diff --git a/OpenCL/m08900-pure.cl b/OpenCL/m08900-pure.cl
index 9744618e0..b2f9a6f65 100644
--- a/OpenCL/m08900-pure.cl
+++ b/OpenCL/m08900-pure.cl
@@ -170,22 +170,18 @@ DECLSPEC void salsa_r (uint4 *TI)
     TT[idx_r2++] = R3;
   }
 
-  idx_r1 = 0;
-  idx_r2 = SCRYPT_R * 4;
+  idx_r2 = 0;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
   for (int i = 0; i < SCRYPT_R; i++)
   {
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
   }
 }
 
-DECLSPEC void scrypt_smix_init (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
+DECLSPEC void scrypt_smix_init (uint4 *X, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
 {
   #define Coord(xd4,y,z) (((xd4) * ySIZE * zSIZE) + ((y) * zSIZE) + (z))
   #define CO Coord(xd4,y,z)
@@ -208,55 +204,15 @@ DECLSPEC void scrypt_smix_init (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
     case 3: V = V3; break;
   }
 
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
-
   for (u32 y = 0; y < ySIZE; y++)
   {
     for (u32 z = 0; z < zSIZE; z++) V[CO] = X[z];
 
     for (u32 i = 0; i < SCRYPT_TMTO; i++) salsa_r (X);
   }
-
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
 }
 
-DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
+DECLSPEC void scrypt_smix_loop (uint4 *X, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
 {
   #define Coord(xd4,y,z) (((xd4) * ySIZE * zSIZE) + ((y) * zSIZE) + (z))
   #define CO Coord(xd4,y,z)
@@ -279,26 +235,6 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
     case 3: V = V3; break;
   }
 
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
-
   for (u32 N_pos = 0; N_pos < 1024; N_pos++)
   {
     const u32 k = X[zSIZE - 4].x & (SCRYPT_N - 1);
@@ -307,6 +243,8 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
 
     const u32 km = k - (y * SCRYPT_TMTO);
 
+    uint4 T[STATE_CNT4];
+
     for (u32 z = 0; z < zSIZE; z++) T[z] = V[CO];
 
     for (u32 i = 0; i < km; i++) salsa_r (T);
@@ -315,26 +253,6 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
 
     salsa_r (X);
   }
-
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
 }
 
 KERNEL_FQ void m08900_init (KERN_ATTR_TMPS (scrypt_tmp_t))
@@ -405,6 +323,40 @@ KERNEL_FQ void m08900_init (KERN_ATTR_TMPS (scrypt_tmp_t))
     tmps[gid].P[k + 0] = tmp0;
     tmps[gid].P[k + 1] = tmp1;
   }
+
+  for (u32 l = 0; l < SCRYPT_CNT4; l += 4)
+  {
+    uint4 T[4];
+
+    T[0] = tmps[gid].P[l + 0];
+    T[1] = tmps[gid].P[l + 1];
+    T[2] = tmps[gid].P[l + 2];
+    T[3] = tmps[gid].P[l + 3];
+
+    T[0] = hc_swap32_4 (T[0]);
+    T[1] = hc_swap32_4 (T[1]);
+    T[2] = hc_swap32_4 (T[2]);
+    T[3] = hc_swap32_4 (T[3]);
+
+    uint4 X[4];
+
+    #ifdef IS_CUDA
+    X[0] = make_uint4 (T[0].x, T[1].y, T[2].z, T[3].w);
+    X[1] = make_uint4 (T[1].x, T[2].y, T[3].z, T[0].w);
+    X[2] = make_uint4 (T[2].x, T[3].y, T[0].z, T[1].w);
+    X[3] = make_uint4 (T[3].x, T[0].y, T[1].z, T[2].w);
+    #else
+    X[0] = (uint4) (T[0].x, T[1].y, T[2].z, T[3].w);
+    X[1] = (uint4) (T[1].x, T[2].y, T[3].z, T[0].w);
+    X[2] = (uint4) (T[2].x, T[3].y, T[0].z, T[1].w);
+    X[3] = (uint4) (T[3].x, T[0].y, T[1].z, T[2].w);
+    #endif
+
+    tmps[gid].P[l + 0] = X[0];
+    tmps[gid].P[l + 1] = X[1];
+    tmps[gid].P[l + 2] = X[2];
+    tmps[gid].P[l + 3] = X[3];
+  }
 }
 
 KERNEL_FQ void m08900_loop_prepare (KERN_ATTR_TMPS (scrypt_tmp_t))
@@ -414,6 +366,7 @@ KERNEL_FQ void m08900_loop_prepare (KERN_ATTR_TMPS (scrypt_tmp_t))
    */
 
   const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
 
   if (gid >= gid_max) return;
 
@@ -425,26 +378,20 @@ KERNEL_FQ void m08900_loop_prepare (KERN_ATTR_TMPS (scrypt_tmp_t))
   GLOBAL_AS uint4 *d_scrypt3_buf = (GLOBAL_AS uint4 *) d_extra3_buf;
 
   uint4 X[STATE_CNT4];
-  uint4 T[STATE_CNT4];
 
   const u32 P_offset = salt_repeat * STATE_CNT4;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) X[z] = hc_swap32_4 (tmps[gid].P[P_offset + z]);
+  for (int z = 0; z < STATE_CNT4; z++) X[z] = tmps[gid].P[P_offset + z];
 
-  scrypt_smix_init (X, T, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
+  scrypt_smix_init (X, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = hc_swap32_4 (X[z]);
+  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = X[z];
 }
 
 KERNEL_FQ void m08900_loop (KERN_ATTR_TMPS (scrypt_tmp_t))
 {
   const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
 
   if (gid >= gid_max) return;
 
@@ -454,21 +401,14 @@ KERNEL_FQ void m08900_loop (KERN_ATTR_TMPS (scrypt_tmp_t))
   GLOBAL_AS uint4 *d_scrypt3_buf = (GLOBAL_AS uint4 *) d_extra3_buf;
 
   uint4 X[STATE_CNT4];
-  uint4 T[STATE_CNT4];
 
   const u32 P_offset = salt_repeat * STATE_CNT4;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) X[z] = hc_swap32_4 (tmps[gid].P[P_offset + z]);
+  for (int z = 0; z < STATE_CNT4; z++) X[z] = tmps[gid].P[P_offset + z];
 
-  scrypt_smix_loop (X, T, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
+  scrypt_smix_loop (X, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = hc_swap32_4 (X[z]);
+  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = X[z];
 }
 
 KERNEL_FQ void m08900_comp (KERN_ATTR_TMPS (scrypt_tmp_t))
@@ -497,35 +437,48 @@ KERNEL_FQ void m08900_comp (KERN_ATTR_TMPS (scrypt_tmp_t))
 
   for (u32 l = 0; l < SCRYPT_CNT4; l += 4)
   {
-    uint4 tmp;
-
-    tmp = tmps[gid].P[l + 0];
+    uint4 X[4];
 
-    w0[0] = tmp.x;
-    w0[1] = tmp.y;
-    w0[2] = tmp.z;
-    w0[3] = tmp.w;
+    X[0] = tmps[gid].P[l + 0];
+    X[1] = tmps[gid].P[l + 1];
+    X[2] = tmps[gid].P[l + 2];
+    X[3] = tmps[gid].P[l + 3];
 
-    tmp = tmps[gid].P[l + 1];
+    uint4 T[4];
 
-    w1[0] = tmp.x;
-    w1[1] = tmp.y;
-    w1[2] = tmp.z;
-    w1[3] = tmp.w;
-
-    tmp = tmps[gid].P[l + 2];
-
-    w2[0] = tmp.x;
-    w2[1] = tmp.y;
-    w2[2] = tmp.z;
-    w2[3] = tmp.w;
-
-    tmp = tmps[gid].P[l + 3];
+    #ifdef IS_CUDA
+    T[0] = make_uint4 (X[0].x, X[3].y, X[2].z, X[1].w);
+    T[1] = make_uint4 (X[1].x, X[0].y, X[3].z, X[2].w);
+    T[2] = make_uint4 (X[2].x, X[1].y, X[0].z, X[3].w);
+    T[3] = make_uint4 (X[3].x, X[2].y, X[1].z, X[0].w);
+    #else
+    T[0] = (uint4) (X[0].x, X[3].y, X[2].z, X[1].w);
+    T[1] = (uint4) (X[1].x, X[0].y, X[3].z, X[2].w);
+    T[2] = (uint4) (X[2].x, X[1].y, X[0].z, X[3].w);
+    T[3] = (uint4) (X[3].x, X[2].y, X[1].z, X[0].w);
+    #endif
 
-    w3[0] = tmp.x;
-    w3[1] = tmp.y;
-    w3[2] = tmp.z;
-    w3[3] = tmp.w;
+    T[0] = hc_swap32_4 (T[0]);
+    T[1] = hc_swap32_4 (T[1]);
+    T[2] = hc_swap32_4 (T[2]);
+    T[3] = hc_swap32_4 (T[3]);
+
+    w0[0] = T[0].x;
+    w0[1] = T[0].y;
+    w0[2] = T[0].z;
+    w0[3] = T[0].w;
+    w1[0] = T[1].x;
+    w1[1] = T[1].y;
+    w1[2] = T[1].z;
+    w1[3] = T[1].w;
+    w2[0] = T[2].x;
+    w2[1] = T[2].y;
+    w2[2] = T[2].z;
+    w2[3] = T[2].w;
+    w3[0] = T[3].x;
+    w3[1] = T[3].y;
+    w3[2] = T[3].z;
+    w3[3] = T[3].w;
 
     sha256_hmac_update_64 (&ctx, w0, w1, w2, w3, 64);
   }
diff --git a/OpenCL/m15700-pure.cl b/OpenCL/m15700-pure.cl
index dfb09edd4..373316024 100644
--- a/OpenCL/m15700-pure.cl
+++ b/OpenCL/m15700-pure.cl
@@ -177,22 +177,18 @@ DECLSPEC void salsa_r (uint4 *TI)
     TT[idx_r2++] = R3;
   }
 
-  idx_r1 = 0;
-  idx_r2 = SCRYPT_R * 4;
+  idx_r2 = 0;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
   for (int i = 0; i < SCRYPT_R; i++)
   {
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
   }
 }
 
-DECLSPEC void scrypt_smix_init (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
+DECLSPEC void scrypt_smix_init (uint4 *X, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
 {
   #define Coord(xd4,y,z) (((xd4) * ySIZE * zSIZE) + ((y) * zSIZE) + (z))
   #define CO Coord(xd4,y,z)
@@ -215,55 +211,15 @@ DECLSPEC void scrypt_smix_init (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
     case 3: V = V3; break;
   }
 
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
-
   for (u32 y = 0; y < ySIZE; y++)
   {
     for (u32 z = 0; z < zSIZE; z++) V[CO] = X[z];
 
     for (u32 i = 0; i < SCRYPT_TMTO; i++) salsa_r (X);
   }
-
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
 }
 
-DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
+DECLSPEC void scrypt_smix_loop (uint4 *X, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
 {
   #define Coord(xd4,y,z) (((xd4) * ySIZE * zSIZE) + ((y) * zSIZE) + (z))
   #define CO Coord(xd4,y,z)
@@ -286,26 +242,6 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
     case 3: V = V3; break;
   }
 
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
-
   for (u32 N_pos = 0; N_pos < 1024; N_pos++)
   {
     const u32 k = X[zSIZE - 4].x & (SCRYPT_N - 1);
@@ -314,6 +250,8 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
 
     const u32 km = k - (y * SCRYPT_TMTO);
 
+    uint4 T[STATE_CNT4];
+
     for (u32 z = 0; z < zSIZE; z++) T[z] = V[CO];
 
     for (u32 i = 0; i < km; i++) salsa_r (T);
@@ -322,26 +260,6 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
 
     salsa_r (X);
   }
-
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
 }
 
 #ifndef KECCAK_ROUNDS
@@ -541,15 +459,50 @@ KERNEL_FQ void m15700_init (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_
     tmps[gid].P[k + 0] = tmp0;
     tmps[gid].P[k + 1] = tmp1;
   }
+
+  for (u32 l = 0; l < SCRYPT_CNT4; l += 4)
+  {
+    uint4 T[4];
+
+    T[0] = tmps[gid].P[l + 0];
+    T[1] = tmps[gid].P[l + 1];
+    T[2] = tmps[gid].P[l + 2];
+    T[3] = tmps[gid].P[l + 3];
+
+    T[0] = hc_swap32_4 (T[0]);
+    T[1] = hc_swap32_4 (T[1]);
+    T[2] = hc_swap32_4 (T[2]);
+    T[3] = hc_swap32_4 (T[3]);
+
+    uint4 X[4];
+
+    #ifdef IS_CUDA
+    X[0] = make_uint4 (T[0].x, T[1].y, T[2].z, T[3].w);
+    X[1] = make_uint4 (T[1].x, T[2].y, T[3].z, T[0].w);
+    X[2] = make_uint4 (T[2].x, T[3].y, T[0].z, T[1].w);
+    X[3] = make_uint4 (T[3].x, T[0].y, T[1].z, T[2].w);
+    #else
+    X[0] = (uint4) (T[0].x, T[1].y, T[2].z, T[3].w);
+    X[1] = (uint4) (T[1].x, T[2].y, T[3].z, T[0].w);
+    X[2] = (uint4) (T[2].x, T[3].y, T[0].z, T[1].w);
+    X[3] = (uint4) (T[3].x, T[0].y, T[1].z, T[2].w);
+    #endif
+
+    tmps[gid].P[l + 0] = X[0];
+    tmps[gid].P[l + 1] = X[1];
+    tmps[gid].P[l + 2] = X[2];
+    tmps[gid].P[l + 3] = X[3];
+  }
 }
 
-KERNEL_FQ void m15700_loop_prepare (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_t))
+KERNEL_FQ void m15700_loop_prepare (KERN_ATTR_TMPS (scrypt_tmp_t))
 {
   /**
    * base
    */
 
   const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
 
   if (gid >= gid_max) return;
 
@@ -561,26 +514,20 @@ KERNEL_FQ void m15700_loop_prepare (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum
   GLOBAL_AS uint4 *d_scrypt3_buf = (GLOBAL_AS uint4 *) d_extra3_buf;
 
   uint4 X[STATE_CNT4];
-  uint4 T[STATE_CNT4];
 
   const u32 P_offset = salt_repeat * STATE_CNT4;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) X[z] = hc_swap32_4 (tmps[gid].P[P_offset + z]);
+  for (int z = 0; z < STATE_CNT4; z++) X[z] = tmps[gid].P[P_offset + z];
 
-  scrypt_smix_init (X, T, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
+  scrypt_smix_init (X, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = hc_swap32_4 (X[z]);
+  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = X[z];
 }
 
-KERNEL_FQ void m15700_loop (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_t))
+KERNEL_FQ void m15700_loop (KERN_ATTR_TMPS (scrypt_tmp_t))
 {
   const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
 
   if (gid >= gid_max) return;
 
@@ -590,21 +537,14 @@ KERNEL_FQ void m15700_loop (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_
   GLOBAL_AS uint4 *d_scrypt3_buf = (GLOBAL_AS uint4 *) d_extra3_buf;
 
   uint4 X[STATE_CNT4];
-  uint4 T[STATE_CNT4];
 
   const u32 P_offset = salt_repeat * STATE_CNT4;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) X[z] = hc_swap32_4 (tmps[gid].P[P_offset + z]);
+  for (int z = 0; z < STATE_CNT4; z++) X[z] = tmps[gid].P[P_offset + z];
 
-  scrypt_smix_loop (X, T, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
+  scrypt_smix_loop (X, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = hc_swap32_4 (X[z]);
+  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = X[z];
 }
 
 KERNEL_FQ void m15700_comp (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_t))
@@ -633,35 +573,48 @@ KERNEL_FQ void m15700_comp (KERN_ATTR_TMPS_ESALT (scrypt_tmp_t, ethereum_scrypt_
 
   for (u32 l = 0; l < SCRYPT_CNT4; l += 4)
   {
-    uint4 tmp;
-
-    tmp = tmps[gid].P[l + 0];
+    uint4 X[4];
 
-    w0[0] = tmp.x;
-    w0[1] = tmp.y;
-    w0[2] = tmp.z;
-    w0[3] = tmp.w;
+    X[0] = tmps[gid].P[l + 0];
+    X[1] = tmps[gid].P[l + 1];
+    X[2] = tmps[gid].P[l + 2];
+    X[3] = tmps[gid].P[l + 3];
 
-    tmp = tmps[gid].P[l + 1];
+    uint4 T[4];
 
-    w1[0] = tmp.x;
-    w1[1] = tmp.y;
-    w1[2] = tmp.z;
-    w1[3] = tmp.w;
-
-    tmp = tmps[gid].P[l + 2];
-
-    w2[0] = tmp.x;
-    w2[1] = tmp.y;
-    w2[2] = tmp.z;
-    w2[3] = tmp.w;
-
-    tmp = tmps[gid].P[l + 3];
+    #ifdef IS_CUDA
+    T[0] = make_uint4 (X[0].x, X[3].y, X[2].z, X[1].w);
+    T[1] = make_uint4 (X[1].x, X[0].y, X[3].z, X[2].w);
+    T[2] = make_uint4 (X[2].x, X[1].y, X[0].z, X[3].w);
+    T[3] = make_uint4 (X[3].x, X[2].y, X[1].z, X[0].w);
+    #else
+    T[0] = (uint4) (X[0].x, X[3].y, X[2].z, X[1].w);
+    T[1] = (uint4) (X[1].x, X[0].y, X[3].z, X[2].w);
+    T[2] = (uint4) (X[2].x, X[1].y, X[0].z, X[3].w);
+    T[3] = (uint4) (X[3].x, X[2].y, X[1].z, X[0].w);
+    #endif
 
-    w3[0] = tmp.x;
-    w3[1] = tmp.y;
-    w3[2] = tmp.z;
-    w3[3] = tmp.w;
+    T[0] = hc_swap32_4 (T[0]);
+    T[1] = hc_swap32_4 (T[1]);
+    T[2] = hc_swap32_4 (T[2]);
+    T[3] = hc_swap32_4 (T[3]);
+
+    w0[0] = T[0].x;
+    w0[1] = T[0].y;
+    w0[2] = T[0].z;
+    w0[3] = T[0].w;
+    w1[0] = T[1].x;
+    w1[1] = T[1].y;
+    w1[2] = T[1].z;
+    w1[3] = T[1].w;
+    w2[0] = T[2].x;
+    w2[1] = T[2].y;
+    w2[2] = T[2].z;
+    w2[3] = T[2].w;
+    w3[0] = T[3].x;
+    w3[1] = T[3].y;
+    w3[2] = T[3].z;
+    w3[3] = T[3].w;
 
     sha256_hmac_update_64 (&ctx, w0, w1, w2, w3, 64);
   }
diff --git a/OpenCL/m22700-pure.cl b/OpenCL/m22700-pure.cl
index 66e1285d9..dfdbbe041 100644
--- a/OpenCL/m22700-pure.cl
+++ b/OpenCL/m22700-pure.cl
@@ -218,22 +218,18 @@ DECLSPEC void salsa_r (uint4 *TI)
     TT[idx_r2++] = R3;
   }
 
-  idx_r1 = 0;
-  idx_r2 = SCRYPT_R * 4;
+  idx_r2 = 0;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
   for (int i = 0; i < SCRYPT_R; i++)
   {
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
-    TI[idx_r2++] = TT[idx_r1++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
+    TI[idx_r1++] = TT[idx_r2++];
   }
 }
 
-DECLSPEC void scrypt_smix_init (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
+DECLSPEC void scrypt_smix_init (uint4 *X, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
 {
   #define Coord(xd4,y,z) (((xd4) * ySIZE * zSIZE) + ((y) * zSIZE) + (z))
   #define CO Coord(xd4,y,z)
@@ -256,55 +252,15 @@ DECLSPEC void scrypt_smix_init (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
     case 3: V = V3; break;
   }
 
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
-
   for (u32 y = 0; y < ySIZE; y++)
   {
     for (u32 z = 0; z < zSIZE; z++) V[CO] = X[z];
 
     for (u32 i = 0; i < SCRYPT_TMTO; i++) salsa_r (X);
   }
-
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
 }
 
-DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
+DECLSPEC void scrypt_smix_loop (uint4 *X, GLOBAL_AS uint4 *V0, GLOBAL_AS uint4 *V1, GLOBAL_AS uint4 *V2, GLOBAL_AS uint4 *V3)
 {
   #define Coord(xd4,y,z) (((xd4) * ySIZE * zSIZE) + ((y) * zSIZE) + (z))
   #define CO Coord(xd4,y,z)
@@ -327,26 +283,6 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
     case 3: V = V3; break;
   }
 
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 1].y, X[i + 2].z, X[i + 3].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 2].y, X[i + 3].z, X[i + 0].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 3].y, X[i + 0].z, X[i + 1].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 0].y, X[i + 1].z, X[i + 2].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
-
   for (u32 N_pos = 0; N_pos < 1024; N_pos++)
   {
     const u32 k = X[zSIZE - 4].x & (SCRYPT_N - 1);
@@ -355,6 +291,8 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
 
     const u32 km = k - (y * SCRYPT_TMTO);
 
+    uint4 T[STATE_CNT4];
+
     for (u32 z = 0; z < zSIZE; z++) T[z] = V[CO];
 
     for (u32 i = 0; i < km; i++) salsa_r (T);
@@ -363,26 +301,6 @@ DECLSPEC void scrypt_smix_loop (uint4 *X, uint4 *T, GLOBAL_AS uint4 *V0, GLOBAL_
 
     salsa_r (X);
   }
-
-  for (u32 i = 0; i < STATE_CNT4; i += 4)
-  {
-    #ifdef IS_CUDA
-    T[0] = make_uint4 (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = make_uint4 (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = make_uint4 (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = make_uint4 (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #else
-    T[0] = (uint4) (X[i + 0].x, X[i + 3].y, X[i + 2].z, X[i + 1].w);
-    T[1] = (uint4) (X[i + 1].x, X[i + 0].y, X[i + 3].z, X[i + 2].w);
-    T[2] = (uint4) (X[i + 2].x, X[i + 1].y, X[i + 0].z, X[i + 3].w);
-    T[3] = (uint4) (X[i + 3].x, X[i + 2].y, X[i + 1].z, X[i + 0].w);
-    #endif
-
-    X[i + 0] = T[0];
-    X[i + 1] = T[1];
-    X[i + 2] = T[2];
-    X[i + 3] = T[3];
-  }
 }
 
 KERNEL_FQ void m22700_init (KERN_ATTR_TMPS (scrypt_tmp_t))
@@ -493,6 +411,40 @@ KERNEL_FQ void m22700_init (KERN_ATTR_TMPS (scrypt_tmp_t))
     tmps[gid].P[k + 0] = tmp0;
     tmps[gid].P[k + 1] = tmp1;
   }
+
+  for (u32 l = 0; l < SCRYPT_CNT4; l += 4)
+  {
+    uint4 T[4];
+
+    T[0] = tmps[gid].P[l + 0];
+    T[1] = tmps[gid].P[l + 1];
+    T[2] = tmps[gid].P[l + 2];
+    T[3] = tmps[gid].P[l + 3];
+
+    T[0] = hc_swap32_4 (T[0]);
+    T[1] = hc_swap32_4 (T[1]);
+    T[2] = hc_swap32_4 (T[2]);
+    T[3] = hc_swap32_4 (T[3]);
+
+    uint4 X[4];
+
+    #ifdef IS_CUDA
+    X[0] = make_uint4 (T[0].x, T[1].y, T[2].z, T[3].w);
+    X[1] = make_uint4 (T[1].x, T[2].y, T[3].z, T[0].w);
+    X[2] = make_uint4 (T[2].x, T[3].y, T[0].z, T[1].w);
+    X[3] = make_uint4 (T[3].x, T[0].y, T[1].z, T[2].w);
+    #else
+    X[0] = (uint4) (T[0].x, T[1].y, T[2].z, T[3].w);
+    X[1] = (uint4) (T[1].x, T[2].y, T[3].z, T[0].w);
+    X[2] = (uint4) (T[2].x, T[3].y, T[0].z, T[1].w);
+    X[3] = (uint4) (T[3].x, T[0].y, T[1].z, T[2].w);
+    #endif
+
+    tmps[gid].P[l + 0] = X[0];
+    tmps[gid].P[l + 1] = X[1];
+    tmps[gid].P[l + 2] = X[2];
+    tmps[gid].P[l + 3] = X[3];
+  }
 }
 
 KERNEL_FQ void m22700_loop_prepare (KERN_ATTR_TMPS (scrypt_tmp_t))
@@ -502,6 +454,7 @@ KERNEL_FQ void m22700_loop_prepare (KERN_ATTR_TMPS (scrypt_tmp_t))
    */
 
   const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
 
   if (gid >= gid_max) return;
 
@@ -513,26 +466,20 @@ KERNEL_FQ void m22700_loop_prepare (KERN_ATTR_TMPS (scrypt_tmp_t))
   GLOBAL_AS uint4 *d_scrypt3_buf = (GLOBAL_AS uint4 *) d_extra3_buf;
 
   uint4 X[STATE_CNT4];
-  uint4 T[STATE_CNT4];
 
   const u32 P_offset = salt_repeat * STATE_CNT4;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) X[z] = hc_swap32_4 (tmps[gid].P[P_offset + z]);
+  for (int z = 0; z < STATE_CNT4; z++) X[z] = tmps[gid].P[P_offset + z];
 
-  scrypt_smix_init (X, T, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
+  scrypt_smix_init (X, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = hc_swap32_4 (X[z]);
+  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = X[z];
 }
 
 KERNEL_FQ void m22700_loop (KERN_ATTR_TMPS (scrypt_tmp_t))
 {
   const u64 gid = get_global_id (0);
+  const u64 lid = get_local_id (0);
 
   if (gid >= gid_max) return;
 
@@ -542,21 +489,14 @@ KERNEL_FQ void m22700_loop (KERN_ATTR_TMPS (scrypt_tmp_t))
   GLOBAL_AS uint4 *d_scrypt3_buf = (GLOBAL_AS uint4 *) d_extra3_buf;
 
   uint4 X[STATE_CNT4];
-  uint4 T[STATE_CNT4];
 
   const u32 P_offset = salt_repeat * STATE_CNT4;
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) X[z] = hc_swap32_4 (tmps[gid].P[P_offset + z]);
+  for (int z = 0; z < STATE_CNT4; z++) X[z] = tmps[gid].P[P_offset + z];
 
-  scrypt_smix_loop (X, T, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
+  scrypt_smix_loop (X, d_scrypt0_buf, d_scrypt1_buf, d_scrypt2_buf, d_scrypt3_buf);
 
-  #ifdef _unroll
-  #pragma unroll
-  #endif
-  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = hc_swap32_4 (X[z]);
+  for (int z = 0; z < STATE_CNT4; z++) tmps[gid].P[P_offset + z] = X[z];
 }
 
 KERNEL_FQ void m22700_comp (KERN_ATTR_TMPS (scrypt_tmp_t))
@@ -665,35 +605,48 @@ KERNEL_FQ void m22700_comp (KERN_ATTR_TMPS (scrypt_tmp_t))
 
   for (u32 l = 0; l < SCRYPT_CNT4; l += 4)
   {
-    uint4 tmp;
-
-    tmp = tmps[gid].P[l + 0];
+    uint4 X[4];
 
-    w0[0] = tmp.x;
-    w0[1] = tmp.y;
-    w0[2] = tmp.z;
-    w0[3] = tmp.w;
+    X[0] = tmps[gid].P[l + 0];
+    X[1] = tmps[gid].P[l + 1];
+    X[2] = tmps[gid].P[l + 2];
+    X[3] = tmps[gid].P[l + 3];
 
-    tmp = tmps[gid].P[l + 1];
+    uint4 T[4];
 
-    w1[0] = tmp.x;
-    w1[1] = tmp.y;
-    w1[2] = tmp.z;
-    w1[3] = tmp.w;
-
-    tmp = tmps[gid].P[l + 2];
-
-    w2[0] = tmp.x;
-    w2[1] = tmp.y;
-    w2[2] = tmp.z;
-    w2[3] = tmp.w;
-
-    tmp = tmps[gid].P[l + 3];
+    #ifdef IS_CUDA
+    T[0] = make_uint4 (X[0].x, X[3].y, X[2].z, X[1].w);
+    T[1] = make_uint4 (X[1].x, X[0].y, X[3].z, X[2].w);
+    T[2] = make_uint4 (X[2].x, X[1].y, X[0].z, X[3].w);
+    T[3] = make_uint4 (X[3].x, X[2].y, X[1].z, X[0].w);
+    #else
+    T[0] = (uint4) (X[0].x, X[3].y, X[2].z, X[1].w);
+    T[1] = (uint4) (X[1].x, X[0].y, X[3].z, X[2].w);
+    T[2] = (uint4) (X[2].x, X[1].y, X[0].z, X[3].w);
+    T[3] = (uint4) (X[3].x, X[2].y, X[1].z, X[0].w);
+    #endif
 
-    w3[0] = tmp.x;
-    w3[1] = tmp.y;
-    w3[2] = tmp.z;
-    w3[3] = tmp.w;
+    T[0] = hc_swap32_4 (T[0]);
+    T[1] = hc_swap32_4 (T[1]);
+    T[2] = hc_swap32_4 (T[2]);
+    T[3] = hc_swap32_4 (T[3]);
+
+    w0[0] = T[0].x;
+    w0[1] = T[0].y;
+    w0[2] = T[0].z;
+    w0[3] = T[0].w;
+    w1[0] = T[1].x;
+    w1[1] = T[1].y;
+    w1[2] = T[1].z;
+    w1[3] = T[1].w;
+    w2[0] = T[2].x;
+    w2[1] = T[2].y;
+    w2[2] = T[2].z;
+    w2[3] = T[2].w;
+    w3[0] = T[3].x;
+    w3[1] = T[3].y;
+    w3[2] = T[3].z;
+    w3[3] = T[3].w;
 
     sha256_hmac_update_64 (&ctx, w0, w1, w2, w3, 64);
   }
diff --git a/hashcat.hctune b/hashcat.hctune
index 4f14768c2..c8ef4f45c 100644
--- a/hashcat.hctune
+++ b/hashcat.hctune
@@ -370,12 +370,13 @@ GeForce_GTX_TITAN                               3       9900    2       A
 ##
 
 DEVICE_TYPE_CPU                                 *       8900    1       N       A
-DEVICE_TYPE_GPU                                 *       8900    1       N       A
 DEVICE_TYPE_CPU                                 *       9300    1       N       A
-DEVICE_TYPE_GPU                                 *       9300    1       N       A
 DEVICE_TYPE_CPU                                 *       15700   1       N       A
-DEVICE_TYPE_GPU                                 *       15700   1       1       A
 DEVICE_TYPE_CPU                                 *       22700   1       N       A
+
+DEVICE_TYPE_GPU                                 *       8900    1       N       A
+DEVICE_TYPE_GPU                                 *       9300    1       N       A
+DEVICE_TYPE_GPU                                 *       15700   1       1       A
 DEVICE_TYPE_GPU                                 *       22700   1       N       A
 
 ## Here's an example of how to manually tune SCRYPT algorithm kernels for your hardware.
@@ -399,7 +400,7 @@ DEVICE_TYPE_GPU                                 *       22700   1       N
 ## 3. Artificial multiplier (--kernel-accel aka -n)
 ##
 ## In order to find these values:
-## 
+##
 ## 1. On startup Hashcat will show: * Device #1: GeForce GTX 980, 3963/4043 MB, 16MCU. The 16 MCU is the number of compute units on that device.
 ## 2. Native thread counts are fixed values: CPU=1, GPU-Intel=8, GPU-AMD=64 (wavefronts), GPU-NVIDIA=32 (warps)
 ##
@@ -412,12 +413,12 @@ DEVICE_TYPE_GPU                                 *       22700   1       N
 ##
 ## How do we deal with this? This is where SCRYPT TMTO(time-memory trde off) kicks in. The SCRYPT algorithm is designed in such a way that we
 ## can pre-compute that 16MB buffer from a self-choosen offset. Details on how this actually works are not important for this process.
-## 
-## What's relevant to us is that we can halve the buffer size, but we pay with twice the computation time. 
+##
+## What's relevant to us is that we can halve the buffer size, but we pay with twice the computation time.
 ## We can repeat this as often as we want. That's why it's a trade-off.
 ##
 ## This mechanic can be manually set using --scrypt-tmto on the commandline, but this is not the best way.
-## 
+##
 ## Back to our problem. We need 8GB of memory but have only ~4GB.
 ## It's not a full 4GB. The OS needs some of it and Hashcat needs some of it to store password candidates and other things.
 ## If you run a headless server it should be safe to subtract a fixed value of 200MB from whatever you have in your GPU.
@@ -426,7 +427,7 @@ DEVICE_TYPE_GPU                                 *       22700   1       N
 ##
 ## (8GB >> 0) = 8GB < 3.8GB = No, Does not fit
 ## (8GB >> 1) = 4GB < 3.8GB = No, Does not fit
-## (8GB >> 2) = 2GB < 3.8GB = Yes! 
+## (8GB >> 2) = 2GB < 3.8GB = Yes!
 ##
 ## This process is automated in Hashcat, but it is important to understand what's happening here.
 ## Because of the light overhead from the OS and Hashcat, we pay a very high price.
@@ -440,7 +441,7 @@ DEVICE_TYPE_GPU                                 *       22700   1       N
 ## Therefore, we do not need to increase the TMTO by another step to fit in VRAM.
 ##
 ## If we cut down our 16 MCU to only 15 MCU or 14 MCU using --kernel-accel(-n), we end up with:
-## 
+##
 ## 16 * 32 * 16777216 = 8589934592 / 2 = 4294967296 = 4.00GB < 3.80GB = Nope, next
 ## 15 * 32 * 16777216 = 8053063680 / 2 = 4026531840 = 3.84GB < 3.80GB = Nope, next
 ## 14 * 32 * 16777216 = 7516192768 / 2 = 3758096384 = 3.58GB < 3.80GB = Yes!
@@ -459,19 +460,24 @@ DEVICE_TYPE_GPU                                 *       22700   1       N
 ## On my GTX980, this improves the performance from 201 H/s to 255 H/s.
 ## Again, there's no need to control this with --scrypt-tmto. Hashcat will realize it has to increase the TMTO again.
 ##
-## All together, you can control all of this by using the -n parameter in the command line. 
+## All together, you can control all of this by using the -n parameter in the command line.
 ## This is not ideal in a production environment because you must use the --force flag.
 ## The best way to set this is by using this Hashcat.hctune file to store it. This avoids the need to bypass any warnings.
 ##
-## Find the ideal -n value, then store it here along with the proper compute device name. 
+## Find the ideal -n value, then store it here along with the proper compute device name.
 ## Formatting guidelines are availabe at the top of this document.
 
 GeForce_GTX_980                                 *       8900    1      28       A
 GeForce_GTX_980                                 *       9300    1     128       A
-GeForce_GTX_980                                 *       15700   1       1       A
+GeForce_GTX_980                                 *       15700   1       2       A
 GeForce_GTX_980                                 *       22700   1      28       A
 
-GeForce_RTX_2080_Ti                             *       8900    1       N       A
+GeForce_RTX_2080_Ti                             *       8900    1      38       A
 GeForce_RTX_2080_Ti                             *       9300    1     544       A
-GeForce_RTX_2080_Ti                             *       15700   1       4       A
-GeForce_RTX_2080_Ti                             *       22700   1       N       A
+GeForce_RTX_2080_Ti                             *       15700   1       8       A
+GeForce_RTX_2080_Ti                             *       22700   1      38       A
+
+gfx900                                          *       8900    1      28       A
+gfx900                                          *       9300    1     384       A
+gfx900                                          *       15700   1       6       A
+gfx900                                          *       22700   1      28       A
diff --git a/src/backend.c b/src/backend.c
index 79c4944c9..a5e33c236 100644
--- a/src/backend.c
+++ b/src/backend.c
@@ -8381,6 +8381,8 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
     device_param->size_st_salts   = size_st_salts;
     device_param->size_st_esalts  = size_st_esalts;
 
+    // extra buffer
+
     u64 size_extra_buffer = 4;
 
     if (module_ctx->module_extra_buffer_size != MODULE_DEFAULT)
diff --git a/src/modules/module_08900.c b/src/modules/module_08900.c
index 94621e983..fe2ea77d4 100644
--- a/src/modules/module_08900.c
+++ b/src/modules/module_08900.c
@@ -250,24 +250,9 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
 
   const u64 tmp_size = 128ULL * scrypt_r * scrypt_p;
 
-  char *unroll = "";
-
-  // NVIDIA GPU
-  if (device_param->opencl_device_vendor_id == VENDOR_ID_NV)
-  {
-    unroll = "-D _unroll";
-  }
-
-  // ROCM
-  if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->has_vperm == true))
-  {
-    unroll = "-D _unroll";
-  }
-
   char *jit_build_options = NULL;
 
-  hc_asprintf (&jit_build_options, "%s -DSCRYPT_N=%u -DSCRYPT_R=%u -DSCRYPT_P=%u -DSCRYPT_TMTO=%" PRIu64 " -DSCRYPT_TMP_ELEM=%" PRIu64,
-    unroll,
+  hc_asprintf (&jit_build_options, "-DSCRYPT_N=%u -DSCRYPT_R=%u -DSCRYPT_P=%u -DSCRYPT_TMTO=%" PRIu64 " -DSCRYPT_TMP_ELEM=%" PRIu64,
     hashes->salts_buf[0].scrypt_N,
     hashes->salts_buf[0].scrypt_r,
     hashes->salts_buf[0].scrypt_p,
diff --git a/src/modules/module_09300.c b/src/modules/module_09300.c
index 2fc876df2..fbf5a6064 100644
--- a/src/modules/module_09300.c
+++ b/src/modules/module_09300.c
@@ -250,24 +250,9 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
 
   const u64 tmp_size = 128ULL * scrypt_r * scrypt_p;
 
-  char *unroll = "";
-
-  // NVIDIA GPU
-  if (device_param->opencl_device_vendor_id == VENDOR_ID_NV)
-  {
-    unroll = "-D _unroll";
-  }
-
-  // ROCM
-  if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->has_vperm == true))
-  {
-    unroll = "-D _unroll";
-  }
-
   char *jit_build_options = NULL;
 
-  hc_asprintf (&jit_build_options, "%s -DSCRYPT_N=%u -DSCRYPT_R=%u -DSCRYPT_P=%u -DSCRYPT_TMTO=%" PRIu64 " -DSCRYPT_TMP_ELEM=%" PRIu64,
-    unroll,
+  hc_asprintf (&jit_build_options, "-DSCRYPT_N=%u -DSCRYPT_R=%u -DSCRYPT_P=%u -DSCRYPT_TMTO=%" PRIu64 " -DSCRYPT_TMP_ELEM=%" PRIu64,
     hashes->salts_buf[0].scrypt_N,
     hashes->salts_buf[0].scrypt_r,
     hashes->salts_buf[0].scrypt_p,
diff --git a/src/modules/module_15700.c b/src/modules/module_15700.c
index bcf314546..fea92056d 100644
--- a/src/modules/module_15700.c
+++ b/src/modules/module_15700.c
@@ -23,7 +23,6 @@ static const u64   KERN_TYPE      = 15700;
 static const u32   OPTI_TYPE      = OPTI_TYPE_ZERO_BYTE;
 static const u64   OPTS_TYPE      = OPTS_TYPE_PT_GENERATE_LE
                                   | OPTS_TYPE_MP_MULTI_DISABLE
-                                  | OPTS_TYPE_NATIVE_THREADS
                                   | OPTS_TYPE_LOOP_PREPARE
                                   | OPTS_TYPE_SELF_TEST_DISABLE
                                   | OPTS_TYPE_ST_HEX;
@@ -73,6 +72,13 @@ u32 module_kernel_loops_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_
   return kernel_loops_max;
 }
 
+u32 module_kernel_threads_max (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
+{
+  const u32 kernel_threads_max = 4;
+
+  return kernel_threads_max;
+}
+
 u64 module_esalt_size (MAYBE_UNUSED const hashconfig_t *hashconfig, MAYBE_UNUSED const user_options_t *user_options, MAYBE_UNUSED const user_options_extra_t *user_options_extra)
 {
   const u64 esalt_size = (const u64) sizeof (ethereum_scrypt_t);
@@ -265,24 +271,9 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
 
   const u64 tmp_size = 128ULL * scrypt_r * scrypt_p;
 
-  char *unroll = "";
-
-  // NVIDIA GPU
-  if (device_param->opencl_device_vendor_id == VENDOR_ID_NV)
-  {
-    unroll = "-D _unroll";
-  }
-
-  // ROCM
-  if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->has_vperm == true))
-  {
-    unroll = "-D _unroll";
-  }
-
   char *jit_build_options = NULL;
 
-  hc_asprintf (&jit_build_options, "%s -DSCRYPT_N=%u -DSCRYPT_R=%u -DSCRYPT_P=%u -DSCRYPT_TMTO=%" PRIu64 " -DSCRYPT_TMP_ELEM=%" PRIu64,
-    unroll,
+  hc_asprintf (&jit_build_options, "-DSCRYPT_N=%u -DSCRYPT_R=%u -DSCRYPT_P=%u -DSCRYPT_TMTO=%" PRIu64 " -DSCRYPT_TMP_ELEM=%" PRIu64,
     hashes->salts_buf[0].scrypt_N,
     hashes->salts_buf[0].scrypt_r,
     hashes->salts_buf[0].scrypt_p,
@@ -507,7 +498,7 @@ void module_init (module_ctx_t *module_ctx)
   module_ctx->module_kernel_accel_min         = MODULE_DEFAULT;
   module_ctx->module_kernel_loops_max         = module_kernel_loops_max;
   module_ctx->module_kernel_loops_min         = module_kernel_loops_min;
-  module_ctx->module_kernel_threads_max       = MODULE_DEFAULT;
+  module_ctx->module_kernel_threads_max       = module_kernel_threads_max;
   module_ctx->module_kernel_threads_min       = MODULE_DEFAULT;
   module_ctx->module_kern_type                = module_kern_type;
   module_ctx->module_kern_type_dynamic        = MODULE_DEFAULT;
diff --git a/src/modules/module_22700.c b/src/modules/module_22700.c
index 46ffb9542..5842118f3 100644
--- a/src/modules/module_22700.c
+++ b/src/modules/module_22700.c
@@ -251,24 +251,9 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
 
   const u64 tmp_size = 128ULL * scrypt_r * scrypt_p;
 
-  char *unroll = "";
-
-  // NVIDIA GPU
-  if (device_param->opencl_device_vendor_id == VENDOR_ID_NV)
-  {
-    unroll = "-D _unroll";
-  }
-
-  // ROCM
-  if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->has_vperm == true))
-  {
-    unroll = "-D _unroll";
-  }
-
   char *jit_build_options = NULL;
 
-  hc_asprintf (&jit_build_options, "%s -DSCRYPT_N=%u -DSCRYPT_R=%u -DSCRYPT_P=%u -DSCRYPT_TMTO=%" PRIu64 " -DSCRYPT_TMP_ELEM=%" PRIu64,
-    unroll,
+  hc_asprintf (&jit_build_options, "-DSCRYPT_N=%u -DSCRYPT_R=%u -DSCRYPT_P=%u -DSCRYPT_TMTO=%" PRIu64 " -DSCRYPT_TMP_ELEM=%" PRIu64,
     hashes->salts_buf[0].scrypt_N,
     hashes->salts_buf[0].scrypt_r,
     hashes->salts_buf[0].scrypt_p,