Cleanup -m 69xx kernels to latest standard

diff --git a/OpenCL/m06900_a0.cl b/OpenCL/m06900_a0.cl
index 9d84e0f..82833bf 100644 (file)
--- a/OpenCL/m06900_a0.cl
+++ b/OpenCL/m06900_a0.cl
@@ -741,18 +741,16 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
    */
 
   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];
-
   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];
-  pw_buf1[3] = pws[gid].i[ 7];
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
 
   const u32 pw_len = pws[gid].pw_len;
 
@@ -769,8 +767,6 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
 
     const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
 
-    u32x w14 = out_len * 8;
-
     u32x data[8];
 
     data[0] = w0[0];
@@ -826,7 +822,7 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
 
     u32x tmp[8];
 
-    if (pw_len > 0)
+    //if (pw_len > 0) // not really SIMD compatible
     {
       PASS0 (state, tmp, state_m, data_m, s_tables);
       PASS2 (state, tmp, state_m, data_m, s_tables);
@@ -838,7 +834,7 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
       SHIFT61 (state, data_m);
     }
 
-    data[0] = w14;
+    data[0] = out_len * 8;
     data[1] = 0;
     data[2] = 0;
     data[3] = 0;
@@ -961,18 +957,16 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
    */
 
   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];
-
   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];
-  pw_buf1[3] = pws[gid].i[ 7];
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
 
   const u32 pw_len = pws[gid].pw_len;
 
@@ -1001,7 +995,9 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
 
     const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
 
-    u32x w14 = out_len * 8;
+    /**
+     * GOST
+     */
 
     u32x data[8];
 
@@ -1058,7 +1054,7 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
 
     u32x tmp[8];
 
-    if (pw_len > 0)
+    //if (pw_len > 0) // not really SIMD compatible
     {
       PASS0 (state, tmp, state_m, data_m, s_tables);
       PASS2 (state, tmp, state_m, data_m, s_tables);
@@ -1070,7 +1066,7 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *  rules_bu
       SHIFT61 (state, data_m);
     }
 
-    data[0] = w14;
+    data[0] = out_len * 8;
     data[1] = 0;
     data[2] = 0;
     data[3] = 0;

diff --git a/OpenCL/m06900_a1.cl b/OpenCL/m06900_a1.cl
index a51f258..f7a0765 100644 (file)
--- a/OpenCL/m06900_a1.cl
+++ b/OpenCL/m06900_a1.cl
@@ -7,6 +7,8 @@
 
 #define _GOST_
 
+#define NEW_SIMD_CODE
+
 #include "include/constants.h"
 #include "include/kernel_vendor.h"
 
@@ -18,9 +20,7 @@
 #include "include/kernel_functions.c"
 #include "OpenCL/types_ocl.c"
 #include "OpenCL/common.c"
-
-#define COMPARE_S "OpenCL/check_single_comp4.c"
-#define COMPARE_M "OpenCL/check_multi_comp4.c"
+#include "OpenCL/simd.c"
 
 __constant u32 c_tables[4][256] =
 {
@@ -290,11 +290,21 @@ __constant u32 c_tables[4][256] =
   }
 };
 
+#if   VECT_SIZE == 1
 #define BOX(i,n,S) (S)[(n)][(i)]
+#elif VECT_SIZE == 2
+#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1])
+#elif VECT_SIZE == 4
+#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3])
+#elif VECT_SIZE == 8
+#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7])
+#elif VECT_SIZE == 16
+#define BOX(i,n,S) (u32x) ((S)[(n)][(i).s0], (S)[(n)][(i).s1], (S)[(n)][(i).s2], (S)[(n)][(i).s3], (S)[(n)][(i).s4], (S)[(n)][(i).s5], (S)[(n)][(i).s6], (S)[(n)][(i).s7], (S)[(n)][(i).s8], (S)[(n)][(i).s9], (S)[(n)][(i).sa], (S)[(n)][(i).sb], (S)[(n)][(i).sc], (S)[(n)][(i).sd], (S)[(n)][(i).se], (S)[(n)][(i).sf])
+#endif
 
 #define _round(k1,k2,tbl)                 \
 {                                         \
-  u32 t;                                  \
+  u32x t;                                 \
   t = (k1) + r;                           \
   l ^= BOX (((t >>  0) & 0xff), 0, tbl) ^ \
        BOX (((t >>  8) & 0xff), 1, tbl) ^ \
@@ -309,8 +319,8 @@ __constant u32 c_tables[4][256] =
 
 #define R(k,h,s,i,t)      \
 {                         \
-  u32 r;                  \
-  u32 l;                  \
+  u32x r;                 \
+  u32x l;                 \
   r = h[i + 0];           \
   l = h[i + 1];           \
   _round (k[0], k[1], t);  \
@@ -379,8 +389,8 @@ __constant u32 c_tables[4][256] =
 
 #define A(x)        \
 {                   \
-  u32 l;          \
-  u32 r;          \
+  u32x l;           \
+  u32x r;           \
   l = x[0] ^ x[2];  \
   r = x[1] ^ x[3];  \
   x[0] = x[2];      \
@@ -395,8 +405,8 @@ __constant u32 c_tables[4][256] =
 
 #define AA(x)       \
 {                   \
-  u32 l;          \
-  u32 r;          \
+  u32x l;           \
+  u32x r;           \
   l    = x[0];      \
   r    = x[2];      \
   x[0] = x[4];      \
@@ -654,8 +664,8 @@ __constant u32 c_tables[4][256] =
 
 #define PASS0(h,s,u,v,t)  \
 {                         \
-  u32 k[8];             \
-  u32 w[8];             \
+  u32x k[8];              \
+  u32x w[8];              \
   X (w, u, v);            \
   P (k, w);               \
   R (k, h, s, 0, t);      \
@@ -665,8 +675,8 @@ __constant u32 c_tables[4][256] =
 
 #define PASS2(h,s,u,v,t)  \
 {                         \
-  u32 k[8];             \
-  u32 w[8];             \
+  u32x k[8];              \
+  u32x w[8];              \
   X (w, u, v);            \
   P (k, w);               \
   R (k, h, s, 2, t);      \
@@ -677,8 +687,8 @@ __constant u32 c_tables[4][256] =
 
 #define PASS4(h,s,u,v,t)  \
 {                         \
-  u32 k[8];             \
-  u32 w[8];             \
+  u32x k[8];              \
+  u32x w[8];              \
   X (w, u, v);            \
   P (k, w);               \
   R (k, h, s, 4, t);      \
@@ -688,8 +698,8 @@ __constant u32 c_tables[4][256] =
 
 #define PASS6(h,s,u,v,t)  \
 {                         \
-  u32 k[8];             \
-  u32 w[8];             \
+  u32x k[8];              \
+  u32x w[8];              \
   X (w, u, v);            \
   P (k, w);               \
   R (k, h, s, 6, t);      \
@@ -727,115 +737,88 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * base
    */
 
-  u32 wordl0[4];
-
-  wordl0[0] = pws[gid].i[ 0];
-  wordl0[1] = pws[gid].i[ 1];
-  wordl0[2] = pws[gid].i[ 2];
-  wordl0[3] = pws[gid].i[ 3];
-
-  u32 wordl1[4];
-
-  wordl1[0] = pws[gid].i[ 4];
-  wordl1[1] = pws[gid].i[ 5];
-  wordl1[2] = pws[gid].i[ 6];
-  wordl1[3] = pws[gid].i[ 7];
-
-  u32 wordl2[4];
-
-  wordl2[0] = 0;
-  wordl2[1] = 0;
-  wordl2[2] = 0;
-  wordl2[3] = 0;
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
 
-  u32 wordl3[4];
-
-  wordl3[0] = 0;
-  wordl3[1] = 0;
-  wordl3[2] = 0;
-  wordl3[3] = 0;
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
 
   const u32 pw_l_len = pws[gid].pw_len;
 
-  if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
-  {
-    switch_buffer_by_offset_le (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
-  }
-
   /**
    * loop
    */
 
-  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
   {
-    const u32 pw_r_len = combs_buf[il_pos].pw_len;
-
-    const u32 pw_len = pw_l_len + pw_r_len;
-
-    u32 wordr0[4];
-
-    wordr0[0] = combs_buf[il_pos].i[0];
-    wordr0[1] = combs_buf[il_pos].i[1];
-    wordr0[2] = combs_buf[il_pos].i[2];
-    wordr0[3] = combs_buf[il_pos].i[3];
-
-    u32 wordr1[4];
-
-    wordr1[0] = combs_buf[il_pos].i[4];
-    wordr1[1] = combs_buf[il_pos].i[5];
-    wordr1[2] = combs_buf[il_pos].i[6];
-    wordr1[3] = combs_buf[il_pos].i[7];
-
-    u32 wordr2[4];
-
-    wordr2[0] = 0;
-    wordr2[1] = 0;
-    wordr2[2] = 0;
-    wordr2[3] = 0;
-
-    u32 wordr3[4];
-
-    wordr3[0] = 0;
-    wordr3[1] = 0;
-    wordr3[2] = 0;
-    wordr3[3] = 0;
+    const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
+
+    const u32x pw_len = pw_l_len + pw_r_len;
+
+    /**
+     * concat password candidate
+     */
+
+    u32x wordl0[4] = { 0 };
+    u32x wordl1[4] = { 0 };
+    u32x wordl2[4] = { 0 };
+    u32x wordl3[4] = { 0 };
+
+    wordl0[0] = pw_buf0[0];
+    wordl0[1] = pw_buf0[1];
+    wordl0[2] = pw_buf0[2];
+    wordl0[3] = pw_buf0[3];
+    wordl1[0] = pw_buf1[0];
+    wordl1[1] = pw_buf1[1];
+    wordl1[2] = pw_buf1[2];
+    wordl1[3] = pw_buf1[3];
+
+    u32x wordr0[4] = { 0 };
+    u32x wordr1[4] = { 0 };
+    u32x wordr2[4] = { 0 };
+    u32x wordr3[4] = { 0 };
+
+    wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
+    wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
+    wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
+    wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
+    wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
+    wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
+    wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
+    wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
 
     if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
     {
-      switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len);
+      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];
+    u32x w0[4];
+    u32x 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];
-
-    u32 w1[4];
-
     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 w2[4];
-
-    w2[0] = wordl2[0] | wordr2[0];
-    w2[1] = wordl2[1] | wordr2[1];
-    w2[2] = wordl2[2] | wordr2[2];
-    w2[3] = wordl2[3] | wordr2[3];
-
-    u32 w3[4];
-
-    w3[0] = wordl3[0] | wordr3[0];
-    w3[1] = wordl3[1] | wordr3[1];
-    w3[2] = pw_len * 8;
-    w3[3] = 0;
-
-    const u32 w14 = pw_len * 8;
+    /**
+     * GOST
+     */
 
-    u32 data[8];
+    u32x data[8];
 
     data[0] = w0[0];
     data[1] = w0[1];
@@ -846,7 +829,7 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     data[6] = w1[2];
     data[7] = w1[3];
 
-    u32 state[16];
+    u32x state[16];
 
     state[ 0] = 0;
     state[ 1] = 0;
@@ -865,8 +848,8 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     state[14] = data[6];
     state[15] = data[7];
 
-    u32 state_m[8];
-    u32 data_m[8];
+    u32x state_m[8];
+    u32x data_m[8];
 
     /* gost1 */
 
@@ -888,9 +871,9 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     data_m[6] = data[6];
     data_m[7] = data[7];
 
-    u32 tmp[8];
+    u32x tmp[8];
 
-    if (pw_len > 0)
+    //if (pw_len > 0) // not really SIMD compatible
     {
       PASS0 (state, tmp, state_m, data_m, s_tables);
       PASS2 (state, tmp, state_m, data_m, s_tables);
@@ -902,7 +885,7 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
       SHIFT61 (state, data_m);
     }
 
-    data[0] = w14;
+    data[0] = pw_len * 8;
     data[1] = 0;
     data[2] = 0;
     data[3] = 0;
@@ -980,12 +963,7 @@ __kernel void m06900_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
     /* store */
 
-    const u32 r0 = state[0];
-    const u32 r1 = state[1];
-    const u32 r2 = state[2];
-    const u32 r3 = state[3];
-
-    #include COMPARE_M
+    COMPARE_M_SIMD (state[0], state[1], state[2], state[3]);
   }
 }
 
@@ -1029,41 +1007,20 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * base
    */
 
-  u32 wordl0[4];
-
-  wordl0[0] = pws[gid].i[ 0];
-  wordl0[1] = pws[gid].i[ 1];
-  wordl0[2] = pws[gid].i[ 2];
-  wordl0[3] = pws[gid].i[ 3];
-
-  u32 wordl1[4];
-
-  wordl1[0] = pws[gid].i[ 4];
-  wordl1[1] = pws[gid].i[ 5];
-  wordl1[2] = pws[gid].i[ 6];
-  wordl1[3] = pws[gid].i[ 7];
-
-  u32 wordl2[4];
-
-  wordl2[0] = 0;
-  wordl2[1] = 0;
-  wordl2[2] = 0;
-  wordl2[3] = 0;
-
-  u32 wordl3[4];
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
 
-  wordl3[0] = 0;
-  wordl3[1] = 0;
-  wordl3[2] = 0;
-  wordl3[3] = 0;
+  pw_buf0[0] = pws[gid].i[0];
+  pw_buf0[1] = pws[gid].i[1];
+  pw_buf0[2] = pws[gid].i[2];
+  pw_buf0[3] = pws[gid].i[3];
+  pw_buf1[0] = pws[gid].i[4];
+  pw_buf1[1] = pws[gid].i[5];
+  pw_buf1[2] = pws[gid].i[6];
+  pw_buf1[3] = pws[gid].i[7];
 
   const u32 pw_l_len = pws[gid].pw_len;
 
-  if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
-  {
-    switch_buffer_by_offset_le (wordl0, wordl1, wordl2, wordl3, combs_buf[0].pw_len);
-  }
-
   /**
    * digest
    */
@@ -1080,76 +1037,70 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
    * loop
    */
 
-  for (u32 il_pos = 0; il_pos < il_cnt; il_pos++)
+  for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
   {
-    const u32 pw_r_len = combs_buf[il_pos].pw_len;
-
-    const u32 pw_len = pw_l_len + pw_r_len;
-
-    u32 wordr0[4];
-
-    wordr0[0] = combs_buf[il_pos].i[0];
-    wordr0[1] = combs_buf[il_pos].i[1];
-    wordr0[2] = combs_buf[il_pos].i[2];
-    wordr0[3] = combs_buf[il_pos].i[3];
-
-    u32 wordr1[4];
-
-    wordr1[0] = combs_buf[il_pos].i[4];
-    wordr1[1] = combs_buf[il_pos].i[5];
-    wordr1[2] = combs_buf[il_pos].i[6];
-    wordr1[3] = combs_buf[il_pos].i[7];
-
-    u32 wordr2[4];
-
-    wordr2[0] = 0;
-    wordr2[1] = 0;
-    wordr2[2] = 0;
-    wordr2[3] = 0;
-
-    u32 wordr3[4];
-
-    wordr3[0] = 0;
-    wordr3[1] = 0;
-    wordr3[2] = 0;
-    wordr3[3] = 0;
+    const u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
+
+    const u32x pw_len = pw_l_len + pw_r_len;
+
+    /**
+     * concat password candidate
+     */
+
+    u32x wordl0[4] = { 0 };
+    u32x wordl1[4] = { 0 };
+    u32x wordl2[4] = { 0 };
+    u32x wordl3[4] = { 0 };
+
+    wordl0[0] = pw_buf0[0];
+    wordl0[1] = pw_buf0[1];
+    wordl0[2] = pw_buf0[2];
+    wordl0[3] = pw_buf0[3];
+    wordl1[0] = pw_buf1[0];
+    wordl1[1] = pw_buf1[1];
+    wordl1[2] = pw_buf1[2];
+    wordl1[3] = pw_buf1[3];
+
+    u32x wordr0[4] = { 0 };
+    u32x wordr1[4] = { 0 };
+    u32x wordr2[4] = { 0 };
+    u32x wordr3[4] = { 0 };
+
+    wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
+    wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
+    wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
+    wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
+    wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
+    wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
+    wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
+    wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
 
     if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
     {
-      switch_buffer_by_offset_le (wordr0, wordr1, wordr2, wordr3, pw_l_len);
+      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];
+    u32x w0[4];
+    u32x 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];
-
-    u32 w1[4];
-
     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 w2[4];
-
-    w2[0] = wordl2[0] | wordr2[0];
-    w2[1] = wordl2[1] | wordr2[1];
-    w2[2] = wordl2[2] | wordr2[2];
-    w2[3] = wordl2[3] | wordr2[3];
-
-    u32 w3[4];
+    /**
+     * GOST
+     */
 
-    w3[0] = wordl3[0] | wordr3[0];
-    w3[1] = wordl3[1] | wordr3[1];
-    w3[2] = pw_len * 8;
-    w3[3] = 0;
-
-    const u32 w14 = pw_len * 8;
-
-    u32 data[8];
+    u32x data[8];
 
     data[0] = w0[0];
     data[1] = w0[1];
@@ -1160,7 +1111,7 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     data[6] = w1[2];
     data[7] = w1[3];
 
-    u32 state[16];
+    u32x state[16];
 
     state[ 0] = 0;
     state[ 1] = 0;
@@ -1179,8 +1130,8 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     state[14] = data[6];
     state[15] = data[7];
 
-    u32 state_m[8];
-    u32 data_m[8];
+    u32x state_m[8];
+    u32x data_m[8];
 
     /* gost1 */
 
@@ -1202,9 +1153,9 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
     data_m[6] = data[6];
     data_m[7] = data[7];
 
-    u32 tmp[8];
+    u32x tmp[8];
 
-    if (pw_len > 0)
+    //if (pw_len > 0) // not really SIMD compatible
     {
       PASS0 (state, tmp, state_m, data_m, s_tables);
       PASS2 (state, tmp, state_m, data_m, s_tables);
@@ -1216,7 +1167,7 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
       SHIFT61 (state, data_m);
     }
 
-    data[0] = w14;
+    data[0] = pw_len * 8;
     data[1] = 0;
     data[2] = 0;
     data[3] = 0;
@@ -1294,12 +1245,7 @@ __kernel void m06900_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
     /* store */
 
-    const u32 r0 = state[0];
-    const u32 r1 = state[1];
-    const u32 r2 = state[2];
-    const u32 r3 = state[3];
-
-    #include COMPARE_S
+    COMPARE_S_SIMD (state[0], state[1], state[2], state[3]);
   }
 }
 

diff --git a/OpenCL/m06900_a3.cl b/OpenCL/m06900_a3.cl
index ac28d32..2453a3b 100644 (file)
--- a/OpenCL/m06900_a3.cl
+++ b/OpenCL/m06900_a3.cl
@@ -715,12 +715,6 @@ static void m06900m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
   const u32 gid = get_global_id (0);
   const u32 lid = get_local_id (0);
 
-  /**
-   * base
-   */
-
-  const u32 w14 = pw_len * 8;
-
   /**
    * loop
    */
@@ -733,6 +727,10 @@ static void m06900m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
     const u32x w0lr = w0l | w0r;
 
+    /**
+     * GOST
+     */
+
     u32x data[8];
 
     data[0] = w0lr;
@@ -788,7 +786,7 @@ static void m06900m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
     u32x tmp[8];
 
-    if (pw_len > 0)
+    //if (pw_len > 0) // not really SIMD compatible
     {
       PASS0 (state, tmp, state_m, data_m, s_tables);
       PASS2 (state, tmp, state_m, data_m, s_tables);
@@ -800,7 +798,7 @@ static void m06900m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
       SHIFT61 (state, data_m);
     }
 
-    data[0] = w14;
+    data[0] = pw_len * 8;
     data[1] = 0;
     data[2] = 0;
     data[3] = 0;
@@ -891,12 +889,6 @@ static void m06900s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
   const u32 gid = get_global_id (0);
   const u32 lid = get_local_id (0);
 
-  /**
-   * base
-   */
-
-  const u32 w14 = pw_len * 8;
-
   /**
    * digest
    */
@@ -921,6 +913,10 @@ static void m06900s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
     const u32x w0lr = w0l | w0r;
 
+    /**
+     * GOST
+     */
+
     u32x data[8];
 
     data[0] = w0lr;
@@ -976,7 +972,7 @@ static void m06900s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
     u32x tmp[8];
 
-    if (pw_len > 0)
+    //if (pw_len > 0) // not really SIMD compatible
     {
       PASS0 (state, tmp, state_m, data_m, s_tables);
       PASS2 (state, tmp, state_m, data_m, s_tables);
@@ -988,7 +984,7 @@ static void m06900s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
       SHIFT61 (state, data_m);
     }
 
-    data[0] = w14;
+    data[0] = pw_len * 8;
     data[1] = 0;
     data[2] = 0;
     data[3] = 0;
@@ -1066,7 +1062,7 @@ static void m06900s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
     /* store */
 
-    COMPARE_S_SIMD (state[0], state[1], state[2], state[3]);
+    COMPARE_M_SIMD (state[0], state[1], state[2], state[3]);
   }
 }