Cleanup -m 77xx kernels to latest standard

OpenCL/m07700_a0.cl

@@ -5,6 +5,9 @@
 
 #define _SAPB_
 
+//too much register pressure
+//#define NEW_SIMD_CODE
+
 #include "include/constants.h"
 #include "include/kernel_vendor.h"
 
@@ -18,9 +21,7 @@
 #include "OpenCL/common.c"
 #include "include/rp_kernel.h"
 #include "OpenCL/rp.c"
-
+#include "OpenCL/simd.c"
-#define COMPARE_S "OpenCL/check_single_comp4.c"
-#define COMPARE_M "OpenCL/check_multi_comp4.c"
 
 #define GETCHAR(a,p)  (((a)[(p) / 4] >> (((p) & 3) * 8)) & 0xff)
 #define PUTCHAR(a,p,c) ((a)[(p) / 4] = (((a)[(p) / 4] & ~(0xff << (((p) & 3) * 8))) | ((c) << (((p) & 3) * 8))))
@@ -235,18 +236,16 @@ __kernel void m07700_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
   if (gid >= gid_max) return;
 
   u32 pw_buf0[4];
-
-  pw_buf0[0] = pws[gid].i[ 0];
-  pw_buf0[1] = pws[gid].i[ 1];
-  pw_buf0[2] = 0;
-  pw_buf0[3] = 0;
-
   u32 pw_buf1[4];
 
-  pw_buf1[0] = 0;
+  pw_buf0[0] = pws[gid].i[0];
-  pw_buf1[1] = 0;
+  pw_buf0[1] = pws[gid].i[1];
-  pw_buf1[2] = 0;
+  pw_buf0[2] = pws[gid].i[2];
-  pw_buf1[3] = 0;
+  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;
 
@@ -271,40 +270,21 @@ __kernel void m07700_m04 (__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)
   {
-    u32 w0[4];
+    u32x w0[4] = { 0 };
+    u32x w1[4] = { 0 };
+    u32x w2[4] = { 0 };
+    u32x w3[4] = { 0 };
 
-    w0[0] = pw_buf0[0];
+    const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
-    w0[1] = pw_buf0[1];
-    w0[2] = 0;
-    w0[3] = 0;
-
-    u32 w1[4];
-
-    w1[0] = 0;
-    w1[1] = 0;
-    w1[2] = 0;
-    w1[3] = 0;
-
-    u32 w2[4];
-
-    w2[0] = 0;
-    w2[1] = 0;
-    w2[2] = 0;
-    w2[3] = 0;
-
-    u32 w3[4];
-
-    w3[0] = 0;
-    w3[1] = 0;
-    w3[2] = 0;
-    w3[3] = 0;
-
-    const u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len);
 
     if (out_len > 8) continue; // otherwise it overflows in waldorf function
 
+    /**
+     * SAP
+     */
+
     w0[0] = sapb_trans (w0[0]);
     w0[1] = sapb_trans (w0[1]);
 
@@ -313,28 +293,22 @@ __kernel void m07700_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
      */
 
     u32 s0[4];
+    u32 s1[4];
+    u32 s2[4];
+    u32 s3[4];
 
     s0[0] = salt_buf0[0];
     s0[1] = salt_buf0[1];
     s0[2] = salt_buf0[2];
     s0[3] = 0;
-
-    u32 s1[4];
-
     s1[0] = 0;
     s1[1] = 0;
     s1[2] = 0;
     s1[3] = 0;
-
-    u32 s2[4];
-
     s2[0] = 0;
     s2[1] = 0;
     s2[2] = 0;
     s2[3] = 0;
-
-    u32 s3[4];
-
     s3[0] = 0;
     s3[1] = 0;
     s3[2] = 0;
@@ -533,13 +507,10 @@ __kernel void m07700_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
     a ^= c;
     b ^= d;
+    c  = 0;
+    d  = 0;
 
-    const u32 r0 = a;
+    COMPARE_M_SIMD (a, b, c, d);
-    const u32 r1 = b;
-    const u32 r2 = 0;
-    const u32 r3 = 0;
-
-    #include COMPARE_M
   }
 }
 
@@ -568,18 +539,16 @@ __kernel void m07700_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
   if (gid >= gid_max) return;
 
   u32 pw_buf0[4];
-
-  pw_buf0[0] = pws[gid].i[ 0];
-  pw_buf0[1] = pws[gid].i[ 1];
-  pw_buf0[2] = 0;
-  pw_buf0[3] = 0;
-
   u32 pw_buf1[4];
 
-  pw_buf1[0] = 0;
+  pw_buf0[0] = pws[gid].i[0];
-  pw_buf1[1] = 0;
+  pw_buf0[1] = pws[gid].i[1];
-  pw_buf1[2] = 0;
+  pw_buf0[2] = pws[gid].i[2];
-  pw_buf1[3] = 0;
+  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;
 
@@ -616,40 +585,21 @@ __kernel void m07700_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)
   {
-    u32 w0[4];
+    u32x w0[4] = { 0 };
+    u32x w1[4] = { 0 };
+    u32x w2[4] = { 0 };
+    u32x w3[4] = { 0 };
 
-    w0[0] = pw_buf0[0];
+    const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
-    w0[1] = pw_buf0[1];
-    w0[2] = 0;
-    w0[3] = 0;
-
-    u32 w1[4];
-
-    w1[0] = 0;
-    w1[1] = 0;
-    w1[2] = 0;
-    w1[3] = 0;
-
-    u32 w2[4];
-
-    w2[0] = 0;
-    w2[1] = 0;
-    w2[2] = 0;
-    w2[3] = 0;
-
-    u32 w3[4];
-
-    w3[0] = 0;
-    w3[1] = 0;
-    w3[2] = 0;
-    w3[3] = 0;
-
-    const u32 out_len = apply_rules (rules_buf[il_pos].cmds, w0, w1, pw_len);
 
     if (out_len > 8) continue; // otherwise it overflows in waldorf function
 
+    /**
+     * SAP
+     */
+
     w0[0] = sapb_trans (w0[0]);
     w0[1] = sapb_trans (w0[1]);
 
@@ -658,28 +608,22 @@ __kernel void m07700_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
      */
 
     u32 s0[4];
+    u32 s1[4];
+    u32 s2[4];
+    u32 s3[4];
 
     s0[0] = salt_buf0[0];
     s0[1] = salt_buf0[1];
     s0[2] = salt_buf0[2];
     s0[3] = 0;
-
-    u32 s1[4];
-
     s1[0] = 0;
     s1[1] = 0;
     s1[2] = 0;
     s1[3] = 0;
-
-    u32 s2[4];
-
     s2[0] = 0;
     s2[1] = 0;
     s2[2] = 0;
     s2[3] = 0;
-
-    u32 s3[4];
-
     s3[0] = 0;
     s3[1] = 0;
     s3[2] = 0;
@@ -878,13 +822,10 @@ __kernel void m07700_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
     a ^= c;
     b ^= d;
+    c  = 0;
+    d  = 0;
 
-    const u32 r0 = a;
+    COMPARE_S_SIMD (a, b, c, d);
-    const u32 r1 = b;
-    const u32 r2 = 0;
-    const u32 r3 = 0;
-
-    #include COMPARE_S
   }
 }
 

OpenCL/m07700_a1.cl

@@ -5,6 +5,9 @@
 
 #define _SAPB_
 
+//too much register pressure
+//#define NEW_SIMD_CODE
+
 #include "include/constants.h"
 #include "include/kernel_vendor.h"
 
@@ -16,9 +19,7 @@
 #include "include/kernel_functions.c"
 #include "OpenCL/types_ocl.c"
 #include "OpenCL/common.c"
-
+#include "OpenCL/simd.c"
-#define COMPARE_S "OpenCL/check_single_comp4.c"
-#define COMPARE_M "OpenCL/check_multi_comp4.c"
 
 #define GETCHAR(a,p)  (((a)[(p) / 4] >> (((p) & 3) * 8)) & 0xff)
 #define PUTCHAR(a,p,c) ((a)[(p) / 4] = (((a)[(p) / 4] & ~(0xff << (((p) & 3) * 8))) | ((c) << (((p) & 3) * 8))))
@@ -232,41 +233,20 @@ __kernel void m07700_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
   if (gid >= gid_max) return;
 
-  u32 wordl0[4];
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
 
-  wordl0[0] = pws[gid].i[ 0];
+  pw_buf0[0] = pws[gid].i[0];
-  wordl0[1] = pws[gid].i[ 1];
+  pw_buf0[1] = pws[gid].i[1];
-  wordl0[2] = 0;
+  pw_buf0[2] = pws[gid].i[2];
-  wordl0[3] = 0;
+  pw_buf0[3] = pws[gid].i[3];
-
+  pw_buf1[0] = pws[gid].i[4];
-  u32 wordl1[4];
+  pw_buf1[1] = pws[gid].i[5];
-
+  pw_buf1[2] = pws[gid].i[6];
-  wordl1[0] = 0;
+  pw_buf1[3] = pws[gid].i[7];
-  wordl1[1] = 0;
-  wordl1[2] = 0;
-  wordl1[3] = 0;
-
-  u32 wordl2[4];
-
-  wordl2[0] = 0;
-  wordl2[1] = 0;
-  wordl2[2] = 0;
-  wordl2[3] = 0;
-
-  u32 wordl3[4];
-
-  wordl3[0] = 0;
-  wordl3[1] = 0;
-  wordl3[2] = 0;
-  wordl3[3] = 0;
 
   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);
-  }
-
   /**
    * salt
    */
@@ -284,87 +264,92 @@ __kernel void m07700_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
   salt_buf0[1] = sapb_trans (salt_buf0[1]);
   salt_buf0[2] = sapb_trans (salt_buf0[2]);
 
-  /**
-   * digest
-   */
-
   /**
    * 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 u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
 
-    const u32 pw_len = pw_l_len + pw_r_len;
+    const u32x pw_len = pw_l_len + pw_r_len;
 
-    u32 wordr0[4];
+    /**
+     * concat password candidate
+     */
 
-    wordr0[0] = combs_buf[il_pos].i[0];
+    u32x wordl0[4] = { 0 };
-    wordr0[1] = combs_buf[il_pos].i[1];
+    u32x wordl1[4] = { 0 };
-    wordr0[2] = 0;
+    u32x wordl2[4] = { 0 };
-    wordr0[3] = 0;
+    u32x wordl3[4] = { 0 };
 
-    u32 wordr1[4];
+    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];
 
-    wordr1[0] = 0;
+    u32x wordr0[4] = { 0 };
-    wordr1[1] = 0;
+    u32x wordr1[4] = { 0 };
-    wordr1[2] = 0;
+    u32x wordr2[4] = { 0 };
-    wordr1[3] = 0;
+    u32x wordr3[4] = { 0 };
 
-    u32 wordr2[4];
+    wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
-
+    wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
-    wordr2[0] = 0;
+    wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
-    wordr2[1] = 0;
+    wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
-    wordr2[2] = 0;
+    wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
-    wordr2[3] = 0;
+    wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
-
+    wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
-    u32 wordr3[4];
+    wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
-
-    wordr3[0] = 0;
-    wordr3[1] = 0;
-    wordr3[2] = 0;
-    wordr3[3] = 0;
 
     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];
 
-    w0[0] = sapb_trans (wordl0[0] | wordr0[0]);
+    w0[0] = wordl0[0] | wordr0[0];
-    w0[1] = sapb_trans (wordl0[1] | wordr0[1]);
+    w0[1] = wordl0[1] | wordr0[1];
-    w0[2] = 0;
+
-    w0[3] = 0;
+    if (pw_len > 8) continue; // otherwise it overflows in waldorf function
+
+    /**
+     * SAP
+     */
+
+    w0[0] = sapb_trans (w0[0]);
+    w0[1] = sapb_trans (w0[1]);
 
     /**
      * append salt
      */
 
     u32 s0[4];
+    u32 s1[4];
+    u32 s2[4];
+    u32 s3[4];
 
     s0[0] = salt_buf0[0];
     s0[1] = salt_buf0[1];
     s0[2] = salt_buf0[2];
     s0[3] = 0;
-
-    u32 s1[4];
-
     s1[0] = 0;
     s1[1] = 0;
     s1[2] = 0;
     s1[3] = 0;
-
-    u32 s2[4];
-
     s2[0] = 0;
     s2[1] = 0;
     s2[2] = 0;
     s2[3] = 0;
-
-    u32 s3[4];
-
     s3[0] = 0;
     s3[1] = 0;
     s3[2] = 0;
@@ -563,13 +548,10 @@ __kernel void m07700_m04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
     a ^= c;
     b ^= d;
+    c  = 0;
+    d  = 0;
 
-    const u32 r0 = a;
+    COMPARE_M_SIMD (a, b, c, d);
-    const u32 r1 = b;
-    const u32 r2 = 0;
-    const u32 r3 = 0;
-
-    #include COMPARE_M
   }
 }
 
@@ -597,41 +579,20 @@ __kernel void m07700_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
   if (gid >= gid_max) return;
 
-  u32 wordl0[4];
+  u32 pw_buf0[4];
+  u32 pw_buf1[4];
 
-  wordl0[0] = pws[gid].i[ 0];
+  pw_buf0[0] = pws[gid].i[0];
-  wordl0[1] = pws[gid].i[ 1];
+  pw_buf0[1] = pws[gid].i[1];
-  wordl0[2] = 0;
+  pw_buf0[2] = pws[gid].i[2];
-  wordl0[3] = 0;
+  pw_buf0[3] = pws[gid].i[3];
-
+  pw_buf1[0] = pws[gid].i[4];
-  u32 wordl1[4];
+  pw_buf1[1] = pws[gid].i[5];
-
+  pw_buf1[2] = pws[gid].i[6];
-  wordl1[0] = 0;
+  pw_buf1[3] = pws[gid].i[7];
-  wordl1[1] = 0;
-  wordl1[2] = 0;
-  wordl1[3] = 0;
-
-  u32 wordl2[4];
-
-  wordl2[0] = 0;
-  wordl2[1] = 0;
-  wordl2[2] = 0;
-  wordl2[3] = 0;
-
-  u32 wordl3[4];
-
-  wordl3[0] = 0;
-  wordl3[1] = 0;
-  wordl3[2] = 0;
-  wordl3[3] = 0;
 
   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);
-  }
-
   /**
    * salt
    */
@@ -665,79 +626,88 @@ __kernel void m07700_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 u32x pw_r_len = pwlenx_create_combt (combs_buf, il_pos);
 
-    const u32 pw_len = pw_l_len + pw_r_len;
+    const u32x pw_len = pw_l_len + pw_r_len;
 
-    u32 wordr0[4];
+    /**
+     * concat password candidate
+     */
 
-    wordr0[0] = combs_buf[il_pos].i[0];
+    u32x wordl0[4] = { 0 };
-    wordr0[1] = combs_buf[il_pos].i[1];
+    u32x wordl1[4] = { 0 };
-    wordr0[2] = 0;
+    u32x wordl2[4] = { 0 };
-    wordr0[3] = 0;
+    u32x wordl3[4] = { 0 };
 
-    u32 wordr1[4];
+    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];
 
-    wordr1[0] = 0;
+    u32x wordr0[4] = { 0 };
-    wordr1[1] = 0;
+    u32x wordr1[4] = { 0 };
-    wordr1[2] = 0;
+    u32x wordr2[4] = { 0 };
-    wordr1[3] = 0;
+    u32x wordr3[4] = { 0 };
 
-    u32 wordr2[4];
+    wordr0[0] = ix_create_combt (combs_buf, il_pos, 0);
-
+    wordr0[1] = ix_create_combt (combs_buf, il_pos, 1);
-    wordr2[0] = 0;
+    wordr0[2] = ix_create_combt (combs_buf, il_pos, 2);
-    wordr2[1] = 0;
+    wordr0[3] = ix_create_combt (combs_buf, il_pos, 3);
-    wordr2[2] = 0;
+    wordr1[0] = ix_create_combt (combs_buf, il_pos, 4);
-    wordr2[3] = 0;
+    wordr1[1] = ix_create_combt (combs_buf, il_pos, 5);
-
+    wordr1[2] = ix_create_combt (combs_buf, il_pos, 6);
-    u32 wordr3[4];
+    wordr1[3] = ix_create_combt (combs_buf, il_pos, 7);
-
-    wordr3[0] = 0;
-    wordr3[1] = 0;
-    wordr3[2] = 0;
-    wordr3[3] = 0;
 
     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];
 
-    w0[0] = sapb_trans (wordl0[0] | wordr0[0]);
+    w0[0] = wordl0[0] | wordr0[0];
-    w0[1] = sapb_trans (wordl0[1] | wordr0[1]);
+    w0[1] = wordl0[1] | wordr0[1];
-    w0[2] = 0;
+
-    w0[3] = 0;
+    if (pw_len > 8) continue; // otherwise it overflows in waldorf function
+
+    /**
+     * SAP
+     */
+
+    w0[0] = sapb_trans (w0[0]);
+    w0[1] = sapb_trans (w0[1]);
 
     /**
      * append salt
      */
 
     u32 s0[4];
+    u32 s1[4];
+    u32 s2[4];
+    u32 s3[4];
 
     s0[0] = salt_buf0[0];
     s0[1] = salt_buf0[1];
     s0[2] = salt_buf0[2];
     s0[3] = 0;
-
-    u32 s1[4];
-
     s1[0] = 0;
     s1[1] = 0;
     s1[2] = 0;
     s1[3] = 0;
-
-    u32 s2[4];
-
     s2[0] = 0;
     s2[1] = 0;
     s2[2] = 0;
     s2[3] = 0;
-
-    u32 s3[4];
-
     s3[0] = 0;
     s3[1] = 0;
     s3[2] = 0;
@@ -936,13 +906,10 @@ __kernel void m07700_s04 (__global pw_t *pws, __global kernel_rule_t *rules_buf,
 
     a ^= c;
     b ^= d;
+    c  = 0;
+    d  = 0;
 
-    const u32 r0 = a;
+    COMPARE_S_SIMD (a, b, c, d);
-    const u32 r1 = b;
-    const u32 r2 = 0;
-    const u32 r3 = 0;
-
-    #include COMPARE_S
   }
 }
 

OpenCL/m07700_a3.cl

@@ -5,6 +5,9 @@
 
 #define _SAPB_
 
+//too much register pressure
+//#define NEW_SIMD_CODE
+
 #include "include/constants.h"
 #include "include/kernel_vendor.h"
 
@@ -16,9 +19,7 @@
 #include "include/kernel_functions.c"
 #include "OpenCL/types_ocl.c"
 #include "OpenCL/common.c"
-
+#include "OpenCL/simd.c"
-#define COMPARE_S "OpenCL/check_single_comp4.c"
-#define COMPARE_M "OpenCL/check_multi_comp4.c"
 
 #define GETCHAR(a,p)  (((a)[(p) / 4] >> (((p) & 3) * 8)) & 0xff)
 #define PUTCHAR(a,p,c) ((a)[(p) / 4] = (((a)[(p) / 4] & ~(0xff << (((p) & 3) * 8))) | ((c) << (((p) & 3) * 8))))
@@ -245,28 +246,22 @@ static void m07700m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
   const u32 salt_len = salt_bufs[salt_pos].salt_len;
 
   u32 s0[4];
+  u32 s1[4];
+  u32 s2[4];
+  u32 s3[4];
 
   s0[0] = salt_buf0[0];
   s0[1] = salt_buf0[1];
   s0[2] = salt_buf0[2];
   s0[3] = 0;
-
-  u32 s1[4];
-
   s1[0] = 0;
   s1[1] = 0;
   s1[2] = 0;
   s1[3] = 0;
-
-  u32 s2[4];
-
   s2[0] = 0;
   s2[1] = 0;
   s2[2] = 0;
   s2[3] = 0;
-
-  u32 s3[4];
-
   s3[0] = 0;
   s3[1] = 0;
   s3[2] = 0;
@@ -282,11 +277,13 @@ static void m07700m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
   u32 w0l = w0[0];
 
-  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 w0r = sapb_trans (bfs_buf[il_pos].i);
+    const u32x w0r = sapb_trans (ix_create_bft (bfs_buf, il_pos));
 
-    w0[0] = w0l | w0r;
+    const u32x w0lr = w0l | w0r;
+
+    w0[0] = w0lr;
 
     u32 t[16];
 
@@ -477,13 +474,10 @@ static void m07700m (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
     a ^= c;
     b ^= d;
+    c  = 0;
+    d  = 0;
 
-    const u32 r0 = a;
+    COMPARE_M_SIMD (a, b, c, d);
-    const u32 r1 = b;
-    const u32 r2 = 0;
-    const u32 r3 = 0;
-
-    #include COMPARE_M
   }
 }
 
@@ -516,28 +510,22 @@ static void m07700s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
   const u32 salt_len = salt_bufs[salt_pos].salt_len;
 
   u32 s0[4];
+  u32 s1[4];
+  u32 s2[4];
+  u32 s3[4];
 
   s0[0] = salt_buf0[0];
   s0[1] = salt_buf0[1];
   s0[2] = salt_buf0[2];
   s0[3] = 0;
-
-  u32 s1[4];
-
   s1[0] = 0;
   s1[1] = 0;
   s1[2] = 0;
   s1[3] = 0;
-
-  u32 s2[4];
-
   s2[0] = 0;
   s2[1] = 0;
   s2[2] = 0;
   s2[3] = 0;
-
-  u32 s3[4];
-
   s3[0] = 0;
   s3[1] = 0;
   s3[2] = 0;
@@ -565,11 +553,13 @@ static void m07700s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
   u32 w0l = w0[0];
 
-  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 w0r = sapb_trans (bfs_buf[il_pos].i);
+    const u32x w0r = sapb_trans (ix_create_bft (bfs_buf, il_pos));
 
-    w0[0] = w0l | w0r;
+    const u32x w0lr = w0l | w0r;
+
+    w0[0] = w0lr;
 
     u32 t[16];
 
@@ -760,13 +750,10 @@ static void m07700s (u32 w0[4], u32 w1[4], u32 w2[4], u32 w3[4], const u32 pw_le
 
     a ^= c;
     b ^= d;
+    c  = 0;
+    d  = 0;
 
-    const u32 r0 = a;
+    COMPARE_S_SIMD (a, b, c, d);
-    const u32 r1 = b;
-    const u32 r2 = 0;
-    const u32 r3 = 0;
-
-    #include COMPARE_S
   }
 }