X-Git-Url: https://www.flypig.org.uk/git/?a=blobdiff_plain;f=OpenCL%2Fm09900_a0.cl;h=fef8b08dbe7cc974eaf2d888783ced8bda1b4b0d;hb=8702d0e3e155e6a61ca1b0dcdc1ec924d92b8436;hp=5100cd48ca551fe987efdbad2c82fac79ffe726d;hpb=db1d480a02048850574c776cb2649eb86cd6993c;p=hashcat.git

diff --git a/OpenCL/m09900_a0.cl b/OpenCL/m09900_a0.cl
index 5100cd4..fef8b08 100644
--- a/OpenCL/m09900_a0.cl
+++ b/OpenCL/m09900_a0.cl
@@ -7,22 +7,16 @@
 
 #define NEW_SIMD_CODE
 
-#include "include/constants.h"
-#include "include/kernel_vendor.h"
-
-#define DGST_R0 0
-#define DGST_R1 3
-#define DGST_R2 2
-#define DGST_R3 1
-
-#include "include/kernel_functions.c"
-#include "OpenCL/types_ocl.c"
-#include "OpenCL/common.c"
-#include "include/rp_kernel.h"
-#include "OpenCL/rp.c"
-#include "OpenCL/simd.c"
-
-__kernel void m09900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+#include "inc_vendor.cl"
+#include "inc_hash_constants.h"
+#include "inc_hash_functions.cl"
+#include "inc_types.cl"
+#include "inc_common.cl"
+#include "inc_rp.h"
+#include "inc_rp.cl"
+#include "inc_simd.cl"
+
+__kernel void m09900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * modifier
@@ -39,14 +33,12 @@ __kernel void m09900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
   if (gid >= gid_max) return;
 
   u32 pw_buf0[4];
+  u32 pw_buf1[4];
 
   pw_buf0[0] = pws[gid].i[ 0];
   pw_buf0[1] = pws[gid].i[ 1];
   pw_buf0[2] = pws[gid].i[ 2];
   pw_buf0[3] = pws[gid].i[ 3];
-
-  u32 pw_buf1[4];
-
   pw_buf1[0] = pws[gid].i[ 4];
   pw_buf1[1] = pws[gid].i[ 5];
   pw_buf1[2] = pws[gid].i[ 6];
@@ -58,7 +50,7 @@ __kernel void m09900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
    * loop
    */
 
-  for (u32 il_pos = 0; il_pos < rules_cnt; il_pos += VECT_SIZE)
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
   {
     u32x w0[4] = { 0 };
     u32x w1[4] = { 0 };
@@ -67,6 +59,10 @@ __kernel void m09900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
 
     apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
 
+    /**
+     * md5
+     */
+
     u32x a = MD5M_A;
     u32x b = MD5M_B;
     u32x c = MD5M_C;
@@ -150,95 +146,90 @@ __kernel void m09900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
     u32x r_c = c;
     u32x r_d = d;
 
-    u32x t0[4];
-    u32x t1[4];
-    u32x t2[4];
-    u32x t3[4];
-
-    t0[0] = 0;
-    t0[1] = 0;
-    t0[2] = 0;
-    t0[3] = 0;
-    t1[0] = 0;
-    t1[1] = 0;
-    t1[2] = 0;
-    t1[3] = 0;
-    t2[0] = 0;
-    t2[1] = 0x80;
-    t2[2] = 0;
-    t2[3] = 0;
-    t3[0] = 0;
-    t3[1] = 0;
-    t3[2] = 100 * 8;
-    t3[3] = 0;
-
-    MD5_STEP (MD5_Fo, a, b, c, d, t0[0], MD5C00, MD5S00);
-    MD5_STEP (MD5_Fo, d, a, b, c, t0[1], MD5C01, MD5S01);
-    MD5_STEP (MD5_Fo, c, d, a, b, t0[2], MD5C02, MD5S02);
-    MD5_STEP (MD5_Fo, b, c, d, a, t0[3], MD5C03, MD5S03);
-    MD5_STEP (MD5_Fo, a, b, c, d, t1[0], MD5C04, MD5S00);
-    MD5_STEP (MD5_Fo, d, a, b, c, t1[1], MD5C05, MD5S01);
-    MD5_STEP (MD5_Fo, c, d, a, b, t1[2], MD5C06, MD5S02);
-    MD5_STEP (MD5_Fo, b, c, d, a, t1[3], MD5C07, MD5S03);
-    MD5_STEP (MD5_Fo, a, b, c, d, t2[0], MD5C08, MD5S00);
-    MD5_STEP (MD5_Fo, d, a, b, c, t2[1], MD5C09, MD5S01);
-    MD5_STEP (MD5_Fo, c, d, a, b, t2[2], MD5C0a, MD5S02);
-    MD5_STEP (MD5_Fo, b, c, d, a, t2[3], MD5C0b, MD5S03);
-    MD5_STEP (MD5_Fo, a, b, c, d, t3[0], MD5C0c, MD5S00);
-    MD5_STEP (MD5_Fo, d, a, b, c, t3[1], MD5C0d, MD5S01);
-    MD5_STEP (MD5_Fo, c, d, a, b, t3[2], MD5C0e, MD5S02);
-    MD5_STEP (MD5_Fo, b, c, d, a, t3[3], MD5C0f, MD5S03);
-
-    MD5_STEP (MD5_Go, a, b, c, d, t0[1], MD5C10, MD5S10);
-    MD5_STEP (MD5_Go, d, a, b, c, t1[2], MD5C11, MD5S11);
-    MD5_STEP (MD5_Go, c, d, a, b, t2[3], MD5C12, MD5S12);
-    MD5_STEP (MD5_Go, b, c, d, a, t0[0], MD5C13, MD5S13);
-    MD5_STEP (MD5_Go, a, b, c, d, t1[1], MD5C14, MD5S10);
-    MD5_STEP (MD5_Go, d, a, b, c, t2[2], MD5C15, MD5S11);
-    MD5_STEP (MD5_Go, c, d, a, b, t3[3], MD5C16, MD5S12);
-    MD5_STEP (MD5_Go, b, c, d, a, t1[0], MD5C17, MD5S13);
-    MD5_STEP (MD5_Go, a, b, c, d, t2[1], MD5C18, MD5S10);
-    MD5_STEP (MD5_Go, d, a, b, c, t3[2], MD5C19, MD5S11);
-    MD5_STEP (MD5_Go, c, d, a, b, t0[3], MD5C1a, MD5S12);
-    MD5_STEP (MD5_Go, b, c, d, a, t2[0], MD5C1b, MD5S13);
-    MD5_STEP (MD5_Go, a, b, c, d, t3[1], MD5C1c, MD5S10);
-    MD5_STEP (MD5_Go, d, a, b, c, t0[2], MD5C1d, MD5S11);
-    MD5_STEP (MD5_Go, c, d, a, b, t1[3], MD5C1e, MD5S12);
-    MD5_STEP (MD5_Go, b, c, d, a, t3[0], MD5C1f, MD5S13);
-
-    MD5_STEP (MD5_H , a, b, c, d, t1[1], MD5C20, MD5S20);
-    MD5_STEP (MD5_H , d, a, b, c, t2[0], MD5C21, MD5S21);
-    MD5_STEP (MD5_H , c, d, a, b, t2[3], MD5C22, MD5S22);
-    MD5_STEP (MD5_H , b, c, d, a, t3[2], MD5C23, MD5S23);
-    MD5_STEP (MD5_H , a, b, c, d, t0[1], MD5C24, MD5S20);
-    MD5_STEP (MD5_H , d, a, b, c, t1[0], MD5C25, MD5S21);
-    MD5_STEP (MD5_H , c, d, a, b, t1[3], MD5C26, MD5S22);
-    MD5_STEP (MD5_H , b, c, d, a, t2[2], MD5C27, MD5S23);
-    MD5_STEP (MD5_H , a, b, c, d, t3[1], MD5C28, MD5S20);
-    MD5_STEP (MD5_H , d, a, b, c, t0[0], MD5C29, MD5S21);
-    MD5_STEP (MD5_H , c, d, a, b, t0[3], MD5C2a, MD5S22);
-    MD5_STEP (MD5_H , b, c, d, a, t1[2], MD5C2b, MD5S23);
-    MD5_STEP (MD5_H , a, b, c, d, t2[1], MD5C2c, MD5S20);
-    MD5_STEP (MD5_H , d, a, b, c, t3[0], MD5C2d, MD5S21);
-    MD5_STEP (MD5_H , c, d, a, b, t3[3], MD5C2e, MD5S22);
-    MD5_STEP (MD5_H , b, c, d, a, t0[2], MD5C2f, MD5S23);
-
-    MD5_STEP (MD5_I , a, b, c, d, t0[0], MD5C30, MD5S30);
-    MD5_STEP (MD5_I , d, a, b, c, t1[3], MD5C31, MD5S31);
-    MD5_STEP (MD5_I , c, d, a, b, t3[2], MD5C32, MD5S32);
-    MD5_STEP (MD5_I , b, c, d, a, t1[1], MD5C33, MD5S33);
-    MD5_STEP (MD5_I , a, b, c, d, t3[0], MD5C34, MD5S30);
-    MD5_STEP (MD5_I , d, a, b, c, t0[3], MD5C35, MD5S31);
-    MD5_STEP (MD5_I , c, d, a, b, t2[2], MD5C36, MD5S32);
-    MD5_STEP (MD5_I , b, c, d, a, t0[1], MD5C37, MD5S33);
-    MD5_STEP (MD5_I , a, b, c, d, t2[0], MD5C38, MD5S30);
-    MD5_STEP (MD5_I , d, a, b, c, t3[3], MD5C39, MD5S31);
-    MD5_STEP (MD5_I , c, d, a, b, t1[2], MD5C3a, MD5S32);
-    MD5_STEP (MD5_I , b, c, d, a, t3[1], MD5C3b, MD5S33);
-    MD5_STEP (MD5_I , a, b, c, d, t1[0], MD5C3c, MD5S30);
-    MD5_STEP (MD5_I , d, a, b, c, t2[3], MD5C3d, MD5S31);
-    MD5_STEP (MD5_I , c, d, a, b, t0[2], MD5C3e, MD5S32);
-    MD5_STEP (MD5_I , b, c, d, a, t2[1], MD5C3f, MD5S33);
+    w0[0] = 0;
+    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] = 0x80;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 100 * 8;
+    w3[3] = 0;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H , a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H , d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H , c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H , b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H , a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H , d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H , c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H , b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H , a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H , d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H , c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H , b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H , a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H , d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H , c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H , b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
 
     a += r_a;
     b += r_b;
@@ -249,15 +240,15 @@ __kernel void m09900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
   }
 }
 
-__kernel void m09900_m08 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m09900_m08 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
 }
 
-__kernel void m09900_m16 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m09900_m16 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
 }
 
-__kernel void m09900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m09900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
   /**
    * modifier
@@ -274,14 +265,12 @@ __kernel void m09900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
   if (gid >= gid_max) return;
 
   u32 pw_buf0[4];
+  u32 pw_buf1[4];
 
   pw_buf0[0] = pws[gid].i[ 0];
   pw_buf0[1] = pws[gid].i[ 1];
   pw_buf0[2] = pws[gid].i[ 2];
   pw_buf0[3] = pws[gid].i[ 3];
-
-  u32 pw_buf1[4];
-
   pw_buf1[0] = pws[gid].i[ 4];
   pw_buf1[1] = pws[gid].i[ 5];
   pw_buf1[2] = pws[gid].i[ 6];
@@ -305,7 +294,7 @@ __kernel void m09900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
    * loop
    */
 
-  for (u32 il_pos = 0; il_pos < rules_cnt; il_pos += VECT_SIZE)
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
   {
     u32x w0[4] = { 0 };
     u32x w1[4] = { 0 };
@@ -314,6 +303,10 @@ __kernel void m09900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
 
     apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
 
+    /**
+     * md5
+     */
+
     u32x a = MD5M_A;
     u32x b = MD5M_B;
     u32x c = MD5M_C;
@@ -397,98 +390,93 @@ __kernel void m09900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
     u32x r_c = c;
     u32x r_d = d;
 
-    u32x t0[4];
-    u32x t1[4];
-    u32x t2[4];
-    u32x t3[4];
-
-    t0[0] = 0;
-    t0[1] = 0;
-    t0[2] = 0;
-    t0[3] = 0;
-    t1[0] = 0;
-    t1[1] = 0;
-    t1[2] = 0;
-    t1[3] = 0;
-    t2[0] = 0;
-    t2[1] = 0x80;
-    t2[2] = 0;
-    t2[3] = 0;
-    t3[0] = 0;
-    t3[1] = 0;
-    t3[2] = 100 * 8;
-    t3[3] = 0;
-
-    MD5_STEP (MD5_Fo, a, b, c, d, t0[0], MD5C00, MD5S00);
-    MD5_STEP (MD5_Fo, d, a, b, c, t0[1], MD5C01, MD5S01);
-    MD5_STEP (MD5_Fo, c, d, a, b, t0[2], MD5C02, MD5S02);
-    MD5_STEP (MD5_Fo, b, c, d, a, t0[3], MD5C03, MD5S03);
-    MD5_STEP (MD5_Fo, a, b, c, d, t1[0], MD5C04, MD5S00);
-    MD5_STEP (MD5_Fo, d, a, b, c, t1[1], MD5C05, MD5S01);
-    MD5_STEP (MD5_Fo, c, d, a, b, t1[2], MD5C06, MD5S02);
-    MD5_STEP (MD5_Fo, b, c, d, a, t1[3], MD5C07, MD5S03);
-    MD5_STEP (MD5_Fo, a, b, c, d, t2[0], MD5C08, MD5S00);
-    MD5_STEP (MD5_Fo, d, a, b, c, t2[1], MD5C09, MD5S01);
-    MD5_STEP (MD5_Fo, c, d, a, b, t2[2], MD5C0a, MD5S02);
-    MD5_STEP (MD5_Fo, b, c, d, a, t2[3], MD5C0b, MD5S03);
-    MD5_STEP (MD5_Fo, a, b, c, d, t3[0], MD5C0c, MD5S00);
-    MD5_STEP (MD5_Fo, d, a, b, c, t3[1], MD5C0d, MD5S01);
-    MD5_STEP (MD5_Fo, c, d, a, b, t3[2], MD5C0e, MD5S02);
-    MD5_STEP (MD5_Fo, b, c, d, a, t3[3], MD5C0f, MD5S03);
-
-    MD5_STEP (MD5_Go, a, b, c, d, t0[1], MD5C10, MD5S10);
-    MD5_STEP (MD5_Go, d, a, b, c, t1[2], MD5C11, MD5S11);
-    MD5_STEP (MD5_Go, c, d, a, b, t2[3], MD5C12, MD5S12);
-    MD5_STEP (MD5_Go, b, c, d, a, t0[0], MD5C13, MD5S13);
-    MD5_STEP (MD5_Go, a, b, c, d, t1[1], MD5C14, MD5S10);
-    MD5_STEP (MD5_Go, d, a, b, c, t2[2], MD5C15, MD5S11);
-    MD5_STEP (MD5_Go, c, d, a, b, t3[3], MD5C16, MD5S12);
-    MD5_STEP (MD5_Go, b, c, d, a, t1[0], MD5C17, MD5S13);
-    MD5_STEP (MD5_Go, a, b, c, d, t2[1], MD5C18, MD5S10);
-    MD5_STEP (MD5_Go, d, a, b, c, t3[2], MD5C19, MD5S11);
-    MD5_STEP (MD5_Go, c, d, a, b, t0[3], MD5C1a, MD5S12);
-    MD5_STEP (MD5_Go, b, c, d, a, t2[0], MD5C1b, MD5S13);
-    MD5_STEP (MD5_Go, a, b, c, d, t3[1], MD5C1c, MD5S10);
-    MD5_STEP (MD5_Go, d, a, b, c, t0[2], MD5C1d, MD5S11);
-    MD5_STEP (MD5_Go, c, d, a, b, t1[3], MD5C1e, MD5S12);
-    MD5_STEP (MD5_Go, b, c, d, a, t3[0], MD5C1f, MD5S13);
-
-    MD5_STEP (MD5_H , a, b, c, d, t1[1], MD5C20, MD5S20);
-    MD5_STEP (MD5_H , d, a, b, c, t2[0], MD5C21, MD5S21);
-    MD5_STEP (MD5_H , c, d, a, b, t2[3], MD5C22, MD5S22);
-    MD5_STEP (MD5_H , b, c, d, a, t3[2], MD5C23, MD5S23);
-    MD5_STEP (MD5_H , a, b, c, d, t0[1], MD5C24, MD5S20);
-    MD5_STEP (MD5_H , d, a, b, c, t1[0], MD5C25, MD5S21);
-    MD5_STEP (MD5_H , c, d, a, b, t1[3], MD5C26, MD5S22);
-    MD5_STEP (MD5_H , b, c, d, a, t2[2], MD5C27, MD5S23);
-    MD5_STEP (MD5_H , a, b, c, d, t3[1], MD5C28, MD5S20);
-    MD5_STEP (MD5_H , d, a, b, c, t0[0], MD5C29, MD5S21);
-    MD5_STEP (MD5_H , c, d, a, b, t0[3], MD5C2a, MD5S22);
-    MD5_STEP (MD5_H , b, c, d, a, t1[2], MD5C2b, MD5S23);
-    MD5_STEP (MD5_H , a, b, c, d, t2[1], MD5C2c, MD5S20);
-    MD5_STEP (MD5_H , d, a, b, c, t3[0], MD5C2d, MD5S21);
-    MD5_STEP (MD5_H , c, d, a, b, t3[3], MD5C2e, MD5S22);
-    MD5_STEP (MD5_H , b, c, d, a, t0[2], MD5C2f, MD5S23);
-
-    MD5_STEP (MD5_I , a, b, c, d, t0[0], MD5C30, MD5S30);
-    MD5_STEP (MD5_I , d, a, b, c, t1[3], MD5C31, MD5S31);
-    MD5_STEP (MD5_I , c, d, a, b, t3[2], MD5C32, MD5S32);
-    MD5_STEP (MD5_I , b, c, d, a, t1[1], MD5C33, MD5S33);
-    MD5_STEP (MD5_I , a, b, c, d, t3[0], MD5C34, MD5S30);
-    MD5_STEP (MD5_I , d, a, b, c, t0[3], MD5C35, MD5S31);
-    MD5_STEP (MD5_I , c, d, a, b, t2[2], MD5C36, MD5S32);
-    MD5_STEP (MD5_I , b, c, d, a, t0[1], MD5C37, MD5S33);
-    MD5_STEP (MD5_I , a, b, c, d, t2[0], MD5C38, MD5S30);
-    MD5_STEP (MD5_I , d, a, b, c, t3[3], MD5C39, MD5S31);
-    MD5_STEP (MD5_I , c, d, a, b, t1[2], MD5C3a, MD5S32);
-    MD5_STEP (MD5_I , b, c, d, a, t3[1], MD5C3b, MD5S33);
-    MD5_STEP (MD5_I , a, b, c, d, t1[0], MD5C3c, MD5S30);
+    w0[0] = 0;
+    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] = 0x80;
+    w2[2] = 0;
+    w2[3] = 0;
+    w3[0] = 0;
+    w3[1] = 0;
+    w3[2] = 100 * 8;
+    w3[3] = 0;
+
+    MD5_STEP (MD5_Fo, a, b, c, d, w0[0], MD5C00, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w0[1], MD5C01, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w0[2], MD5C02, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w0[3], MD5C03, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w1[0], MD5C04, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w1[1], MD5C05, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w1[2], MD5C06, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w1[3], MD5C07, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w2[0], MD5C08, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w2[1], MD5C09, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w2[2], MD5C0a, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w2[3], MD5C0b, MD5S03);
+    MD5_STEP (MD5_Fo, a, b, c, d, w3[0], MD5C0c, MD5S00);
+    MD5_STEP (MD5_Fo, d, a, b, c, w3[1], MD5C0d, MD5S01);
+    MD5_STEP (MD5_Fo, c, d, a, b, w3[2], MD5C0e, MD5S02);
+    MD5_STEP (MD5_Fo, b, c, d, a, w3[3], MD5C0f, MD5S03);
+
+    MD5_STEP (MD5_Go, a, b, c, d, w0[1], MD5C10, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w1[2], MD5C11, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w2[3], MD5C12, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w0[0], MD5C13, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w1[1], MD5C14, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w2[2], MD5C15, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w3[3], MD5C16, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w1[0], MD5C17, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w2[1], MD5C18, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w3[2], MD5C19, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w0[3], MD5C1a, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w2[0], MD5C1b, MD5S13);
+    MD5_STEP (MD5_Go, a, b, c, d, w3[1], MD5C1c, MD5S10);
+    MD5_STEP (MD5_Go, d, a, b, c, w0[2], MD5C1d, MD5S11);
+    MD5_STEP (MD5_Go, c, d, a, b, w1[3], MD5C1e, MD5S12);
+    MD5_STEP (MD5_Go, b, c, d, a, w3[0], MD5C1f, MD5S13);
+
+    MD5_STEP (MD5_H , a, b, c, d, w1[1], MD5C20, MD5S20);
+    MD5_STEP (MD5_H , d, a, b, c, w2[0], MD5C21, MD5S21);
+    MD5_STEP (MD5_H , c, d, a, b, w2[3], MD5C22, MD5S22);
+    MD5_STEP (MD5_H , b, c, d, a, w3[2], MD5C23, MD5S23);
+    MD5_STEP (MD5_H , a, b, c, d, w0[1], MD5C24, MD5S20);
+    MD5_STEP (MD5_H , d, a, b, c, w1[0], MD5C25, MD5S21);
+    MD5_STEP (MD5_H , c, d, a, b, w1[3], MD5C26, MD5S22);
+    MD5_STEP (MD5_H , b, c, d, a, w2[2], MD5C27, MD5S23);
+    MD5_STEP (MD5_H , a, b, c, d, w3[1], MD5C28, MD5S20);
+    MD5_STEP (MD5_H , d, a, b, c, w0[0], MD5C29, MD5S21);
+    MD5_STEP (MD5_H , c, d, a, b, w0[3], MD5C2a, MD5S22);
+    MD5_STEP (MD5_H , b, c, d, a, w1[2], MD5C2b, MD5S23);
+    MD5_STEP (MD5_H , a, b, c, d, w2[1], MD5C2c, MD5S20);
+    MD5_STEP (MD5_H , d, a, b, c, w3[0], MD5C2d, MD5S21);
+    MD5_STEP (MD5_H , c, d, a, b, w3[3], MD5C2e, MD5S22);
+    MD5_STEP (MD5_H , b, c, d, a, w0[2], MD5C2f, MD5S23);
+
+    MD5_STEP (MD5_I , a, b, c, d, w0[0], MD5C30, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w1[3], MD5C31, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w3[2], MD5C32, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w1[1], MD5C33, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w3[0], MD5C34, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w0[3], MD5C35, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w2[2], MD5C36, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w0[1], MD5C37, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w2[0], MD5C38, MD5S30);
+    MD5_STEP (MD5_I , d, a, b, c, w3[3], MD5C39, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w1[2], MD5C3a, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w3[1], MD5C3b, MD5S33);
+    MD5_STEP (MD5_I , a, b, c, d, w1[0], MD5C3c, MD5S30);
 
     if (MATCHES_NONE_VS ((a + r_a), search[0])) continue;
 
-    MD5_STEP (MD5_I , d, a, b, c, t2[3], MD5C3d, MD5S31);
-    MD5_STEP (MD5_I , c, d, a, b, t0[2], MD5C3e, MD5S32);
-    MD5_STEP (MD5_I , b, c, d, a, t2[1], MD5C3f, MD5S33);
+    MD5_STEP (MD5_I , d, a, b, c, w2[3], MD5C3d, MD5S31);
+    MD5_STEP (MD5_I , c, d, a, b, w0[2], MD5C3e, MD5S32);
+    MD5_STEP (MD5_I , b, c, d, a, w2[1], MD5C3f, MD5S33);
 
     a += r_a;
     b += r_b;
@@ -499,10 +487,10 @@ __kernel void m09900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
   }
 }
 
-__kernel void m09900_s08 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m09900_s08 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
 }
 
-__kernel void m09900_s16 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 rules_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
+__kernel void m09900_s16 (__global pw_t *pws, __global kernel_rule_t *  rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
 {
 }