defined (__APPLE__)
+        if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
+        {
+          if (device_param->skipped == false) need_iokit = true;
+        }
+        #endif
+
         if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
         {
           if ((device_param->opencl_platform_vendor_id == VENDOR_ID_AMD) && (device_param->opencl_device_vendor_id == VENDOR_ID_AMD))
@@ -7059,7 +7125,12 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
 
       if (device_param->is_opencl == false) continue;
 
+      if (user_options->backend_info == false)
+      {
+        // do not ignore in case -I because user expects a value also for skipped devices
+
         if (device_param->skipped == true) continue;
+      }
 
       /**
        * create context for each device
@@ -7283,6 +7354,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
   backend_ctx->need_nvml    = need_nvml;
   backend_ctx->need_nvapi   = need_nvapi;
   backend_ctx->need_sysfs   = need_sysfs;
+  backend_ctx->need_iokit   = need_iokit;
 
   backend_ctx->comptime     = comptime;
 
@@ -8626,29 +8698,11 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
     #endif
 
     /**
-     * device_name_chksum
+     * device_name_chksum_amp_mp
      */
 
-    char *device_name_chksum        = (char *) hcmalloc (HCBUFSIZ_TINY);
     char *device_name_chksum_amp_mp = (char *) hcmalloc (HCBUFSIZ_TINY);
 
-    // The kernel source can depend on some JiT compiler macros which themself depend on the attack_modes.
-    // ATM this is relevant only for ATTACK_MODE_ASSOCIATION which slightly modifies ATTACK_MODE_STRAIGHT kernels.
-
-    const u32 extra_value = (user_options->attack_mode == ATTACK_MODE_ASSOCIATION) ? ATTACK_MODE_ASSOCIATION : ATTACK_MODE_NONE;
-
-    const size_t dnclen = snprintf (device_name_chksum, HCBUFSIZ_TINY, "%d-%d-%d-%u-%s-%s-%s-%d-%u-%u",
-      backend_ctx->comptime,
-      backend_ctx->cuda_driver_version,
-      device_param->is_opencl,
-      device_param->opencl_platform_vendor_id,
-      device_param->device_name,
-      device_param->opencl_device_version,
-      device_param->opencl_driver_version,
-      device_param->vector_width,
-      hashconfig->kern_type,
-      extra_value);
-
     const size_t dnclen_amp_mp = snprintf (device_name_chksum_amp_mp, HCBUFSIZ_TINY, "%d-%d-%d-%u-%s-%s-%s",
       backend_ctx->comptime,
       backend_ctx->cuda_driver_version,
@@ -8660,12 +8714,6 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
     md5_ctx_t md5_ctx;
 
-    md5_init   (&md5_ctx);
-    md5_update (&md5_ctx, (u32 *) device_name_chksum, dnclen);
-    md5_final  (&md5_ctx);
-
-    snprintf (device_name_chksum, HCBUFSIZ_TINY, "%08x", md5_ctx.h[0]);
-
     md5_init   (&md5_ctx);
     md5_update (&md5_ctx, (u32 *) device_name_chksum_amp_mp, dnclen_amp_mp);
     md5_final  (&md5_ctx);
@@ -8884,6 +8932,38 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       if (user_options->quiet == false) event_log_warning (hashcat_ctx, "* Device #%u: build_options_module '%s'", device_id + 1, build_options_module_buf);
       #endif
 
+      /**
+       * device_name_chksum
+       */
+
+      char *device_name_chksum = (char *) hcmalloc (HCBUFSIZ_TINY);
+
+      // The kernel source can depend on some JiT compiler macros which themself depend on the attack_modes.
+      // ATM this is relevant only for ATTACK_MODE_ASSOCIATION which slightly modifies ATTACK_MODE_STRAIGHT kernels.
+
+      const u32 extra_value = (user_options->attack_mode == ATTACK_MODE_ASSOCIATION) ? ATTACK_MODE_ASSOCIATION : ATTACK_MODE_NONE;
+
+      const size_t dnclen = snprintf (device_name_chksum, HCBUFSIZ_TINY, "%d-%d-%d-%u-%s-%s-%s-%d-%u-%u-%s",
+        backend_ctx->comptime,
+        backend_ctx->cuda_driver_version,
+        device_param->is_opencl,
+        device_param->opencl_platform_vendor_id,
+        device_param->device_name,
+        device_param->opencl_device_version,
+        device_param->opencl_driver_version,
+        device_param->vector_width,
+        hashconfig->kern_type,
+        extra_value,
+        build_options_module_buf);
+
+      md5_ctx_t md5_ctx;
+
+      md5_init   (&md5_ctx);
+      md5_update (&md5_ctx, (u32 *) device_name_chksum, dnclen);
+      md5_final  (&md5_ctx);
+
+      snprintf (device_name_chksum, HCBUFSIZ_TINY, "%08x", md5_ctx.h[0]);
+
       /**
        * kernel source filename
        */
@@ -8921,6 +9001,8 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       }
 
       hcfree (build_options_module_buf);
+
+      hcfree (device_name_chksum);
     }
 
     /**
@@ -9032,7 +9114,6 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
       if (hc_clUnloadPlatformCompiler (hashcat_ctx, platform_id) == -1) return -1;
     }
 
-    hcfree (device_name_chksum);
     hcfree (device_name_chksum_amp_mp);
 
     // some algorithm collide too fast, make that impossible

src/bitmap.c

@@ -86,6 +86,7 @@ int bitmap_ctx_init (hashcat_ctx_t *hashcat_ctx)
   if (user_options->show          == true) return 0;
   if (user_options->usage         == true) return 0;
   if (user_options->version       == true) return 0;
+  if (user_options->identify      == true) return 0;
 
   bitmap_ctx->enabled = true;
 

src/cpt.c

@@ -22,6 +22,7 @@ int cpt_ctx_init (hashcat_ctx_t *hashcat_ctx)
   if (user_options->show          == true) return 0;
   if (user_options->usage         == true) return 0;
   if (user_options->version       == true) return 0;
+  if (user_options->identify      == true) return 0;
 
   cpt_ctx->enabled = true;
 

src/cpu_crc32.c

@@ -88,7 +88,7 @@ u32 cpu_crc32_buffer (const u8 *buf, const size_t length)
   return crc ^ 0xffffffff;
 }
 
-int cpu_crc32 (const char *filename, u8 keytab[64])
+int cpu_crc32 (const char *filename, u8 *keytab, const size_t keytabsz)
 {
   u32 crc = ~0U;
 
@@ -110,10 +110,10 @@ int cpu_crc32 (const char *filename, u8 keytab[64])
   {
     crc = crc32tab[(crc ^ buf[fpos]) & 0xff] ^ (crc >> 8);
 
-    keytab[kpos++] += (crc >> 24) & 0xff; if (kpos >= 64) kpos = 0;
-    keytab[kpos++] += (crc >> 16) & 0xff; if (kpos >= 64) kpos = 0;
-    keytab[kpos++] += (crc >>  8) & 0xff; if (kpos >= 64) kpos = 0;
-    keytab[kpos++] += (crc >>  0) & 0xff; if (kpos >= 64) kpos = 0;
+    keytab[kpos++] += (crc >> 24) & 0xff; if (kpos >= keytabsz) kpos = 0;
+    keytab[kpos++] += (crc >> 16) & 0xff; if (kpos >= keytabsz) kpos = 0;
+    keytab[kpos++] += (crc >>  8) & 0xff; if (kpos >= keytabsz) kpos = 0;
+    keytab[kpos++] += (crc >>  0) & 0xff; if (kpos >= keytabsz) kpos = 0;
   }
 
   hcfree (buf);

src/debugfile.c

@@ -97,6 +97,7 @@ int debugfile_init (hashcat_ctx_t *hashcat_ctx)
   if (user_options->progress_only == true) return 0;
   if (user_options->usage         == true) return 0;
   if (user_options->version       == true) return 0;
+  if (user_options->identify      == true) return 0;
   if (user_options->debug_mode    == 0)    return 0;
 
   debugfile_ctx->enabled = true;

src/dictstat.c

@@ -64,6 +64,7 @@ int dictstat_init (hashcat_ctx_t *hashcat_ctx)
   if (user_options->show          == true) return 0;
   if (user_options->usage         == true) return 0;
   if (user_options->version       == true) return 0;
+  if (user_options->identify      == true) return 0;
 
   if (user_options->attack_mode == ATTACK_MODE_BF) return 0;
 

src/emu_inc_truecrypt_crc32.c

@@ -1,11 +0,0 @@
-/**
- * Author......: See docs/credits.txt
- * License.....: MIT
- */
-
-#include "common.h"
-#include "types.h"
-#include "emu_general.h"
-
-#include "inc_truecrypt_crc32.cl"
-

src/emu_inc_truecrypt_keyfile.c

@@ -1,11 +0,0 @@
-/**
- * Author......: See docs/credits.txt
- * License.....: MIT
- */
-
-#include "common.h"
-#include "types.h"
-#include "emu_general.h"
-
-#include "inc_truecrypt_keyfile.cl"
-

src/emu_inc_truecrypt_xts.c

@@ -1,11 +0,0 @@
-/**
- * Author......: See docs/credits.txt
- * License.....: MIT
- */
-
-#include "common.h"
-#include "types.h"
-#include "emu_general.h"
-
-#include "inc_truecrypt_xts.cl"
-

src/emu_inc_veracrypt_xts.c

@@ -1,11 +0,0 @@
-/**
- * Author......: See docs/credits.txt
- * License.....: MIT
- */
-
-#include "common.h"
-#include "types.h"
-#include "emu_general.h"
-
-#include "inc_veracrypt_xts.cl"
-

src/ext_ADL.c

@@ -5,9 +5,285 @@
 
 #include "common.h"
 #include "types.h"
+#include "memory.h"
+#include "event.h"
 #include "ext_ADL.h"
 
+#include "dynloader.h"
+
 void *HC_API_CALL ADL_Main_Memory_Alloc (const int iSize)
 {
   return malloc ((size_t) iSize);
 }
+
+int adl_init (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  memset (adl, 0, sizeof (ADL_PTR));
+
+  #if defined (_WIN)
+  adl->lib = hc_dlopen ("atiadlxx.dll");
+
+  if (!adl->lib)
+  {
+    adl->lib = hc_dlopen ("atiadlxy.dll");
+  }
+  #elif defined (__CYGWIN__)
+  adl->lib = hc_dlopen ("atiadlxx.dll");
+
+  if (!adl->lib)
+  {
+    adl->lib = hc_dlopen ("atiadlxy.dll");
+  }
+  #elif defined (_POSIX)
+  adl->lib = hc_dlopen ("libatiadlxx.so");
+  #endif
+
+  if (!adl->lib)
+  {
+    //if (user_options->quiet == false)
+    //  event_log_error (hashcat_ctx, "Load of ADL library failed. Proceeding without ADL HWMon enabled.");
+
+    return -1;
+  }
+
+  HC_LOAD_FUNC(adl, ADL_Main_Control_Destroy, ADL_MAIN_CONTROL_DESTROY, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Main_Control_Create, ADL_MAIN_CONTROL_CREATE, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Adapter_NumberOfAdapters_Get, ADL_ADAPTER_NUMBEROFADAPTERS_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Adapter_AdapterInfo_Get, ADL_ADAPTER_ADAPTERINFO_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Display_DisplayInfo_Get, ADL_DISPLAY_DISPLAYINFO_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Adapter_ID_Get, ADL_ADAPTER_ID_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Adapter_VideoBiosInfo_Get, ADL_ADAPTER_VIDEOBIOSINFO_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_ThermalDevices_Enum, ADL_OVERDRIVE5_THERMALDEVICES_ENUM, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_Temperature_Get, ADL_OVERDRIVE5_TEMPERATURE_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_Temperature_Get, ADL_OVERDRIVE6_TEMPERATURE_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_CurrentActivity_Get, ADL_OVERDRIVE5_CURRENTACTIVITY_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_FanSpeedInfo_Get, ADL_OVERDRIVE5_FANSPEEDINFO_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive5_FanSpeed_Get, ADL_OVERDRIVE5_FANSPEED_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_FanSpeed_Get, ADL_OVERDRIVE6_FANSPEED_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Adapter_Active_Get, ADL_ADAPTER_ACTIVE_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive_Caps, ADL_OVERDRIVE_CAPS, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_Capabilities_Get, ADL_OVERDRIVE6_CAPABILITIES_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_StateInfo_Get, ADL_OVERDRIVE6_STATEINFO_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_CurrentStatus_Get, ADL_OVERDRIVE6_CURRENTSTATUS_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_TargetTemperatureData_Get, ADL_OVERDRIVE6_TARGETTEMPERATUREDATA_GET, ADL, 0);
+  HC_LOAD_FUNC(adl, ADL_Overdrive6_TargetTemperatureRangeInfo_Get, ADL_OVERDRIVE6_TARGETTEMPERATURERANGEINFO_GET, ADL, 0);
+
+  return 0;
+}
+
+void adl_close (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  if (adl)
+  {
+    if (adl->lib)
+      hc_dlclose (adl->lib);
+
+    hcfree (adl);
+  }
+}
+
+int hm_ADL_Main_Control_Destroy (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Main_Control_Destroy ();
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Main_Control_Destroy(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Main_Control_Create (void *hashcat_ctx, ADL_MAIN_MALLOC_CALLBACK callback, int iEnumConnectedAdapters)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Main_Control_Create (callback, iEnumConnectedAdapters);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Main_Control_Create(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Adapter_NumberOfAdapters_Get (void *hashcat_ctx, int *lpNumAdapters)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Adapter_NumberOfAdapters_Get (lpNumAdapters);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Adapter_NumberOfAdapters_Get(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Adapter_AdapterInfo_Get (void *hashcat_ctx, LPAdapterInfo lpInfo, int iInputSize)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Adapter_AdapterInfo_Get (lpInfo, iInputSize);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Adapter_AdapterInfo_Get(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Overdrive5_Temperature_Get (void *hashcat_ctx, int iAdapterIndex, int iThermalControllerIndex, ADLTemperature *lpTemperature)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Overdrive5_Temperature_Get (iAdapterIndex, iThermalControllerIndex, lpTemperature);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Overdrive5_Temperature_Get(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Overdrive6_Temperature_Get (void *hashcat_ctx, int iAdapterIndex, int *iTemperature)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Overdrive6_Temperature_Get (iAdapterIndex, iTemperature);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Overdrive6_Temperature_Get(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Overdrive_CurrentActivity_Get (void *hashcat_ctx, int iAdapterIndex, ADLPMActivity *lpActivity)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Overdrive5_CurrentActivity_Get (iAdapterIndex, lpActivity);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Overdrive5_CurrentActivity_Get(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Overdrive5_FanSpeed_Get (void *hashcat_ctx, int iAdapterIndex, int iThermalControllerIndex, ADLFanSpeedValue *lpFanSpeedValue)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Overdrive5_FanSpeed_Get (iAdapterIndex, iThermalControllerIndex, lpFanSpeedValue);
+
+  if ((ADL_rc != ADL_OK) && (ADL_rc != ADL_ERR_NOT_SUPPORTED)) // exception allowed only here
+  {
+    event_log_error (hashcat_ctx, "ADL_Overdrive5_FanSpeed_Get(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Overdrive6_FanSpeed_Get (void *hashcat_ctx, int iAdapterIndex, ADLOD6FanSpeedInfo *lpFanSpeedInfo)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Overdrive6_FanSpeed_Get (iAdapterIndex, lpFanSpeedInfo);
+
+  if ((ADL_rc != ADL_OK) && (ADL_rc != ADL_ERR_NOT_SUPPORTED)) // exception allowed only here
+  {
+    event_log_error (hashcat_ctx, "ADL_Overdrive6_FanSpeed_Get(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Overdrive_Caps (void *hashcat_ctx, int iAdapterIndex, int *od_supported, int *od_enabled, int *od_version)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Overdrive_Caps (iAdapterIndex, od_supported, od_enabled, od_version);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Overdrive_Caps(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_ADL_Overdrive6_TargetTemperatureData_Get (void *hashcat_ctx, int iAdapterIndex, int *cur_temp, int *default_temp)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  ADL_PTR *adl = (ADL_PTR *) hwmon_ctx->hm_adl;
+
+  const int ADL_rc = adl->ADL_Overdrive6_TargetTemperatureData_Get (iAdapterIndex, cur_temp, default_temp);
+
+  if (ADL_rc != ADL_OK)
+  {
+    event_log_error (hashcat_ctx, "ADL_Overdrive6_TargetTemperatureData_Get(): %d", ADL_rc);
+
+    return -1;
+  }
+
+  return 0;
+}

src/ext_iokit.c

@@ -0,0 +1,305 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#include "common.h"
+#include "types.h"
+#include "memory.h"
+#include "shared.h"
+#include "event.h"
+#include "ext_iokit.h"
+
+#if defined (__APPLE__)
+#include <IOKit/IOKitLib.h>
+
+UInt32 hm_IOKIT_strtoul (char *str, int size, int base)
+{
+  int i;
+
+  UInt32 total = 0;
+
+  for (i = 0; i < size; i++)
+  {
+    if (base == 16)
+    {
+      total += str[i] << (size - 1 - i) * 8;
+    }
+    else
+    {
+      total += (unsigned char)(str[i] << (size - 1 - i) * 8);
+    }
+  }
+  return total;
+}
+
+void hm_IOKIT_ultostr (char *str, UInt32 val)
+{
+  str[0] = '\0';
+
+  sprintf (str, "%c%c%c%c", (unsigned int)(val >> 24), (unsigned int)(val >> 16), (unsigned int)(val >> 8), (unsigned int)(val));
+}
+
+kern_return_t hm_IOKIT_SMCOpen (void *hashcat_ctx, io_connect_t *conn)
+{
+  kern_return_t result;
+  io_iterator_t iterator;
+  io_object_t device;
+
+  CFMutableDictionaryRef matchingDictionary = IOServiceMatching ("AppleSMC");
+
+  result = IOServiceGetMatchingServices (kIOMasterPortDefault, matchingDictionary, &iterator);
+
+  if (result != kIOReturnSuccess)
+  {
+    event_log_error (hashcat_ctx, "IOServiceGetMatchingServices(): %08x", result);
+
+    return 1;
+  }
+
+  device = IOIteratorNext (iterator);
+
+  IOObjectRelease (iterator);
+
+  if (device == 0)
+  {
+    event_log_error (hashcat_ctx, "hm_IOKIT_SMCOpen(): no SMC found.");
+
+    return 1;
+  }
+
+  result = IOServiceOpen (device, mach_task_self(), 0, conn);
+
+  IOObjectRelease (device);
+
+  if (result != kIOReturnSuccess)
+  {
+    event_log_error (hashcat_ctx, "IOServiceOpen(): %08x", result);
+
+    return 1;
+  }
+
+  return kIOReturnSuccess;
+}
+
+kern_return_t hm_IOKIT_SMCClose (io_connect_t conn)
+{
+  return IOServiceClose (conn);
+}
+
+kern_return_t hm_IOKIT_SMCCall (int index, SMCKeyData_t *inData, SMCKeyData_t *outData, io_connect_t conn)
+{
+  size_t inDataSize  = sizeof (SMCKeyData_t);
+  size_t outDataSize = sizeof (SMCKeyData_t);
+
+  #if MAC_OS_X_VERSION_10_5
+  return IOConnectCallStructMethod (conn, index, inData, inDataSize, outData, &outDataSize);
+  #else
+  return IOConnectMethodStructureIStructureO (conn, index, inDataSize, &outDataSize, inData, outData);
+  #endif
+}
+
+kern_return_t hm_IOKIT_SMCReadKey (UInt32Char_t key, SMCVal_t *val, io_connect_t conn)
+{
+  SMCKeyData_t inData;
+  SMCKeyData_t outData;
+
+  memset (&inData,  0, sizeof (SMCKeyData_t));
+  memset (&outData, 0, sizeof (SMCKeyData_t));
+  memset (val,      0, sizeof (SMCVal_t));
+
+  inData.key = hm_IOKIT_strtoul (key, 4, 16);
+
+  inData.data8 = SMC_CMD_READ_KEYINFO;
+
+  if (hm_IOKIT_SMCCall (KERNEL_INDEX_SMC, &inData, &outData, conn) != kIOReturnSuccess) return 1;
+
+  val->dataSize = outData.keyInfo.dataSize;
+
+  hm_IOKIT_ultostr (val->dataType, outData.keyInfo.dataType);
+
+  inData.keyInfo.dataSize = val->dataSize;
+
+  inData.data8 = SMC_CMD_READ_BYTES;
+
+  if (hm_IOKIT_SMCCall (KERNEL_INDEX_SMC, &inData, &outData, conn) != kIOReturnSuccess) return 1;
+
+  memcpy(val->bytes, outData.bytes, sizeof(outData.bytes));
+
+  return kIOReturnSuccess;
+}
+
+int hm_IOKIT_SMCGetSensorGraphicHot (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  IOKIT_PTR *iokit = hwmon_ctx->hm_iokit;
+
+  SMCVal_t val;
+
+  memset (&val, 0, sizeof (SMCVal_t));
+
+  if (hm_IOKIT_SMCReadKey (HM_IOKIT_SMC_SENSOR_GRAPHICS_HOT, &val, iokit->conn) == kIOReturnSuccess)
+  {
+    int alarm = -1;
+
+    if (val.dataSize > 0)
+    {
+      if (strcmp(val.dataType, DATATYPE_UINT8) == 0)
+      {
+        alarm = hm_IOKIT_strtoul ((char *)val.bytes, val.dataSize, 10);
+      }
+    }
+
+    return alarm;
+  }
+
+  return -1;
+}
+
+int hm_IOKIT_SMCGetTemperature (void *hashcat_ctx, char *key, double *temp)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  IOKIT_PTR *iokit = hwmon_ctx->hm_iokit;
+
+  SMCVal_t val;
+
+  memset (&val, 0, sizeof (SMCVal_t));
+
+  if (hm_IOKIT_SMCReadKey (key, &val, iokit->conn) == kIOReturnSuccess)
+  {
+    if (val.dataSize > 0)
+    {
+      if (strcmp(val.dataType, DATATYPE_SP78) == 0)
+      {
+        // convert sp78 value to temperature
+        int intValue = val.bytes[0] * 256 + (unsigned char)val.bytes[1];
+
+        *temp = (intValue / 256.0);
+
+        return 1;
+      }
+    }
+  }
+
+  // read failed
+
+  *temp = 0.0;
+
+  return -1;
+}
+
+bool hm_IOKIT_SMCGetFanRPM (char *key, io_connect_t conn, float *ret)
+{
+  SMCVal_t val;
+
+  memset (&val, 0, sizeof (SMCVal_t));
+
+  if (hm_IOKIT_SMCReadKey (key, &val, conn) == kIOReturnSuccess)
+  {
+    if (val.dataSize > 0)
+    {
+      if (strcmp(val.dataType, DATATYPE_FLT) == 0)
+      {
+        *ret = *(float *) val.bytes;
+
+        return true;
+      }
+
+      if (strcmp(val.dataType, DATATYPE_FPE2) == 0)
+      {
+        // convert fpe2 value to RPM
+        *ret = ntohs (*(UInt16*)val.bytes) / 4.0;
+
+        return true;
+      }
+    }
+  }
+
+  // read failed
+  *ret = -1.f;
+
+  return false;
+}
+
+int hm_IOKIT_get_fan_speed_current (void *hashcat_ctx, char *fan_speed_buf)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  IOKIT_PTR *iokit = hwmon_ctx->hm_iokit;
+
+  SMCVal_t val;
+
+  UInt32Char_t key;
+
+  memset (&val, 0, sizeof (SMCVal_t));
+
+  if (hm_IOKIT_SMCReadKey ("FNum", &val, iokit->conn) == kIOReturnSuccess)
+  {
+    int totalFans = hm_IOKIT_strtoul ((char *)val.bytes, val.dataSize, 10);
+
+    if (totalFans <= 0) return -1;
+
+    char tmp_buf[16];
+
+    for (int i = 0; i < totalFans; i++)
+    {
+      int fan_speed = 0;
+      float actual_speed  = 0.0f;
+      float maximum_speed = 0.0f;
+
+      memset (&key, 0, sizeof (UInt32Char_t));
+      sprintf (key, "F%dAc", i);
+      hm_IOKIT_SMCGetFanRPM (key, iokit->conn, &actual_speed);
+      if (actual_speed < 0.f) continue;
+
+      memset (&key, 0, sizeof (UInt32Char_t));
+      sprintf (key, "F%dMx", i);
+      hm_IOKIT_SMCGetFanRPM (key, iokit->conn, &maximum_speed);
+      if (maximum_speed < 0.f) continue;
+
+      fan_speed = (actual_speed / maximum_speed) * 100.f;
+
+      memset   (tmp_buf, 0, sizeof (tmp_buf));
+      snprintf (tmp_buf, sizeof (tmp_buf) - 1, "Fan%d: %d%%, ", i, fan_speed);
+      strncat  (fan_speed_buf, tmp_buf, strlen (tmp_buf));
+    }
+
+    // remove last two bytes
+    size_t out_len = strlen (fan_speed_buf);
+    if (out_len > 2) fan_speed_buf[out_len-2] = '\0';
+  }
+
+  return 1;
+}
+
+bool iokit_init (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  IOKIT_PTR *iokit = hwmon_ctx->hm_iokit;
+
+  memset (iokit, 0, sizeof (IOKIT_PTR));
+
+  if (hm_IOKIT_SMCOpen (hashcat_ctx, &iokit->conn) == kIOReturnSuccess) return true;
+
+  hcfree (hwmon_ctx->hm_iokit);
+
+  hwmon_ctx->hm_iokit = NULL;
+
+  return false;
+}
+
+bool iokit_close (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  IOKIT_PTR *iokit = hwmon_ctx->hm_iokit;
+
+  hm_IOKIT_SMCClose (iokit->conn);
+
+  return true;
+}
+
+#endif // __APPLE__

src/ext_nvapi.c

@@ -5,4 +5,237 @@
 
 #include "common.h"
 #include "types.h"
+#include "memory.h"
+#include "event.h"
 #include "ext_nvapi.h"
+
+#include "dynloader.h"
+
+int nvapi_init (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  memset (nvapi, 0, sizeof (NVAPI_PTR));
+
+  #if defined (_WIN)
+
+  #if defined (_WIN64)
+  nvapi->lib = hc_dlopen ("nvapi64.dll");
+  #else
+  nvapi->lib = hc_dlopen ("nvapi.dll");
+  #endif
+
+  #else
+
+  #if defined (__CYGWIN__)
+
+  #if defined (__x86_x64__)
+  nvapi->lib = hc_dlopen ("nvapi64.dll");
+  #else
+  nvapi->lib = hc_dlopen ("nvapi.dll");
+  #endif
+
+  #else
+  nvapi->lib = hc_dlopen ("nvapi.so"); // uhm yes, but .. yeah
+  #endif
+
+  #endif
+
+  if (!nvapi->lib)
+  {
+    //if (user_options->quiet == false)
+    //  event_log_error (hashcat_ctx, "Load of NVAPI library failed. Proceeding without NVAPI HWMon enabled.");
+
+    return -1;
+  }
+
+  HC_LOAD_FUNC(nvapi, nvapi_QueryInterface,             NVAPI_QUERYINTERFACE,             NVAPI,                0);
+  HC_LOAD_ADDR(nvapi, NvAPI_Initialize,                 NVAPI_INITIALIZE,                 nvapi_QueryInterface, 0x0150E828U, NVAPI, 0);
+  HC_LOAD_ADDR(nvapi, NvAPI_Unload,                     NVAPI_UNLOAD,                     nvapi_QueryInterface, 0xD22BDD7EU, NVAPI, 0);
+  HC_LOAD_ADDR(nvapi, NvAPI_GetErrorMessage,            NVAPI_GETERRORMESSAGE,            nvapi_QueryInterface, 0x6C2D048CU, NVAPI, 0);
+  HC_LOAD_ADDR(nvapi, NvAPI_EnumPhysicalGPUs,           NVAPI_ENUMPHYSICALGPUS,           nvapi_QueryInterface, 0xE5AC921FU, NVAPI, 0);
+  HC_LOAD_ADDR(nvapi, NvAPI_GPU_GetPerfPoliciesInfo,    NVAPI_GPU_GETPERFPOLICIESINFO,    nvapi_QueryInterface, 0x409D9841U, NVAPI, 0);
+  HC_LOAD_ADDR(nvapi, NvAPI_GPU_GetPerfPoliciesStatus,  NVAPI_GPU_GETPERFPOLICIESSTATUS,  nvapi_QueryInterface, 0x3D358A0CU, NVAPI, 0);
+  HC_LOAD_ADDR(nvapi, NvAPI_GPU_GetBusId,               NVAPI_GPU_GETBUSID,               nvapi_QueryInterface, 0x1BE0B8E5U, NVAPI, 0);
+  HC_LOAD_ADDR(nvapi, NvAPI_GPU_GetBusSlotId,           NVAPI_GPU_GETBUSSLOTID,           nvapi_QueryInterface, 0x2A0A350FU, NVAPI, 0);
+
+  return 0;
+}
+
+void nvapi_close (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  if (nvapi)
+  {
+    if (nvapi->lib)
+      hc_dlclose (nvapi->lib);
+
+    hcfree (nvapi);
+  }
+}
+
+void hm_NvAPI_GetErrorMessage (NVAPI_PTR *nvapi, const NvAPI_Status NvAPI_rc, NvAPI_ShortString string)
+{
+  nvapi->NvAPI_GetErrorMessage (NvAPI_rc, string);
+}
+
+int hm_NvAPI_Initialize (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  const NvAPI_Status NvAPI_rc = (NvAPI_Status) nvapi->NvAPI_Initialize ();
+
+  if (NvAPI_rc == NVAPI_LIBRARY_NOT_FOUND) return -1;
+
+  if (NvAPI_rc != NVAPI_OK)
+  {
+    NvAPI_ShortString string = { 0 };
+
+    hm_NvAPI_GetErrorMessage (nvapi, NvAPI_rc, string);
+
+    event_log_error (hashcat_ctx, "NvAPI_Initialize(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NvAPI_Unload (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  const NvAPI_Status NvAPI_rc = (NvAPI_Status) nvapi->NvAPI_Unload ();
+
+  if (NvAPI_rc != NVAPI_OK)
+  {
+    NvAPI_ShortString string = { 0 };
+
+    hm_NvAPI_GetErrorMessage (nvapi, NvAPI_rc, string);
+
+    event_log_error (hashcat_ctx, "NvAPI_Unload(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NvAPI_EnumPhysicalGPUs (void *hashcat_ctx, NvPhysicalGpuHandle nvGPUHandle[NVAPI_MAX_PHYSICAL_GPUS], NvU32 *pGpuCount)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  const NvAPI_Status NvAPI_rc = (NvAPI_Status) nvapi->NvAPI_EnumPhysicalGPUs (nvGPUHandle, pGpuCount);
+
+  if (NvAPI_rc != NVAPI_OK)
+  {
+    NvAPI_ShortString string = { 0 };
+
+    hm_NvAPI_GetErrorMessage (nvapi, NvAPI_rc, string);
+
+    event_log_error (hashcat_ctx, "NvAPI_EnumPhysicalGPUs(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NvAPI_GPU_GetPerfPoliciesInfo (void *hashcat_ctx, NvPhysicalGpuHandle hPhysicalGpu, NV_GPU_PERF_POLICIES_INFO_PARAMS_V1 *perfPolicies_info)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  const NvAPI_Status NvAPI_rc = (NvAPI_Status) nvapi->NvAPI_GPU_GetPerfPoliciesInfo (hPhysicalGpu, perfPolicies_info);
+
+  if (NvAPI_rc != NVAPI_OK)
+  {
+    NvAPI_ShortString string = { 0 };
+
+    hm_NvAPI_GetErrorMessage (nvapi, NvAPI_rc, string);
+
+    event_log_error (hashcat_ctx, "NvAPI_GPU_GetPerfPoliciesInfo(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NvAPI_GPU_GetPerfPoliciesStatus (void *hashcat_ctx, NvPhysicalGpuHandle hPhysicalGpu, NV_GPU_PERF_POLICIES_STATUS_PARAMS_V1 *perfPolicies_status)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  const NvAPI_Status NvAPI_rc = (NvAPI_Status) nvapi->NvAPI_GPU_GetPerfPoliciesStatus (hPhysicalGpu, perfPolicies_status);
+
+  if (NvAPI_rc != NVAPI_OK)
+  {
+    NvAPI_ShortString string = { 0 };
+
+    hm_NvAPI_GetErrorMessage (nvapi, NvAPI_rc, string);
+
+    event_log_error (hashcat_ctx, "NvAPI_GPU_GetPerfPoliciesStatus(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NvAPI_GPU_GetBusId (void *hashcat_ctx, NvPhysicalGpuHandle hPhysicalGpu, NvU32 *pBusId)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  const NvAPI_Status NvAPI_rc = (NvAPI_Status) nvapi->NvAPI_GPU_GetBusId (hPhysicalGpu, pBusId);
+
+  if (NvAPI_rc != NVAPI_OK)
+  {
+    NvAPI_ShortString string = { 0 };
+
+    hm_NvAPI_GetErrorMessage (nvapi, NvAPI_rc, string);
+
+    event_log_error (hashcat_ctx, "NvAPI_GPU_GetBusId(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NvAPI_GPU_GetBusSlotId (void *hashcat_ctx, NvPhysicalGpuHandle hPhysicalGpu, NvU32 *pBusSlotId)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVAPI_PTR *nvapi = (NVAPI_PTR *) hwmon_ctx->hm_nvapi;
+
+  const NvAPI_Status NvAPI_rc = (NvAPI_Status) nvapi->NvAPI_GPU_GetBusSlotId (hPhysicalGpu, pBusSlotId);
+
+  if (NvAPI_rc != NVAPI_OK)
+  {
+    NvAPI_ShortString string = { 0 };
+
+    hm_NvAPI_GetErrorMessage (nvapi, NvAPI_rc, string);
+
+    event_log_error (hashcat_ctx, "NvAPI_GPU_GetBusSlotId(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}

src/ext_nvml.c

@@ -5,4 +5,390 @@
 
 #include "common.h"
 #include "types.h"
+#include "memory.h"
+#include "shared.h"
+#include "event.h"
 #include "ext_nvml.h"
+
+#include "dynloader.h"
+
+#if defined (__CYGWIN__)
+#include <sys/cygwin.h>
+#endif
+
+int nvml_init (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  memset (nvml, 0, sizeof (NVML_PTR));
+
+  #if defined (_WIN)
+
+  nvml->lib = hc_dlopen ("nvml.dll");
+
+  if (!nvml->lib)
+  {
+    DWORD BufferSize = 1024;
+
+    DWORD Type = REG_SZ;
+
+    char *Buffer = (char *) hcmalloc (BufferSize + 1);
+
+    HKEY hKey = 0;
+
+    if (RegOpenKeyExA (HKEY_LOCAL_MACHINE, "SOFTWARE\\NVIDIA Corporation\\Global\\NVSMI", 0, KEY_QUERY_VALUE, &hKey) == ERROR_SUCCESS)
+    {
+      if (RegQueryValueExA (hKey, "NVSMIPATH", NULL, &Type, (LPBYTE)Buffer, &BufferSize) == ERROR_SUCCESS)
+      {
+        Buffer[BufferSize] = 0;
+      }
+      else
+      {
+        //if (user_options->quiet == false)
+        //  event_log_error (hashcat_ctx, "NVML library load failed. Proceeding without NVML HWMon enabled.");
+
+        return -1;
+      }
+
+      RegCloseKey (hKey);
+    }
+    else
+    {
+      //if (user_options->quiet == false)
+      //  event_log_error (hashcat_ctx, "NVML library load failed. Proceeding without NVML HWMon enabled.");
+
+      return -1;
+    }
+
+    strcat (Buffer, "\\nvml.dll");
+
+    nvml->lib = hc_dlopen (Buffer);
+
+    hcfree (Buffer);
+  }
+
+  #elif defined (__CYGWIN__)
+
+  nvml->lib = hc_dlopen ("nvml.dll");
+
+  if (!nvml->lib)
+  {
+    HCFILE nvml_lib;
+
+    if (hc_fopen (&nvml_lib, "/proc/registry/HKEY_LOCAL_MACHINE/SOFTWARE/NVIDIA Corporation/Global/NVSMI/NVSMIPATH", "rb") == false)
+    {
+      //if (user_options->quiet == false)
+      //  event_log_error (hashcat_ctx, "NVML library load failed: %m. Proceeding without NVML HWMon enabled.");
+
+      return -1;
+    }
+
+    char *nvml_winpath, *nvml_cygpath;
+
+    nvml_winpath = (char *) hcmalloc (100);
+
+    hc_fread (nvml_winpath, 100, 1, &nvml_lib);
+
+    hc_fclose (&nvml_lib);
+
+    ssize_t size = cygwin_conv_path (CCP_WIN_A_TO_POSIX | CCP_PROC_CYGDRIVE, nvml_winpath, NULL, 0);
+
+    if (size > 0)
+    {
+      nvml_cygpath = (char *) hcmalloc (size + 9);
+
+      cygwin_conv_path (CCP_WIN_A_TO_POSIX | CCP_PROC_CYGDRIVE, nvml_winpath, nvml_cygpath, size);
+    }
+    else
+    {
+      //if (user_options->quiet == false)
+      //  event_log_error (hashcat_ctx, "Could not find NVML on this system. Proceeding without NVML HWMon enabled.");
+
+      return -1;
+    }
+
+    strcat (nvml_cygpath, "/nvml.dll");
+
+    nvml->lib = hc_dlopen (nvml_cygpath);
+  }
+
+  #elif defined (_POSIX)
+
+  nvml->lib = hc_dlopen ("libnvidia-ml.so");
+
+  if (!nvml->lib)
+  {
+    nvml->lib = hc_dlopen ("libnvidia-ml.so.1");
+  }
+
+  #endif
+
+  if (!nvml->lib)
+  {
+    //if (user_options->quiet == false)
+    //  event_log_error (hashcat_ctx, "NVML library load failed. Proceeding without NVML HWMon enabled.");
+
+    return -1;
+  }
+
+  HC_LOAD_FUNC(nvml, nvmlErrorString, NVML_ERROR_STRING, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlInit, NVML_INIT, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlShutdown, NVML_SHUTDOWN, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetCount, NVML_DEVICE_GET_COUNT, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetName, NVML_DEVICE_GET_NAME, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetHandleByIndex, NVML_DEVICE_GET_HANDLE_BY_INDEX, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetTemperature, NVML_DEVICE_GET_TEMPERATURE, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetFanSpeed, NVML_DEVICE_GET_FAN_SPEED, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetUtilizationRates, NVML_DEVICE_GET_UTILIZATION_RATES, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetClockInfo, NVML_DEVICE_GET_CLOCKINFO, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetTemperatureThreshold, NVML_DEVICE_GET_THRESHOLD, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetCurrPcieLinkGeneration, NVML_DEVICE_GET_CURRPCIELINKGENERATION, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetCurrPcieLinkWidth, NVML_DEVICE_GET_CURRPCIELINKWIDTH, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetCurrentClocksThrottleReasons, NVML_DEVICE_GET_CURRENTCLOCKSTHROTTLEREASONS, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetSupportedClocksThrottleReasons, NVML_DEVICE_GET_SUPPORTEDCLOCKSTHROTTLEREASONS, NVML, 0);
+  HC_LOAD_FUNC(nvml, nvmlDeviceGetPciInfo, NVML_DEVICE_GET_PCIINFO, NVML, 0);
+
+  return 0;
+}
+
+void nvml_close (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  if (nvml)
+  {
+    if (nvml->lib)
+      hc_dlclose (nvml->lib);
+
+    hcfree (nvml);
+  }
+}
+
+const char *hm_NVML_nvmlErrorString (NVML_PTR *nvml, const nvmlReturn_t nvml_rc)
+{
+  return nvml->nvmlErrorString (nvml_rc);
+}
+
+int hm_NVML_nvmlInit (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = (nvmlReturn_t) nvml->nvmlInit ();
+
+  if (nvml_rc != NVML_SUCCESS && nvml_rc != NVML_ERROR_DRIVER_NOT_LOADED)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlInit(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlShutdown (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = (nvmlReturn_t) nvml->nvmlShutdown ();
+
+  if (nvml_rc != NVML_SUCCESS && nvml_rc != NVML_ERROR_DRIVER_NOT_LOADED)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlShutdown(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetCount (void *hashcat_ctx, unsigned int *deviceCount)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetCount (deviceCount);
+
+  if (nvml_rc != NVML_SUCCESS && nvml_rc != NVML_ERROR_DRIVER_NOT_LOADED)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetCount(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetHandleByIndex (void *hashcat_ctx, unsigned int device_index, nvmlDevice_t *device)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetHandleByIndex (device_index, device);
+
+  if (nvml_rc != NVML_SUCCESS)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetHandleByIndex(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetTemperature (void *hashcat_ctx, nvmlDevice_t device, nvmlTemperatureSensors_t sensorType, unsigned int *temp)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetTemperature (device, sensorType, temp);
+
+  if (nvml_rc != NVML_SUCCESS)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetTemperature(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetFanSpeed (void *hashcat_ctx, nvmlDevice_t device, unsigned int *speed)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetFanSpeed (device, speed);
+
+  if (nvml_rc != NVML_SUCCESS)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetFanSpeed(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetUtilizationRates (void *hashcat_ctx, nvmlDevice_t device, nvmlUtilization_t *utilization)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetUtilizationRates (device, utilization);
+
+  if (nvml_rc != NVML_SUCCESS)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetUtilizationRates(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetClockInfo (void *hashcat_ctx, nvmlDevice_t device, nvmlClockType_t type, unsigned int *clockfreq)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetClockInfo (device, type, clockfreq);
+
+  if (nvml_rc != NVML_SUCCESS)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetClockInfo(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetTemperatureThreshold (void *hashcat_ctx, nvmlDevice_t device, nvmlTemperatureThresholds_t thresholdType, unsigned int *temp)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetTemperatureThreshold (device, thresholdType, temp);
+
+  if (nvml_rc != NVML_SUCCESS)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetTemperatureThreshold(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetCurrPcieLinkWidth (void *hashcat_ctx, nvmlDevice_t device, unsigned int *currLinkWidth)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetCurrPcieLinkWidth (device, currLinkWidth);
+
+  if (nvml_rc != NVML_SUCCESS)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetCurrPcieLinkWidth(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}
+
+int hm_NVML_nvmlDeviceGetPciInfo (void *hashcat_ctx, nvmlDevice_t device, nvmlPciInfo_t *pci)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  NVML_PTR *nvml = (NVML_PTR *) hwmon_ctx->hm_nvml;
+
+  const nvmlReturn_t nvml_rc = nvml->nvmlDeviceGetPciInfo (device, pci);
+
+  if (nvml_rc != NVML_SUCCESS)
+  {
+    const char *string = hm_NVML_nvmlErrorString (nvml, nvml_rc);
+
+    event_log_error (hashcat_ctx, "nvmlDeviceGetPciInfo(): %s", string);
+
+    return -1;
+  }
+
+  return 0;
+}

src/ext_sysfs.c

@@ -4,4 +4,440 @@
  */
 
 #include "common.h"
+#include "types.h"
+#include "memory.h"
+#include "shared.h"
+#include "event.h"
+#include "folder.h"
 #include "ext_sysfs.h"
+
+bool sysfs_init (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  SYSFS_PTR *sysfs = (SYSFS_PTR *) hwmon_ctx->hm_sysfs;
+
+  memset (sysfs, 0, sizeof (SYSFS_PTR));
+
+  char *path;
+
+  hc_asprintf (&path, "%s", SYS_BUS_PCI_DEVICES);
+
+  const bool r = hc_path_read (path);
+
+  hcfree (path);
+
+  return r;
+}
+
+void sysfs_close (void *hashcat_ctx)
+{
+  hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;
+
+  SYSFS_PTR *sysfs = (SYSFS_PTR *) hwmon_ctx->hm_sysfs;
+
+  if (sysfs)
+  {
+    hcfree (sysfs);
+  }
+}
+
+char *hm_SYSFS_get_syspath_device (void *hashcat_ctx, const int backend_device_idx)
+{
+  backend_ctx_t *backend_ctx = ((hashcat_ctx_t *) hashcat_ctx)->backend_ctx;
+
+  hc_device_param_t *device_param = &backend_ctx->devices_param[backend_device_idx];
+
+  char *syspath;
+
+  hc_asprintf (&syspath, "%s/0000:%02x:%02x.%01x", SYS_BUS_PCI_DEVICES, device_param->pcie_bus, device_param->pcie_device, device_param->pcie_function);
+
+  return syspath;
+}
+
+char *hm_SYSFS_get_syspath_hwmon (void *hashcat_ctx, const int backend_device_idx)
+{
+  char *syspath = hm_SYSFS_get_syspath_device (hashcat_ctx, backend_device_idx);
+
+  if (syspath == NULL)
+  {
+    event_log_error (hashcat_ctx, "hm_SYSFS_get_syspath_device() failed.");
+
+    return NULL;
+  }
+
+  char *hwmon = (char *) hcmalloc (HCBUFSIZ_TINY);
+
+  snprintf (hwmon, HCBUFSIZ_TINY, "%s/hwmon", syspath);
+
+  char *hwmonN = first_file_in_directory (hwmon);
+
+  if (hwmonN == NULL)
+  {
+    event_log_error (hashcat_ctx, "First_file_in_directory() failed.");
+
+    hcfree (syspath);
+
+    hcfree (hwmon);
+    hcfree (hwmonN);
+
+    return NULL;
+  }
+
+  snprintf (hwmon, HCBUFSIZ_TINY, "%s/hwmon/%s", syspath, hwmonN);
+
+  hcfree (syspath);
+
+  hcfree (hwmonN);
+
+  return hwmon;
+}
+
+int hm_SYSFS_get_fan_speed_current (void *hashcat_ctx, const int backend_device_idx, int *val)
+{
+  char *syspath = hm_SYSFS_get_syspath_hwmon (hashcat_ctx, backend_device_idx);
+
+  if (syspath == NULL) return -1;
+
+  char *path_cur;
+  char *path_max;
+
+  hc_asprintf (&path_cur, "%s/pwm1",     syspath);
+  hc_asprintf (&path_max, "%s/pwm1_max", syspath);
+
+  hcfree (syspath);
+
+  HCFILE fp_cur;
+
+  if (hc_fopen (&fp_cur, path_cur, "r") == false)
+  {
+    event_log_error (hashcat_ctx, "%s: %s", path_cur, strerror (errno));
+
+    hcfree (path_cur);
+    hcfree (path_max);
+
+    return -1;
+  }
+
+  int pwm1_cur = 0;
+
+  if (hc_fscanf (&fp_cur, "%d", &pwm1_cur) != 1)
+  {
+    hc_fclose (&fp_cur);
+
+    event_log_error (hashcat_ctx, "%s: unexpected data.", path_cur);
+
+    hcfree (path_cur);
+    hcfree (path_max);
+
+    return -1;
+  }
+
+  hc_fclose (&fp_cur);
+
+  HCFILE fp_max;
+
+  if (hc_fopen (&fp_max, path_max, "r") == false)
+  {
+    event_log_error (hashcat_ctx, "%s: %s", path_max, strerror (errno));
+
+    hcfree (path_cur);
+    hcfree (path_max);
+
+    return -1;
+  }
+
+  int pwm1_max = 0;
+
+  if (hc_fscanf (&fp_max, "%d", &pwm1_max) != 1)
+  {
+    hc_fclose (&fp_max);
+
+    event_log_error (hashcat_ctx, "%s: unexpected data.", path_max);
+
+    hcfree (path_cur);
+    hcfree (path_max);
+
+    return -1;
+  }
+
+  hc_fclose (&fp_max);
+
+  if (pwm1_max == 0)
+  {
+    event_log_error (hashcat_ctx, "%s: pwm1_max cannot be 0.", path_max);
+
+    hcfree (path_cur);
+    hcfree (path_max);
+
+    return -1;
+  }
+
+  const float p1 = (float) pwm1_max / 100.0F;
+
+  const float pwm1_percent = (float) pwm1_cur / p1;
+
+  *val = (int) pwm1_percent;
+
+  hcfree (path_cur);
+  hcfree (path_max);
+
+  return 0;
+}
+
+int hm_SYSFS_get_temperature_current (void *hashcat_ctx, const int backend_device_idx, int *val)
+{
+  char *syspath = hm_SYSFS_get_syspath_hwmon (hashcat_ctx, backend_device_idx);
+
+  if (syspath == NULL) return -1;
+
+  char *path;
+
+  hc_asprintf (&path, "%s/temp1_input", syspath);
+
+  hcfree (syspath);
+
+  HCFILE fp;
+
+  if (hc_fopen (&fp, path, "r") == false)
+  {
+    event_log_error (hashcat_ctx, "%s: %s", path, strerror (errno));
+
+    hcfree (path);
+
+    return -1;
+  }
+
+  int temperature = 0;
+
+  if (hc_fscanf (&fp, "%d", &temperature) != 1)
+  {
+    hc_fclose (&fp);
+
+    event_log_error (hashcat_ctx, "%s: unexpected data.", path);
+
+    hcfree (path);
+
+    return -1;
+  }
+
+  hc_fclose (&fp);
+
+  *val = temperature / 1000;
+
+  hcfree (path);
+
+  return 0;
+}
+
+int hm_SYSFS_get_pp_dpm_sclk (void *hashcat_ctx, const int backend_device_idx, int *val)
+{
+  char *syspath = hm_SYSFS_get_syspath_device (hashcat_ctx, backend_device_idx);
+
+  if (syspath == NULL) return -1;
+
+  char *path;
+
+  hc_asprintf (&path, "%s/pp_dpm_sclk", syspath);
+
+  hcfree (syspath);
+
+  HCFILE fp;
+
+  if (hc_fopen (&fp, path, "r") == false)
+  {
+    event_log_error (hashcat_ctx, "%s: %s", path, strerror (errno));
+
+    hcfree (path);
+
+    return -1;
+  }
+
+  int clockfreq = 0;
+
+  while (!hc_feof (&fp))
+  {
+    char buf[HCBUFSIZ_TINY] = { 0 };
+
+    char *ptr = hc_fgets (buf, sizeof (buf), &fp);
+
+    if (ptr == NULL) continue;
+
+    size_t len = strlen (ptr);
+
+    if (len < 2) continue;
+
+    if (ptr[len - 2] != '*') continue;
+
+    int profile = 0;
+
+    int rc = sscanf (ptr, "%d: %dMHz", &profile, &clockfreq);
+
+    if (rc == 2) break;
+  }
+
+  hc_fclose (&fp);
+
+  *val = clockfreq;
+
+  hcfree (path);
+
+  return 0;
+}
+
+int hm_SYSFS_get_pp_dpm_mclk (void *hashcat_ctx, const int backend_device_idx, int *val)
+{
+  char *syspath = hm_SYSFS_get_syspath_device (hashcat_ctx, backend_device_idx);
+
+  if (syspath == NULL) return -1;
+
+  char *path;
+
+  hc_asprintf (&path, "%s/pp_dpm_mclk", syspath);
+
+  hcfree (syspath);
+
+  HCFILE fp;
+
+  if (hc_fopen (&fp, path, "r") == false)
+  {
+    event_log_error (hashcat_ctx, "%s: %s", path, strerror (errno));
+
+    hcfree (path);
+
+    return -1;
+  }
+
+  int clockfreq = 0;
+
+  while (!hc_feof (&fp))
+  {
+    char buf[HCBUFSIZ_TINY];
+
+    char *ptr = hc_fgets (buf, sizeof (buf), &fp);
+
+    if (ptr == NULL) continue;
+
+    size_t len = strlen (ptr);
+
+    if (len < 2) continue;
+
+    if (ptr[len - 2] != '*') continue;
+
+    int profile = 0;
+
+    int rc = sscanf (ptr, "%d: %dMHz", &profile, &clockfreq);
+
+    if (rc == 2) break;
+  }
+
+  hc_fclose (&fp);
+
+  *val = clockfreq;
+
+  hcfree (path);
+
+  return 0;
+}
+
+int hm_SYSFS_get_pp_dpm_pcie (void *hashcat_ctx, const int backend_device_idx, int *val)
+{
+  char *syspath = hm_SYSFS_get_syspath_device (hashcat_ctx, backend_device_idx);
+
+  if (syspath == NULL) return -1;
+
+  char *path;
+
+  hc_asprintf (&path, "%s/current_link_width", syspath);
+
+  hcfree (syspath);
+
+  HCFILE fp;
+
+  if (hc_fopen (&fp, path, "r") == false)
+  {
+    event_log_error (hashcat_ctx, "%s: %s", path, strerror (errno));
+
+    hcfree (path);
+
+    return -1;
+  }
+
+  int lanes = 0;
+
+  while (!hc_feof (&fp))
+  {
+    char buf[HCBUFSIZ_TINY];
+
+    char *ptr = hc_fgets (buf, sizeof (buf), &fp);
+
+    if (ptr == NULL) continue;
+
+    size_t len = strlen (ptr);
+
+    if (len < 2) continue;
+
+    int rc = sscanf (ptr, "%d", &lanes);
+
+    if (rc == 1) break;
+  }
+
+  hc_fclose (&fp);
+
+  *val = lanes;
+
+  hcfree (path);
+
+  return 0;
+}
+
+int hm_SYSFS_get_gpu_busy_percent (void *hashcat_ctx, const int backend_device_idx, int *val)
+{
+  char *syspath = hm_SYSFS_get_syspath_device (hashcat_ctx, backend_device_idx);
+
+  if (syspath == NULL) return -1;
+
+  char *path;
+
+  hc_asprintf (&path, "%s/gpu_busy_percent", syspath);
+
+  hcfree (syspath);
+
+  HCFILE fp;
+
+  if (hc_fopen (&fp, path, "r") == false)
+  {
+    event_log_error (hashcat_ctx, "%s: %s", path, strerror (errno));
+
+    hcfree (path);
+
+    return -1;
+  }
+
+  int util = 0;
+
+  while (!hc_feof (&fp))
+  {
+    char buf[HCBUFSIZ_TINY];
+
+    char *ptr = hc_fgets (buf, sizeof (buf), &fp);
+
+    if (ptr == NULL) continue;
+
+    size_t len = strlen (ptr);
+
+    if (len < 1) continue;
+
+    int rc = sscanf (ptr, "%d", &util);
+
+    if (rc == 1) break;
+  }
+
+  hc_fclose (&fp);
+
+  *val = util;
+
+  hcfree (path);
+
+  return 0;
+}

src/hashcat.c

@@ -1271,6 +1271,8 @@ bool autodetect_hashmode_test (hashcat_ctx_t *hashcat_ctx)
   {
     HCFILE fp;
 
+    int error_count = 0;
+
     if (hc_fopen (&fp, hashfile, "rb") == false) return false;
 
     char *line_buf = (char *) hcmalloc (HCBUFSIZ_LARGE);
@@ -1301,6 +1303,17 @@ bool autodetect_hashmode_test (hashcat_ctx_t *hashcat_ctx)
 
         break;
       }
+
+      // abort this list after 100 errors
+
+      if (error_count == 100)
+      {
+        break;
+      }
+      else
+      {
+        error_count++;
+      }
     }
 
     hcfree (line_buf);
@@ -1463,43 +1476,87 @@ int hashcat_session_execute (hashcat_ctx_t *hashcat_ctx)
     }
 
     if (modes_cnt > 1)
+    {
+      if (user_options->machine_readable == false)
       {
         event_log_info (hashcat_ctx, "The following %d hash-modes match the structure of your input hash:", modes_cnt);
         event_log_info (hashcat_ctx, NULL);
         event_log_info (hashcat_ctx, "      # | Name                                                | Category");
         event_log_info (hashcat_ctx, "  ======+=====================================================+======================================");
+      }
 
       for (int i = 0; i < modes_cnt; i++)
+      {
+        if (user_options->machine_readable == false)
         {
           event_log_info (hashcat_ctx, "%7u | %-51s | %s", usage_sort_buf[i].hash_mode, usage_sort_buf[i].hash_name, strhashcategory (usage_sort_buf[i].hash_category));
+        }
+        else
+        {
+          event_log_info (hashcat_ctx, "%u", usage_sort_buf[i].hash_mode);
+        }
 
         hcfree (usage_sort_buf[i].hash_name);
       }
 
-      event_log_info (hashcat_ctx, NULL);
-
-      event_log_error (hashcat_ctx, "Please specify the hash-mode with -m [hash-mode].");
-
       hcfree (usage_sort_buf);
 
+      if (user_options->machine_readable == false) event_log_info (hashcat_ctx, NULL);
+
+      if (user_options->identify == false)
+      {
+        event_log_error (hashcat_ctx, "Please specify the hash-mode with -m [hash-mode].");
+
         return -1;
       }
 
+      return 0;
+    }
+
     // modes_cnt == 1
 
+    if (user_options->identify == false)
+    {
       event_log_warning (hashcat_ctx, "Hash-mode was not specified with -m. Attempting to auto-detect hash mode.");
       event_log_warning (hashcat_ctx, "The following mode was auto-detected as the only one matching your input hash:");
-    event_log_warning (hashcat_ctx, "\n%u | %s | %s\n", usage_sort_buf[0].hash_mode, usage_sort_buf[0].hash_name, strhashcategory (usage_sort_buf[0].hash_category));
+    }
+
+    if (user_options->identify == true)
+    {
+      if (user_options->machine_readable == true)
+      {
+        event_log_info (hashcat_ctx, "%u", usage_sort_buf[0].hash_mode);
+      }
+      else
+      {
+        event_log_info (hashcat_ctx, "The following hash-mode match the structure of your input hash:");
+        event_log_info (hashcat_ctx, NULL);
+        event_log_info (hashcat_ctx, "      # | Name                                                | Category");
+        event_log_info (hashcat_ctx, "  ======+=====================================================+======================================");
+        event_log_info (hashcat_ctx, "%7u | %-51s | %s", usage_sort_buf[0].hash_mode, usage_sort_buf[0].hash_name, strhashcategory (usage_sort_buf[0].hash_category));
+        event_log_info (hashcat_ctx, NULL);
+      }
+    }
+    else
+    {
+      event_log_info (hashcat_ctx, "\n%u | %s | %s\n", usage_sort_buf[0].hash_mode, usage_sort_buf[0].hash_name, strhashcategory (usage_sort_buf[0].hash_category));
+    }
+
+    if (user_options->identify == false)
+    {
       event_log_warning (hashcat_ctx, "NOTE: Auto-detect is best effort. The correct hash-mode is NOT guaranteed!");
       event_log_warning (hashcat_ctx, "Do NOT report auto-detect issues unless you are certain of the hash type.");
       event_log_warning (hashcat_ctx, NULL);
-
-    user_options->autodetect = false;
+    }
 
     user_options->hash_mode = usage_sort_buf[0].hash_mode;
 
     hcfree (usage_sort_buf[0].hash_name);
     hcfree (usage_sort_buf);
+
+    if (user_options->identify == true) return 0;
+
+    user_options->autodetect = false;
   }
 
   /**
@@ -1731,6 +1788,8 @@ int hashcat_get_status (hashcat_ctx_t *hashcat_ctx, hashcat_status_t *hashcat_st
   #ifdef WITH_BRAIN
   hashcat_status->brain_session               = status_get_brain_session              (hashcat_ctx);
   hashcat_status->brain_attack                = status_get_brain_attack               (hashcat_ctx);
+  hashcat_status->brain_rx_all                = status_get_brain_rx_all               (hashcat_ctx);
+  hashcat_status->brain_tx_all                = status_get_brain_tx_all               (hashcat_ctx);
   #endif
   hashcat_status->status_string               = status_get_status_string              (hashcat_ctx);
   hashcat_status->status_number               = status_get_status_number              (hashcat_ctx);
@@ -1762,6 +1821,9 @@ int hashcat_get_status (hashcat_ctx_t *hashcat_ctx, hashcat_status_t *hashcat_st
     device_info->exec_msec_dev                  = status_get_exec_msec_dev                  (hashcat_ctx, device_id);
     device_info->speed_sec_dev                  = status_get_speed_sec_dev                  (hashcat_ctx, device_id);
     device_info->guess_candidates_dev           = status_get_guess_candidates_dev           (hashcat_ctx, device_id);
+    #if defined(__APPLE__)
+    device_info->hwmon_fan_dev                  = status_get_hwmon_fan_dev                  (hashcat_ctx);
+    #endif
     device_info->hwmon_dev                      = status_get_hwmon_dev                      (hashcat_ctx, device_id);
     device_info->corespeed_dev                  = status_get_corespeed_dev                  (hashcat_ctx, device_id);
     device_info->memoryspeed_dev                = status_get_memoryspeed_dev                (hashcat_ctx, device_id);
@@ -1783,8 +1845,6 @@ int hashcat_get_status (hashcat_ctx_t *hashcat_ctx, hashcat_status_t *hashcat_st
     device_info->brain_link_send_bytes_dev      = status_get_brain_link_send_bytes_dev      (hashcat_ctx, device_id);
     device_info->brain_link_recv_bytes_sec_dev  = status_get_brain_link_recv_bytes_sec_dev  (hashcat_ctx, device_id);
     device_info->brain_link_send_bytes_sec_dev  = status_get_brain_link_send_bytes_sec_dev  (hashcat_ctx, device_id);
-    hashcat_status->brain_rx_all   = status_get_brain_rx_all   (hashcat_ctx);
-    hashcat_status->brain_tx_all   = status_get_brain_tx_all   (hashcat_ctx);
     #endif
   }
 

src/induct.c

@@ -50,6 +50,7 @@ int induct_ctx_init (hashcat_ctx_t *hashcat_ctx)
   if (user_options->progress_only == true) return 0;
   if (user_options->usage         == true) return 0;
   if (user_options->version       == true) return 0;
+  if (user_options->identify      == true) return 0;
 
   if ((user_options->attack_mode != ATTACK_MODE_STRAIGHT)
    && (user_options->attack_mode != ATTACK_MODE_ASSOCIATION)) return 0;

src/loopback.c

@@ -71,6 +71,7 @@ int loopback_init (hashcat_ctx_t *hashcat_ctx)
   if (user_options->progress_only == true) return 0;
   if (user_options->usage         == true) return 0;
   if (user_options->version       == true) return 0;
+  if (user_options->identify      == true) return 0;
 
   loopback_ctx->enabled  = true;
   loopback_ctx->fp.pfp   = NULL;

src/main.c

@@ -189,6 +189,7 @@ static void main_outerloop_starting (MAYBE_UNUSED hashcat_ctx_t *hashcat_ctx, MA
   if (user_options->stdout_flag    == true) return;
   if (user_options->backend_info   == true) return;
   if (user_options->speed_only     == true) return;
+  if (user_options->identify       == true) return;
 
   if ((user_options_extra->wordlist_mode == WL_MODE_FILE) || (user_options_extra->wordlist_mode == WL_MODE_MASK))
   {

src/modules/module_06211.c

@@ -59,7 +59,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -204,13 +206,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06212.c

@@ -59,7 +59,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -204,13 +206,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06213.c

@@ -59,7 +59,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -202,13 +204,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06221.c

@@ -60,7 +60,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -138,8 +140,6 @@ int module_hash_init_selftest (MAYBE_UNUSED const hashconfig_t *hashconfig, hash
 
   const int parser_status = module_hash_decode (hashconfig, hash->digest, hash->salt, hash->esalt, hash->hook_salt, hash->hash_info, tmpdata, st_hash_len / 2);
 
-
-
   hcfree (tmpdata);
 
   return parser_status;
@@ -189,13 +189,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06222.c

@@ -60,7 +60,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -138,8 +140,6 @@ int module_hash_init_selftest (MAYBE_UNUSED const hashconfig_t *hashconfig, hash
 
   const int parser_status = module_hash_decode (hashconfig, hash->digest, hash->salt, hash->esalt, hash->hook_salt, hash->hash_info, tmpdata, st_hash_len / 2);
 
-
-
   hcfree (tmpdata);
 
   return parser_status;
@@ -189,13 +189,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06223.c

@@ -60,7 +60,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -187,13 +189,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06231.c

@@ -59,7 +59,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -197,13 +199,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06232.c

@@ -59,7 +59,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -197,13 +199,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06233.c

@@ -59,7 +59,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -193,13 +195,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06241.c

@@ -60,7 +60,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -198,13 +200,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06242.c

@@ -60,7 +60,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -198,13 +200,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_06243.c

@@ -60,7 +60,9 @@ typedef struct tc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -198,13 +200,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) tc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    tc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping

src/modules/module_07000.c

@@ -243,7 +243,7 @@ void module_init (module_ctx_t *module_ctx)
   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_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
   module_ctx->module_pw_min                   = MODULE_DEFAULT;
   module_ctx->module_salt_max                 = MODULE_DEFAULT;
   module_ctx->module_salt_min                 = MODULE_DEFAULT;

src/modules/module_07400.c

@@ -413,7 +413,7 @@ void module_init (module_ctx_t *module_ctx)
   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_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
   module_ctx->module_pw_min                   = MODULE_DEFAULT;
   module_ctx->module_salt_max                 = MODULE_DEFAULT;
   module_ctx->module_salt_min                 = MODULE_DEFAULT;

src/modules/module_07401.c

@@ -458,7 +458,7 @@ void module_init (module_ctx_t *module_ctx)
   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_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
   module_ctx->module_pw_min                   = MODULE_DEFAULT;
   module_ctx->module_salt_max                 = MODULE_DEFAULT;
   module_ctx->module_salt_min                 = MODULE_DEFAULT;

src/modules/module_07800.c

@@ -219,7 +219,7 @@ void module_init (module_ctx_t *module_ctx)
   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_DEFAULT;
+  module_ctx->module_pw_max                   = module_pw_max;
   module_ctx->module_pw_min                   = MODULE_DEFAULT;
   module_ctx->module_salt_max                 = MODULE_DEFAULT;
   module_ctx->module_salt_min                 = MODULE_DEFAULT;

src/modules/module_08900.c

@@ -243,11 +243,6 @@ u64 module_extra_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UN
   return tmp_size;
 }
 
-bool module_jit_cache_disable (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra, MAYBE_UNUSED const hashes_t *hashes, MAYBE_UNUSED const hc_device_param_t *device_param)
-{
-  return true;
-}
-
 bool module_warmup_disable (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 true;
@@ -438,7 +433,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
   module_ctx->module_hook_size                = MODULE_DEFAULT;
   module_ctx->module_jit_build_options        = module_jit_build_options;
-  module_ctx->module_jit_cache_disable        = module_jit_cache_disable;
+  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_kernel_loops_max;

src/modules/module_09300.c

@@ -235,11 +235,6 @@ u64 module_extra_tmp_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UN
   return tmp_size;
 }
 
-bool module_jit_cache_disable (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra, MAYBE_UNUSED const hashes_t *hashes, MAYBE_UNUSED const hc_device_param_t *device_param)
-{
-  return true;
-}
-
 bool module_warmup_disable (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 true;
@@ -395,7 +390,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_hook_salt_size           = MODULE_DEFAULT;
   module_ctx->module_hook_size                = MODULE_DEFAULT;
   module_ctx->module_jit_build_options        = module_jit_build_options;
-  module_ctx->module_jit_cache_disable        = module_jit_cache_disable;
+  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_kernel_loops_max;

src/modules/module_13711.c

@@ -65,7 +65,9 @@ typedef struct vc
 {
   u32 salt_buf[32];
   u32 data_buf[112];
-  u32 keyfile_buf[16];
+  u32 keyfile_buf16[16];
+  u32 keyfile_buf32[32];
+  u32 keyfile_enabled;
   u32 signature;
 
   keyboard_layout_mapping_t keyboard_layout_mapping_buf[256];
@@ -145,7 +147,7 @@ u32 module_pw_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED con
   // this overrides the reductions of PW_MAX in case optimized kernel is selected
   // IOW, even in optimized kernel mode it support length 64
 
-  const u32 pw_max = 64;
+  const u32 pw_max = 64; // VC nowadays support 128, but RipeMD160 for container were removed before update from 64->128
 
   return pw_max;
 }
@@ -214,13 +216,16 @@ int module_hash_binary_parse (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE
     {
       if (hc_path_read (keyfile))
       {
-        cpu_crc32 (keyfile, (u8 *) vc->keyfile_buf);
+        cpu_crc32 (keyfile, (u8 *) vc->keyfile_buf16,  64);
+        cpu_crc32 (keyfile, (u8 *) vc->keyfile_buf32, 128);
       }
 
       keyfile = strtok_r ((char *) NULL, ",", &saveptr);
     }
 
     hcfree (keyfiles);
+
+    vc->keyfile_enabled = 1;
   }
 
   // keyboard layout mapping