X-Git-Url: https://www.flypig.org.uk/git/?a=blobdiff_plain;f=OpenCL%2Fm00400.cl;h=9aeaf87292bb7f4a407a759481073692c9de3b07;hb=51e1d11b2d6dff4ae331cff6098169455733c412;hp=46500bd169ea01d5374e8cc08bf01a92de35a886;hpb=0bf4e3c34a6a799ccc34f403bed70119574ca9c8;p=hashcat.git

diff --git a/OpenCL/m00400.cl b/OpenCL/m00400.cl
index 46500bd..9aeaf87 100644
--- a/OpenCL/m00400.cl
+++ b/OpenCL/m00400.cl
@@ -5,22 +5,19 @@
 
 #define _MD5_
 
-#include "include/constants.h"
-#include "include/kernel_vendor.h"
+#define NEW_SIMD_CODE
 
-#define DGST_R0 0
-#define DGST_R1 1
-#define DGST_R2 2
-#define DGST_R3 3
+#include "inc_vendor.cl"
+#include "inc_hash_constants.h"
+#include "inc_hash_functions.cl"
+#include "inc_types.cl"
+#include "inc_common.cl"
+#include "inc_simd.cl"
 
-#include "include/kernel_functions.c"
-#include "types_ocl.c"
-#include "common.c"
+#define COMPARE_S "inc_comp_single.cl"
+#define COMPARE_M "inc_comp_multi.cl"
 
-#define COMPARE_S "check_single_comp4.c"
-#define COMPARE_M "check_multi_comp4.c"
-
-static void md5_transform (const u32 w0[4], const u32 w1[4], const u32 w2[4], const u32 w3[4], u32 digest[4])
+void md5_transform_S (const u32 w0[4], const u32 w1[4], const u32 w2[4], const u32 w3[4], u32 digest[4])
 {
   u32 a = digest[0];
   u32 b = digest[1];
@@ -44,7 +41,103 @@ static void md5_transform (const u32 w0[4], const u32 w1[4], const u32 w2[4], co
   u32 we_t = w3[2];
   u32 wf_t = 0;
 
-  u32 tmp2;
+  MD5_STEP_S (MD5_Fo, a, b, c, d, w0_t, MD5C00, MD5S00);
+  MD5_STEP_S (MD5_Fo, d, a, b, c, w1_t, MD5C01, MD5S01);
+  MD5_STEP_S (MD5_Fo, c, d, a, b, w2_t, MD5C02, MD5S02);
+  MD5_STEP_S (MD5_Fo, b, c, d, a, w3_t, MD5C03, MD5S03);
+  MD5_STEP_S (MD5_Fo, a, b, c, d, w4_t, MD5C04, MD5S00);
+  MD5_STEP_S (MD5_Fo, d, a, b, c, w5_t, MD5C05, MD5S01);
+  MD5_STEP_S (MD5_Fo, c, d, a, b, w6_t, MD5C06, MD5S02);
+  MD5_STEP_S (MD5_Fo, b, c, d, a, w7_t, MD5C07, MD5S03);
+  MD5_STEP_S (MD5_Fo, a, b, c, d, w8_t, MD5C08, MD5S00);
+  MD5_STEP_S (MD5_Fo, d, a, b, c, w9_t, MD5C09, MD5S01);
+  MD5_STEP_S (MD5_Fo, c, d, a, b, wa_t, MD5C0a, MD5S02);
+  MD5_STEP_S (MD5_Fo, b, c, d, a, wb_t, MD5C0b, MD5S03);
+  MD5_STEP_S (MD5_Fo, a, b, c, d, wc_t, MD5C0c, MD5S00);
+  MD5_STEP_S (MD5_Fo, d, a, b, c, wd_t, MD5C0d, MD5S01);
+  MD5_STEP_S (MD5_Fo, c, d, a, b, we_t, MD5C0e, MD5S02);
+  MD5_STEP_S (MD5_Fo, b, c, d, a, wf_t, MD5C0f, MD5S03);
+
+  MD5_STEP_S (MD5_Go, a, b, c, d, w1_t, MD5C10, MD5S10);
+  MD5_STEP_S (MD5_Go, d, a, b, c, w6_t, MD5C11, MD5S11);
+  MD5_STEP_S (MD5_Go, c, d, a, b, wb_t, MD5C12, MD5S12);
+  MD5_STEP_S (MD5_Go, b, c, d, a, w0_t, MD5C13, MD5S13);
+  MD5_STEP_S (MD5_Go, a, b, c, d, w5_t, MD5C14, MD5S10);
+  MD5_STEP_S (MD5_Go, d, a, b, c, wa_t, MD5C15, MD5S11);
+  MD5_STEP_S (MD5_Go, c, d, a, b, wf_t, MD5C16, MD5S12);
+  MD5_STEP_S (MD5_Go, b, c, d, a, w4_t, MD5C17, MD5S13);
+  MD5_STEP_S (MD5_Go, a, b, c, d, w9_t, MD5C18, MD5S10);
+  MD5_STEP_S (MD5_Go, d, a, b, c, we_t, MD5C19, MD5S11);
+  MD5_STEP_S (MD5_Go, c, d, a, b, w3_t, MD5C1a, MD5S12);
+  MD5_STEP_S (MD5_Go, b, c, d, a, w8_t, MD5C1b, MD5S13);
+  MD5_STEP_S (MD5_Go, a, b, c, d, wd_t, MD5C1c, MD5S10);
+  MD5_STEP_S (MD5_Go, d, a, b, c, w2_t, MD5C1d, MD5S11);
+  MD5_STEP_S (MD5_Go, c, d, a, b, w7_t, MD5C1e, MD5S12);
+  MD5_STEP_S (MD5_Go, b, c, d, a, wc_t, MD5C1f, MD5S13);
+
+  MD5_STEP_S (MD5_H , a, b, c, d, w5_t, MD5C20, MD5S20);
+  MD5_STEP_S (MD5_H , d, a, b, c, w8_t, MD5C21, MD5S21);
+  MD5_STEP_S (MD5_H , c, d, a, b, wb_t, MD5C22, MD5S22);
+  MD5_STEP_S (MD5_H , b, c, d, a, we_t, MD5C23, MD5S23);
+  MD5_STEP_S (MD5_H , a, b, c, d, w1_t, MD5C24, MD5S20);
+  MD5_STEP_S (MD5_H , d, a, b, c, w4_t, MD5C25, MD5S21);
+  MD5_STEP_S (MD5_H , c, d, a, b, w7_t, MD5C26, MD5S22);
+  MD5_STEP_S (MD5_H , b, c, d, a, wa_t, MD5C27, MD5S23);
+  MD5_STEP_S (MD5_H , a, b, c, d, wd_t, MD5C28, MD5S20);
+  MD5_STEP_S (MD5_H , d, a, b, c, w0_t, MD5C29, MD5S21);
+  MD5_STEP_S (MD5_H , c, d, a, b, w3_t, MD5C2a, MD5S22);
+  MD5_STEP_S (MD5_H , b, c, d, a, w6_t, MD5C2b, MD5S23);
+  MD5_STEP_S (MD5_H , a, b, c, d, w9_t, MD5C2c, MD5S20);
+  MD5_STEP_S (MD5_H , d, a, b, c, wc_t, MD5C2d, MD5S21);
+  MD5_STEP_S (MD5_H , c, d, a, b, wf_t, MD5C2e, MD5S22);
+  MD5_STEP_S (MD5_H , b, c, d, a, w2_t, MD5C2f, MD5S23);
+
+  MD5_STEP_S (MD5_I , a, b, c, d, w0_t, MD5C30, MD5S30);
+  MD5_STEP_S (MD5_I , d, a, b, c, w7_t, MD5C31, MD5S31);
+  MD5_STEP_S (MD5_I , c, d, a, b, we_t, MD5C32, MD5S32);
+  MD5_STEP_S (MD5_I , b, c, d, a, w5_t, MD5C33, MD5S33);
+  MD5_STEP_S (MD5_I , a, b, c, d, wc_t, MD5C34, MD5S30);
+  MD5_STEP_S (MD5_I , d, a, b, c, w3_t, MD5C35, MD5S31);
+  MD5_STEP_S (MD5_I , c, d, a, b, wa_t, MD5C36, MD5S32);
+  MD5_STEP_S (MD5_I , b, c, d, a, w1_t, MD5C37, MD5S33);
+  MD5_STEP_S (MD5_I , a, b, c, d, w8_t, MD5C38, MD5S30);
+  MD5_STEP_S (MD5_I , d, a, b, c, wf_t, MD5C39, MD5S31);
+  MD5_STEP_S (MD5_I , c, d, a, b, w6_t, MD5C3a, MD5S32);
+  MD5_STEP_S (MD5_I , b, c, d, a, wd_t, MD5C3b, MD5S33);
+  MD5_STEP_S (MD5_I , a, b, c, d, w4_t, MD5C3c, MD5S30);
+  MD5_STEP_S (MD5_I , d, a, b, c, wb_t, MD5C3d, MD5S31);
+  MD5_STEP_S (MD5_I , c, d, a, b, w2_t, MD5C3e, MD5S32);
+  MD5_STEP_S (MD5_I , b, c, d, a, w9_t, MD5C3f, MD5S33);
+
+  digest[0] += a;
+  digest[1] += b;
+  digest[2] += c;
+  digest[3] += d;
+}
+
+void md5_transform_V (const u32x w0[4], const u32x w1[4], const u32x w2[4], const u32x w3[4], u32x digest[4])
+{
+  u32x a = digest[0];
+  u32x b = digest[1];
+  u32x c = digest[2];
+  u32x d = digest[3];
+
+  u32x w0_t = w0[0];
+  u32x w1_t = w0[1];
+  u32x w2_t = w0[2];
+  u32x w3_t = w0[3];
+  u32x w4_t = w1[0];
+  u32x w5_t = w1[1];
+  u32x w6_t = w1[2];
+  u32x w7_t = w1[3];
+  u32x w8_t = w2[0];
+  u32x w9_t = w2[1];
+  u32x wa_t = w2[2];
+  u32x wb_t = w2[3];
+  u32x wc_t = w3[0];
+  u32x wd_t = w3[1];
+  u32x we_t = w3[2];
+  u32x wf_t = 0;
 
   MD5_STEP (MD5_Fo, a, b, c, d, w0_t, MD5C00, MD5S00);
   MD5_STEP (MD5_Fo, d, a, b, c, w1_t, MD5C01, MD5S01);
@@ -80,22 +173,22 @@ static void md5_transform (const u32 w0[4], const u32 w1[4], const u32 w2[4], co
   MD5_STEP (MD5_Go, c, d, a, b, w7_t, MD5C1e, MD5S12);
   MD5_STEP (MD5_Go, b, c, d, a, wc_t, MD5C1f, MD5S13);
 
-  MD5_STEP (MD5_H1, a, b, c, d, w5_t, MD5C20, MD5S20);
-  MD5_STEP (MD5_H2, d, a, b, c, w8_t, MD5C21, MD5S21);
-  MD5_STEP (MD5_H1, c, d, a, b, wb_t, MD5C22, MD5S22);
-  MD5_STEP (MD5_H2, b, c, d, a, we_t, MD5C23, MD5S23);
-  MD5_STEP (MD5_H1, a, b, c, d, w1_t, MD5C24, MD5S20);
-  MD5_STEP (MD5_H2, d, a, b, c, w4_t, MD5C25, MD5S21);
-  MD5_STEP (MD5_H1, c, d, a, b, w7_t, MD5C26, MD5S22);
-  MD5_STEP (MD5_H2, b, c, d, a, wa_t, MD5C27, MD5S23);
-  MD5_STEP (MD5_H1, a, b, c, d, wd_t, MD5C28, MD5S20);
-  MD5_STEP (MD5_H2, d, a, b, c, w0_t, MD5C29, MD5S21);
-  MD5_STEP (MD5_H1, c, d, a, b, w3_t, MD5C2a, MD5S22);
-  MD5_STEP (MD5_H2, b, c, d, a, w6_t, MD5C2b, MD5S23);
-  MD5_STEP (MD5_H1, a, b, c, d, w9_t, MD5C2c, MD5S20);
-  MD5_STEP (MD5_H2, d, a, b, c, wc_t, MD5C2d, MD5S21);
-  MD5_STEP (MD5_H1, c, d, a, b, wf_t, MD5C2e, MD5S22);
-  MD5_STEP (MD5_H2, b, c, d, a, w2_t, MD5C2f, MD5S23);
+  MD5_STEP (MD5_H , a, b, c, d, w5_t, MD5C20, MD5S20);
+  MD5_STEP (MD5_H , d, a, b, c, w8_t, MD5C21, MD5S21);
+  MD5_STEP (MD5_H , c, d, a, b, wb_t, MD5C22, MD5S22);
+  MD5_STEP (MD5_H , b, c, d, a, we_t, MD5C23, MD5S23);
+  MD5_STEP (MD5_H , a, b, c, d, w1_t, MD5C24, MD5S20);
+  MD5_STEP (MD5_H , d, a, b, c, w4_t, MD5C25, MD5S21);
+  MD5_STEP (MD5_H , c, d, a, b, w7_t, MD5C26, MD5S22);
+  MD5_STEP (MD5_H , b, c, d, a, wa_t, MD5C27, MD5S23);
+  MD5_STEP (MD5_H , a, b, c, d, wd_t, MD5C28, MD5S20);
+  MD5_STEP (MD5_H , d, a, b, c, w0_t, MD5C29, MD5S21);
+  MD5_STEP (MD5_H , c, d, a, b, w3_t, MD5C2a, MD5S22);
+  MD5_STEP (MD5_H , b, c, d, a, w6_t, MD5C2b, MD5S23);
+  MD5_STEP (MD5_H , a, b, c, d, w9_t, MD5C2c, MD5S20);
+  MD5_STEP (MD5_H , d, a, b, c, wc_t, MD5C2d, MD5S21);
+  MD5_STEP (MD5_H , c, d, a, b, wf_t, MD5C2e, MD5S22);
+  MD5_STEP (MD5_H , b, c, d, a, w2_t, MD5C2f, MD5S23);
 
   MD5_STEP (MD5_I , a, b, c, d, w0_t, MD5C30, MD5S30);
   MD5_STEP (MD5_I , d, a, b, c, w7_t, MD5C31, MD5S31);
@@ -120,7 +213,7 @@ static void md5_transform (const u32 w0[4], const u32 w1[4], const u32 w2[4], co
   digest[3] += d;
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00400_init (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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 m00400_init (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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)
 {
   /**
    * base
@@ -196,7 +289,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00400_init (__gl
   block3[2] = block_len * 8;
   block3[3] = 0;
 
-  append_0x80_4x4 (block0, block1, block2, block3, block_len);
+  append_0x80_4x4_S (block0, block1, block2, block3, block_len);
 
   /**
    * init
@@ -209,7 +302,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00400_init (__gl
   digest[2] = MD5M_C;
   digest[3] = MD5M_D;
 
-  md5_transform (block0, block1, block2, block3, digest);
+  md5_transform_S (block0, block1, block2, block3, digest);
 
   tmps[gid].digest_buf[0] = digest[0];
   tmps[gid].digest_buf[1] = digest[1];
@@ -217,7 +310,7 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00400_init (__gl
   tmps[gid].digest_buf[3] = digest[3];
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00400_loop (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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 m00400_loop (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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)
 {
   /**
    * base
@@ -227,75 +320,61 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00400_loop (__gl
 
   if (gid >= gid_max) return;
 
-  u32 w0[4];
-
-  w0[0] = pws[gid].i[ 0];
-  w0[1] = pws[gid].i[ 1];
-  w0[2] = pws[gid].i[ 2];
-  w0[3] = pws[gid].i[ 3];
-
-  u32 w1[4];
-
-  w1[0] = pws[gid].i[ 4];
-  w1[1] = pws[gid].i[ 5];
-  w1[2] = pws[gid].i[ 6];
-  w1[3] = pws[gid].i[ 7];
-
-  u32 w2[4];
-
-  w2[0] = pws[gid].i[ 8];
-  w2[1] = pws[gid].i[ 9];
+  u32x w0[4];
+  u32x w1[4];
+  u32x w2[4];
+
+  w0[0] = packv (pws, i, gid, 0);
+  w0[1] = packv (pws, i, gid, 1);
+  w0[2] = packv (pws, i, gid, 2);
+  w0[3] = packv (pws, i, gid, 3);
+  w1[0] = packv (pws, i, gid, 4);
+  w1[1] = packv (pws, i, gid, 5);
+  w1[2] = packv (pws, i, gid, 6);
+  w1[3] = packv (pws, i, gid, 7);
+  w2[0] = packv (pws, i, gid, 8);
+  w2[1] = packv (pws, i, gid, 9);
   w2[2] = 0;
   w2[3] = 0;
 
-  const u32 pw_len = pws[gid].pw_len;
+  u32x pw_len = packvf (pws, pw_len, gid);
 
-  /**
-   * digest
-   */
+  u32x digest[4];
 
-  u32 digest[4];
-
-  digest[0] = tmps[gid].digest_buf[0];
-  digest[1] = tmps[gid].digest_buf[1];
-  digest[2] = tmps[gid].digest_buf[2];
-  digest[3] = tmps[gid].digest_buf[3];
+  digest[0] = packv (tmps, digest_buf, gid, 0);
+  digest[1] = packv (tmps, digest_buf, gid, 1);
+  digest[2] = packv (tmps, digest_buf, gid, 2);
+  digest[3] = packv (tmps, digest_buf, gid, 3);
 
   /**
    * loop
    */
 
-  u32 block_len = (16 + pw_len);
+  u32x block_len = (16 + pw_len);
 
-  u32 block0[4];
+  u32x block0[4];
+  u32x block1[4];
+  u32x block2[4];
+  u32x block3[4];
 
   block0[0] = 0;
   block0[1] = 0;
   block0[2] = 0;
   block0[3] = 0;
-
-  u32 block1[4];
-
   block1[0] = w0[0];
   block1[1] = w0[1];
   block1[2] = w0[2];
   block1[3] = w0[3];
-
-  u32 block2[4];
-
   block2[0] = w1[0];
   block2[1] = w1[1];
   block2[2] = w1[2];
   block2[3] = w1[3];
-
-  u32 block3[4];
-
   block3[0] = w2[0];
   block3[1] = w2[1];
   block3[2] = block_len * 8;
   block3[3] = 0;
 
-  append_0x80_4x4 (block0, block1, block2, block3, block_len);
+  append_0x80_4x4_VV (block0, block1, block2, block3, block_len);
 
   /**
    * init
@@ -313,27 +392,26 @@ __kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00400_loop (__gl
     digest[2] = MD5M_C;
     digest[3] = MD5M_D;
 
-    md5_transform (block0, block1, block2, block3, digest);
+    md5_transform_V (block0, block1, block2, block3, digest);
   }
 
-  tmps[gid].digest_buf[0] = digest[0];
-  tmps[gid].digest_buf[1] = digest[1];
-  tmps[gid].digest_buf[2] = digest[2];
-  tmps[gid].digest_buf[3] = digest[3];
+  unpackv (tmps, digest_buf, gid, 0, digest[0]);
+  unpackv (tmps, digest_buf, gid, 1, digest[1]);
+  unpackv (tmps, digest_buf, gid, 2, digest[2]);
+  unpackv (tmps, digest_buf, gid, 3, digest[3]);
 }
 
-__kernel void __attribute__((reqd_work_group_size (64, 1, 1))) m00400_comp (__global pw_t *pws, __global gpu_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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 m00400_comp (__global pw_t *pws, __global kernel_rule_t *rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global phpass_tmp_t *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
    */
 
   const u32 gid = get_global_id (0);
+  const u32 lid = get_local_id (0);
 
   if (gid >= gid_max) return;
 
-  const u32 lid = get_local_id (0);
-
   /**
    * digest
    */