v3.9.2.3

Makefile.am

@@ -163,6 +163,7 @@ cpuminer_SOURCES = \
   algo/sha/sph_sha2.c \
   algo/sha/sph_sha2big.c \
   algo/sha/sha2-hash-4way.c \
+  algo/sha/sha256_hash_11way.c \
   algo/sha/sha2.c \
   algo/sha/sha256t-gate.c \
   algo/sha/sha256t-4way.c \

RELEASE_NOTES

@@ -38,7 +38,13 @@ supported.
 Change Log
 ----------
 
-v3.9.2.1
+v3.9.2.3
+
+Another cpu-affinity fix.
+Disabled test code that fails to compile on some CPUs with limited
+AVX512 capabilities.
+
+v3.9.2.2
 
 Fixed some day one cpu-affinity issues.
 

algo-gate-api.c

@@ -345,9 +345,9 @@ const char* const algo_alias_map[][2] =
   { NULL,                NULL           }   
 };
 
-// if arg is a valid alias for a known algo it is updated with the proper name.
+// if arg is a valid alias for a known algo it is updated with the proper
-// No validation of the algo or alias is done, It is the responsinility of the
+// name. No validation of the algo or alias is done, It is the responsinility
-// calling function to validate the algo after return.
+// of the calling function to validate the algo after return.
 void get_algo_alias( char** algo_or_alias )
 {
   int i;
@@ -362,3 +362,21 @@ void get_algo_alias( char** algo_or_alias )
 
 #undef ALIAS
 #undef PROPER
+
+bool submit_solution( struct work *work, void *hash,
+                      struct thr_info *thr, int lane )
+{
+     work_set_target_ratio( work, hash );
+     if ( submit_work( thr, work ) )
+     {
+         applog( LOG_NOTICE, "Share %d submitted by thread %d, lane %d.",
+                 accepted_share_count + rejected_share_count + 1,
+                 thr->id, lane );
+         return true;
+     }
+     else
+          applog( LOG_WARNING, "Failed to submit share." );
+     return false;
+}
+
+

algo-gate-api.h

@@ -196,8 +196,9 @@ void four_way_not_tested();
 int null_scanhash();
 
 // The one and only, a callback for scanhash.
-
+bool submit_solution( struct work *work, void *hash,
-
+                      struct thr_info *thr, int lane );
+ 
 bool submit_work( struct thr_info *thr, const struct work *work_in );
 
 // displays warning

algo/sha/sha2-hash-4way.h

@@ -122,12 +122,11 @@ typedef struct {
 } sha256_11way_context;
 
 void sha256_11way_init( sha256_11way_context *ctx );
-void sha256_11way( sha256_11way_context *ctx, const void *datax,
+void sha256_11way_update( sha256_11way_context *ctx, const void *datax,
-	                 void *datay, void *dataz, size_t len );
+	                 const void *datay, const void *dataz, size_t len );
 void sha256_11way_close( sha256_11way_context *ctx, void *dstx, void *dstyx,
 	                 void *dstz  );
 
-
+#endif  // __AVX2__
-#endif
+#endif  // __SSE2__
-#endif
+#endif  // SHA256_4WAY_H__
-#endif

algo/sha/sha256_hash_11way.c

@@ -9,7 +9,7 @@
 // naming convention for variables and macros
 // VARx: AVX2 8 way 32 bit
 // VARy: MMX 2 way 32 bit
-// VARz: 32 bit integer
+// VARz: scalar integer 32 bit
 
 
 static const uint32_t H256[8] =
@@ -18,7 +18,7 @@ static const uint32_t H256[8] =
         0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
 };
 
-static const uont32_t K256[64] = 
+static const uint32_t K256[64] = 
 {
         0x428A2F98, 0x71374491, 0xB5C0FBCF, 0xE9B5DBA5,
         0x3956C25B, 0x59F111F1, 0x923F82A4, 0xAB1C5ED5,
@@ -57,29 +57,25 @@ static const uont32_t K256[64] =
 
 #define MAJz(X, Y, Z)  ( ( (X) & (Y) ) | ( ( (X) | (Y) ) & (Z) ) )
 
-
 #define BSG2_0x(x) \
    _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,2), mm256_ror_32(x,13) ), mm256_ror_32( x,22) )
+       mm256_ror_32(x,2), mm256_ror_32(x,13) ), _mm256_srli_epi32(x,22) )
 
 #define BSG2_0y(x) \
    _mm_xor_si64( _mm_xor_si64( \
-       mm64_ror_32(x,2), mm64_ror_32(x,13) ), mm64_ror_32( x,22) )
+       mm64_ror_32(x,2), mm64_ror_32(x,13) ), _mm_srli_pi32(x,22) )
-
-#define BSG2_0z(x)  ( ( ror_32(x,2) ^ ror_32(x,13) ) ^ ror_32(x,22) )
 
+#define BSG2_0z(x)  ( ror_32(x,2) ^ ror_32(x,13)  ^ ((x)>>22) )
 
 #define BSG2_1x(x) \
    _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,6), mm256_ror_32(x,11) ), mm256_ror_32( x,25) )
+       mm256_ror_32(x,6), mm256_ror_32(x,11) ), _mm256_srli_epi32(x,25) )
 
 #define BSG2_1y(x) \
    _mm_xor_si64( _mm_xor_si64( \
-       mm64_ror_32(x,6), mm64_ror_32(x,11) ), mm64_ror_32( x,25) )
+       mm64_ror_32(x,6), mm64_ror_32(x,11) ), _mm_srli_pi32(x,25) )
-
-#define BSG2_1z(x) \
-      (mm256_ror_32(x,6) ^ mm256_ror_32(x,11) ^ mm256_ror_32( x,25) )
 
+#define BSG2_1z(x)   ( ror_32(x,6) ^ ror_32(x,11) ^ ((x)>>25) )
 
 #define SSG2_0x(x) \
    _mm256_xor_si256( _mm256_xor_si256( \
@@ -87,30 +83,35 @@ static const uont32_t K256[64] =
 
 #define SSG2_0y(x) \
    _mm_xor_si64( _mm_xor_si64( \
-       mm64_ror_32(x,7), mm64_ror_32(x,18) ), _mm64_srli_pi32(x,3) )
+       mm64_ror_32(x,7), mm64_ror_32(x,18) ), _mm_srli_pi32(x,3) )
-
-#define SSG2_0z(x)  ( ror_32(x,7) ^ ror_32(x,18) ^ ((x)>>3) )
 
+#define SSG2_0z(x)  (( ror_32(x,7) ^ ror_32(x,18) ) ^ ((x)>>3) )
 
 #define SSG2_1x(x) \
    _mm256_xor_si256( _mm256_xor_si256( \
        mm256_ror_32(x,17), mm256_ror_32(x,19) ), _mm256_srli_epi32(x,10) )
 
+#define SSG2_1y(x) \
+   _mm_xor_si64( _mm_xor_si64( \
+       mm64_ror_32(x,17), mm64_ror_32(x,19) ), _mm_srli_pi32(x,10) )
+
+#define SSG2_1z(x)   ( ror_32(x,17) ^ ror_32(x,19)  ^ ((x)>>10) )
+
 #define SHA2x_MEXP( a, b, c, d ) \
      _mm256_add_epi32( _mm256_add_epi32( _mm256_add_epi32( \
-                 SSG2_1x( Wx[a] ), Wx[b] ), SSG2_0x( Wx[c] ) ), Wx[d] );
+                 SSG2_1x( Wx[a] ), Wx[b] ), SSG2_0x( Wx[c] ) ), Wx[d] )
 
 #define SHA2y_MEXP( a, b, c, d ) \
      _mm_add_pi32( _mm_add_pi32( _mm_add_pi32( \
-                 SSG2_1y( Wy[a] ), Wy[b] ), SSG2_0y( Wy[c] ) ), Wy[d] );
+                 SSG2_1y( Wy[a] ), Wy[b] ), SSG2_0y( Wy[c] ) ), Wy[d] )
 
 #define SHA2z_MEXP( a, b, c, d ) \
-               ( SSG2_1z( Wz[a] ) + Wz[b] + SSG2_0z( Wz[c] ) + Wz[d] );
+               ( SSG2_1z( Wz[a] ) + Wz[b] + SSG2_0z( Wz[c] ) + Wz[d] )
 
 
 #define SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx, \
 	                  Ay, By, Cy, Dy, Ey, Fy, Gy, Hy, \
-		          Ax, Bx, Cz, Dz, Ez, Fz, Gz, Hz, i, j) \
+		          Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz, i, j) \
 do { \
   __m256i T1x, T2x; \
   __m64 T1y, T2y; \
@@ -119,22 +120,22 @@ do { \
         _mm256_add_epi32( Hx, BSG2_1x(Ex) ), CHx(Ex, Fx, Gx) ), \
                           _mm256_set1_epi32( K256[( (j)+(i) )] ) ), Wx[i] ); \
   T1y = _mm_add_pi32( _mm_add_pi32( _mm_add_pi32( \
-        _mm_add_pi32( H, BSG2_1x(Ey) ), CHx(Ey, Fy, Gy) ), \
+        _mm_add_pi32( Hy, BSG2_1y(Ey) ), CHy(Ey, Fy, Gy) ), \
                           _mm_set1_pi32( K256[( (j)+(i) )] ) ), Wy[i] ); \
   T1z = Hz + BSG2_1z( Ez ) + CHz( Ez, Fz, Gz ) + K256[ ((j)+(i)) ] + Wz[i]; \
   T2x = _mm256_add_epi32( BSG2_0x(Ax), MAJx(Ax, Bx, Cx) ); \
-  T2y = _mm256_add_epi32( BSG2_0y(Ay), MAJy(Ay, By, Cy) ); \
+  T2y = _mm_add_pi32( BSG2_0y(Ay), MAJy(Ay, By, Cy) ); \
-  T2z = BSG2_0z( Az ) + MAJz( Az, Bz, Cz ); \ \
+  T2z = BSG2_0z( Az ) + MAJz( Az, Bz, Cz ); \
   Dx  = _mm256_add_epi32( Dx,  T1x ); \
-  Dy  = _mm256_add_epi32( Dy,  T1y ); \
+  Dy  = _mm_add_pi32( Dy, T1y ); \
   Dz  = Dz + T1z; \
   Hx  = _mm256_add_epi32( T1x, T2x ); \
-  Hy = _mm256_add_epi32( T1y, T2y ); \
+  Hy  = _mm_add_pi32( T1y, T2y ); \
-  Hz = T1z + T2z; \
+  Hz  = T1z + T2z; \
 } while (0)
 	
-sha256_8way_round( __m256i *inx, __m256i rx[8], __m64 *iny, __m64 *ry[8],
+void sha256_11way_round( __m256i *inx, __m256i rx[8], __m64 *iny, __m64 ry[8],
-	           uint32_t inz, uint32_t *rz[8] )
+                         uint32_t *inz, uint32_t rz[8] )
 {
    __m256i Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx;
    __m256i Wx[16];
@@ -169,43 +170,43 @@ sha256_8way_round( __m256i *inx, __m256i rx[8], __m64 *iny, __m64 *ry[8],
 
    Wx[ 6] = mm256_bswap_32( inx[ 6] );
    Wy[ 6] =  mm64_bswap_32( iny[ 6] );
-   Wz[ 6] =       bswap_32( inx[ 6] );
+   Wz[ 6] =       bswap_32( inz[ 6] );
 
    Wx[ 7] = mm256_bswap_32( inx[ 7] );
    Wy[ 7] =  mm64_bswap_32( iny[ 7] );
-   Wz[ 7] =       bswap_32( inx[ 7] );
+   Wz[ 7] =       bswap_32( inz[ 7] );
 
    Wx[ 8] = mm256_bswap_32( inx[ 8] );
    Wy[ 8] =  mm64_bswap_32( iny[ 8] );
-   Wz[ 8] =       bswap_32( inx[ 8] );
+   Wz[ 8] =       bswap_32( inz[ 8] );
 
    Wx[ 9] = mm256_bswap_32( inx[ 9] );
    Wy[ 9] =  mm64_bswap_32( iny[ 9] );
-   Wz[ 9] =       bswap_32( inx[ 9] );
+   Wz[ 9] =       bswap_32( inz[ 9] );
 
    Wx[10] = mm256_bswap_32( inx[10] );
    Wy[10] =  mm64_bswap_32( iny[10] );
-   Wz[10] =       bswap_32( inx[10] );
+   Wz[10] =       bswap_32( inz[10] );
 
    Wx[11] = mm256_bswap_32( inx[11] );
    Wy[11] =  mm64_bswap_32( iny[11] );
-   Wz[11] =       bswap_32( inx[11] );
+   Wz[11] =       bswap_32( inz[11] );
 
    Wx[12] = mm256_bswap_32( inx[12] );
    Wy[12] =  mm64_bswap_32( iny[12] );
-   Wz[12] =       bswap_32( inx[12] );
+   Wz[12] =       bswap_32( inz[12] );
 
    Wx[13] = mm256_bswap_32( inx[13] );
    Wy[13] =  mm64_bswap_32( iny[13] );
-   Wz[13] =       bswap_32( inx[13] );
+   Wz[13] =       bswap_32( inz[13] );
 
    Wx[14] = mm256_bswap_32( inx[14] );
    Wy[14] =  mm64_bswap_32( iny[14] );
-   Wz[14] =       bswap_32( inx[14] );
+   Wz[14] =       bswap_32( inz[14] );
 
    Wx[15] = mm256_bswap_32( inx[15] );
    Wy[15] =  mm64_bswap_32( iny[15] );
-   Wz[15] =       bswap_32( inx[15] );
+   Wz[15] =       bswap_32( inz[15] );
 
    SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx,
                      Ay, By, Cy, Dy, Ey, Fy, Gy, Hy,
@@ -325,52 +326,52 @@ sha256_8way_round( __m256i *inx, __m256i rx[8], __m64 *iny, __m64 *ry[8],
 
       SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx,
                         Ay, By, Cy, Dy, Ey, Fy, Gy, Hy,
-			Az, By, Cz, Dz, Ez, Fy, Gz, Hz,	 0, j );
+			Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz,	 0, j );
       SHA2s_11WAY_STEP( Hx, Ax, Bx, Cx, Dx, Ex, Fx, Gx,
 		        Hy, Ay, By, Cy, Dy, Ey, Fy, Gy,
-		       	HZ, Az, By, Cz, Dz, Ez, Fy, Gz,  1, j );
+		       	Hz, Az, Bz, Cz, Dz, Ez, Fz, Gz,  1, j );
       SHA2s_11WAY_STEP( Gx, Hx, Ax, Bx, Cx, Dx, Ex, Fx,
 		        Gy, Hy, Ay, By, Cy, Dy, Ey, Fy,
-		       	Gz, HZ, Az, By, Cz, Dz, Ez, Fy,  2, j );
+		       	Gz, Hz, Az, Bz, Cz, Dz, Ez, Fz,  2, j );
       SHA2s_11WAY_STEP( Fx, Gx, Hx, Ax, Bx, Cx, Dx, Ex,
 		        Fy, Gy, Hy, Ay, By, Cy, Dy, Ey,
-		       	Fz, Gz, HZ, Az, By, Cz, Dz, Ez,  3, j );
+		       	Fz, Gz, Hz, Az, Bz, Cz, Dz, Ez,  3, j );
       SHA2s_11WAY_STEP( Ex, Fx, Gx, Hx, Ax, Bx, Cx, Dx,
 		        Ey, Fy, Gy, Hy, Ay, By, Cy, Dy,
-		       	Ez, Fz, Gz, HZ, Az, By, Cz, Dz,  4, j );
+		       	Ez, Fz, Gz, Hz, Az, Bz, Cz, Dz,  4, j );
       SHA2s_11WAY_STEP( Dx, Ex, Fx, Gx, Hx, Ax, Bx, Cx,
 		        Dy, Ey, Fy, Gy, Hy, Ay, By, Cy,
-		       	Dz, Ez, Fz, Gz, HZ, Az, By, Cz,  5, j );
+		       	Dz, Ez, Fz, Gz, Hz, Az, Bz, Cz,  5, j );
       SHA2s_11WAY_STEP( Cx, Dx, Ex, Fx, Gx, Hx, Ax, Bx,
 		        Cy, Dy, Ey, Fy, Gy, Hy, Ay, By,
-		       	Cz, Dz, Ez, Fz, Gz, HZ, Az, By,  6, j );
+		       	Cz, Dz, Ez, Fz, Gz, Hz, Az, Bz,  6, j );
       SHA2s_11WAY_STEP( Bx, Cx, Dx, Ex, Fx, Gx, Hx, Ax,
 		        By, Cy, Dy, Ey, Fy, Gy, Hy, Ay,
-		       	Bz, Cz, Dz, Ez, Fz, Gz, HZ, Az,  7, j );
+		       	Bz, Cz, Dz, Ez, Fz, Gz, Hz, Az,  7, j );
       SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx,
                         Ay, By, Cy, Dy, Ey, Fy, Gy, Hy,
-                        Az, By, Cz, Dz, Ez, Fy, Gz, Hz,  8, j );
+                        Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz,  8, j );
       SHA2s_11WAY_STEP( Hx, Ax, Bx, Cx, Dx, Ex, Fx, Gx, 
                         Hy, Ay, By, Cy, Dy, Ey, Fy, Gy, 
-                        HZ, Az, By, Cz, Dz, Ez, Fy, Gz,  9, j );
+                        Hz, Az, Bz, Cz, Dz, Ez, Fz, Gz,  9, j );
       SHA2s_11WAY_STEP( Gx, Hx, Ax, Bx, Cx, Dx, Ex, Fx, 
                         Gy, Hy, Ay, By, Cy, Dy, Ey, Fy, 
-                        Gz, HZ, Az, By, Cz, Dz, Ez, Fy, 10, j );
+                        Gz, Hz, Az, Bz, Cz, Dz, Ez, Fz, 10, j );
       SHA2s_11WAY_STEP( Fx, Gx, Hx, Ax, Bx, Cx, Dx, Ex, 
                         Fy, Gy, Hy, Ay, By, Cy, Dy, Ey, 
-                        Fz, Gz, HZ, Az, By, Cz, Dz, Ez, 11, j );
+                        Fz, Gz, Hz, Az, Bz, Cz, Dz, Ez, 11, j );
       SHA2s_11WAY_STEP( Ex, Fx, Gx, Hx, Ax, Bx, Cx, Dx, 
                         Ey, Fy, Gy, Hy, Ay, By, Cy, Dy, 
-                        Ez, Fz, Gz, HZ, Az, By, Cz, Dz, 12, j );
+                        Ez, Fz, Gz, Hz, Az, Bz, Cz, Dz, 12, j );
       SHA2s_11WAY_STEP( Dx, Ex, Fx, Gx, Hx, Ax, Bx, Cx, 
                         Dy, Ey, Fy, Gy, Hy, Ay, By, Cy, 
-                        Dz, Ez, Fz, Gz, HZ, Az, By, Cz, 13, j );
+                        Dz, Ez, Fz, Gz, Hz, Az, Bz, Cz, 13, j );
       SHA2s_11WAY_STEP( Cx, Dx, Ex, Fx, Gx, Hx, Ax, Bx, 
                         Cy, Dy, Ey, Fy, Gy, Hy, Ay, By, 
-                        Cz, Dz, Ez, Fz, Gz, HZ, Az, By, 14, j );
+                        Cz, Dz, Ez, Fz, Gz, Hz, Az, Bz, 14, j );
       SHA2s_11WAY_STEP( Bx, Cx, Dx, Ex, Fx, Gx, Hx, Ax, 
                         By, Cy, Dy, Ey, Fy, Gy, Hy, Ay, 
-                        Bz, Cz, Dz, Ez, Fz, Gz, HZ, Az, 15, j );
+                        Bz, Cz, Dz, Ez, Fz, Gz, Hz, Az, 15, j );
    }
 
    rx[0] = _mm256_add_epi32( rx[0], Ax );
@@ -384,7 +385,7 @@ sha256_8way_round( __m256i *inx, __m256i rx[8], __m64 *iny, __m64 *ry[8],
    rz[3] =                   rz[3]+ Dz;
    rx[4] = _mm256_add_epi32( rx[4], Ex );
    ry[4] =     _mm_add_pi32( ry[4], Ey );
-   rz[4] =                   rz[4], Ez;
+   rz[4] =                   rz[4]+ Ez;
    rx[5] = _mm256_add_epi32( rx[5], Fx );
    ry[5] =     _mm_add_pi32( ry[5], Fy );
    rz[5] =                   rz[5]+ Fz;
@@ -397,7 +398,7 @@ sha256_8way_round( __m256i *inx, __m256i rx[8], __m64 *iny, __m64 *ry[8],
 
 }
 
-void sha256_8way_init( sha256_11way_context *ctx )
+void sha256_11way_init( sha256_11way_context *ctx )
 {
    ctx->count_high = ctx->count_low = 0;
    ctx->valx[0] = _mm256_set1_epi32( H256[0] );
@@ -416,12 +417,12 @@ void sha256_8way_init( sha256_11way_context *ctx )
    ctx->valy[6] =     _mm_set1_pi32( H256[0] );
    ctx->valx[7] = _mm256_set1_epi32( H256[0] );
    ctx->valy[7] =     _mm_set1_pi32( H256[0] );
-   memscpy( ctx->valz, H256, 32 );
+   memcpy( ctx->valz, H256, 32 );
 }
 
 
-void sha256_11way( sha256_11way_context *ctx, const void *datax,
+void sha256_11way_update( sha256_11way_context *ctx, const void *datax,
-	          const void *datay, const void *dataz, size_t len )
+	                  const void *datay, const void *dataz, size_t len )
 {
    __m256i  *vdatax = (__m256i*) datax;
     __m64   *vdatay = (__m64*)   datay;
@@ -440,26 +441,26 @@ void sha256_11way( sha256_11way_context *ctx, const void *datax,
          clen = len;
       memcpy_256( ctx->bufx + (ptr>>2), vdatax + (ptr>>2), clen>>2 );
       memcpy_64 ( ctx->bufy + (ptr>>2), vdatay + (ptr>>2), clen>>2 );
-      memcpy    ( ctx->bufz +  ptr,     sdataz +  ptr,     clen    );
+      memcpy    ( ctx->bufz +  ptr,     idataz +  ptr,     clen    );
       ptr += clen;
       len -= clen;
       if ( ptr == buf_size )
       {
          sha256_11way_round( ctx->bufx, ctx->valx,
 			     ctx->bufy, ctx->valy,
-			     ctx->bufz, ctx->valzx, );
+			     ctx->bufz, ctx->valz );
          ptr = 0;
       }
-      clow = sc->count_low;
+      clow = ctx->count_low;
       clow2 = clow + clen;
-      sc->count_low = clow2;
+      ctx->count_low = clow2;
       if ( clow2 < clow )
-         sc->count_high++;
+         ctx->count_high++;
    }
 }
 
 
-void sha256_11way_close( sha256_11way_context *ctx, void *dstx, void dsty,
+void sha256_11way_close( sha256_11way_context *ctx, void *dstx, void *dsty,
 	                                            void *dstz)
 {
     unsigned ptr, u;
@@ -487,9 +488,9 @@ void sha256_11way_close( sha256_11way_context *ctx, void *dstx, void dsty,
     }
     else
     {
-        memset_zero_256( ctx->bufx + (ptr>>2), (pad - ptr) >> 2 );
+        memset_zero_256( ctx->bufx + (ptr>>2),    (pad - ptr) >> 2 );
-        memset_zero_64(  ctx->bufy + (ptr>>2), (pad - ptr) >> 2 );
+        memset_zero_64(  ctx->bufy + (ptr>>2),    (pad - ptr) >> 2 );
-        memset(        ctx->bufz + (ptr>>2), 0 (pad - ptr) >> 2 );
+        memset(          ctx->bufz + (ptr>>2), 0, (pad - ptr) >> 2 );
     }
 
     low = ctx->count_low;
@@ -511,9 +512,9 @@ void sha256_11way_close( sha256_11way_context *ctx, void *dstx, void dsty,
     ctx->bufz[ ( pad+4 ) >> 2 ] =
                  bswap_32( low );
 
-    sha256_8way_round( ctx->bufx, ctx->valx,
+    sha256_11way_round( ctx->bufx, ctx->valx,
 		       ctx->bufy, ctx->valy,
-		       ctx->bufz, ctx->valz,  );
+		       ctx->bufz, ctx->valz  );
 
     for ( u = 0; u < 8; u ++ )
     {
@@ -523,4 +524,4 @@ void sha256_11way_close( sha256_11way_context *ctx, void *dstx, void dsty,
    }
 }
 
-
+#endif

algo/sha/sha256t-4way.c

@@ -5,6 +5,136 @@
 #include <stdio.h>
 #include "sha2-hash-4way.h"
 
+#if defined(SHA256T_11WAY)
+
+static __thread sha256_11way_context sha256_ctx11 __attribute__ ((aligned (64)));
+
+void sha256t_11way_hash( void *outx, void *outy, void *outz, const void *inpx,
+	                 const void *inpy, const void*inpz )
+{
+   uint32_t hashx[8*8] __attribute__ ((aligned (64)));
+   uint32_t hashy[8*2] __attribute__ ((aligned (64)));
+   uint32_t hashz[8]   __attribute__ ((aligned (64)));
+   sha256_11way_context ctx;
+   const void *inpx64 = inpx+(64<<3);
+   const void *inpy64 = inpy+(64<<1);
+   const void *inpz64 = inpz+ 64;
+
+   memcpy( &ctx, &sha256_ctx11, sizeof ctx );
+   sha256_11way_update( &ctx, inpx64, inpy64, inpz64,  16 );
+   sha256_11way_close( &ctx, hashx, hashy, hashz );
+
+   sha256_11way_init( &ctx );
+   sha256_11way_update( &ctx, hashx, hashy, hashz, 32 );
+   sha256_11way_close( &ctx, hashx, hashy, hashz );
+
+   sha256_11way_init( &ctx );
+   sha256_11way_update( &ctx, hashx, hashy, hashz, 32 );
+   sha256_11way_close( &ctx, outx, outy, outz );
+}
+
+int scanhash_sha256t_11way( int thr_id, struct work *work, uint32_t max_nonce,
+	                    uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t datax[20*8]  __attribute__ ((aligned (64)));
+   uint32_t datay[20*2]  __attribute__ ((aligned (32)));
+   uint32_t dataz[20]    __attribute__ ((aligned (32)));
+   uint32_t hashx[8*8]   __attribute__ ((aligned (32)));
+   uint32_t hashy[8*2]   __attribute__ ((aligned (32)));
+   uint32_t hashz[8]     __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
+   uint32_t *hash7;
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t Htarg = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   uint32_t n = first_nonce;
+   __m256i  *noncex = (__m256i*) datax + 19;
+   __m64    *noncey = (__m64*)   datay + 19;
+   uint32_t *noncez = (uint32_t*)dataz + 19;
+   /* int */ thr_id = mythr->id;  // thr_id arg is deprecated
+   int i;
+   const uint64_t htmax[] = {           0,
+                                      0xF,
+                                     0xFF,
+                                    0xFFF,
+                                   0xFFFF,
+                               0x10000000 };
+   const uint32_t masks[] = {  0xFFFFFFFF,
+                               0xFFFFFFF0,
+                               0xFFFFFF00,
+                               0xFFFFF000,
+                               0xFFFF0000,
+                                        0 };
+
+   // Use dataz (scalar) to stage bswapped data for the vectors.
+   casti_m256i( dataz, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
+   casti_m256i( dataz, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
+   casti_m128i( dataz, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
+
+   mm256_interleave_8x32( datax, dataz, dataz, dataz, dataz,
+                                 dataz, dataz, dataz, dataz, 640 );
+   mm64_interleave_2x32( datay, dataz, dataz, 640 );
+
+   sha256_11way_init( &sha256_ctx11 );
+   sha256_11way_update( &sha256_ctx11, datax, datay, dataz, 64 );
+
+   for ( int m = 0; m < 6; m++ ) if ( Htarg <= htmax[m] )
+   {
+      uint32_t mask = masks[m];
+      do
+      {
+        *noncex = mm256_bswap_32(
+		 _mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n ) );
+        *noncey = mm64_bswap_32( _mm_set_pi32( n+9, n+8 ) );
+        *noncez = bswap_32( n+10 );
+
+       	pdata[19] = n;
+
+        sha256t_11way_hash( hashx, hashy, hashz, datax, datay, dataz );
+
+        if ( opt_benchmark ) { n += 11; continue; }
+
+        hash7 = &(hashx[7<<3]); 
+        for ( i = 0; i < 8; i++ ) if ( !( hash7[ i ] & mask ) )
+        { 
+            // deinterleave hash for lane
+            mm256_extract_lane_8x32( lane_hash, hashx, i, 256 );
+            if ( fulltest( lane_hash, ptarget ) )
+            {
+	       pdata[19] = n + i;
+               submit_solution( work, lane_hash, mythr, i );
+            }
+	}
+
+	hash7 = &(hashy[7<<1]);
+        for( i = 0; i < 2; i++ ) if ( !(hash7[ 0] & mask ) )
+        {
+            mm64_extract_lane_2x32( lane_hash, hashy, i, 256 );
+  	    if ( fulltest( lane_hash, ptarget ) )
+            {
+               pdata[19] = n + 8 + i;
+               submit_solution( work, lane_hash, mythr, i+8 );
+            }
+	 }
+
+	 if ( !(hashz[7] & mask ) && fulltest( hashz, ptarget ) )
+         {
+            pdata[19] = n+10;
+            submit_solution( work, hashz, mythr, 10 );
+         }
+         n += 11;
+
+      } while ( (n < max_nonce-12) && !work_restart[thr_id].restart );
+      break;
+   }
+    
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+#endif
+
 #if defined(SHA256T_8WAY)
 
 static __thread sha256_8way_context sha256_ctx8 __attribute__ ((aligned (64)));
@@ -29,7 +159,7 @@ void sha256t_8way_hash( void* output, const void* input )
 }
 
 int scanhash_sha256t_8way( int thr_id, struct work *work, uint32_t max_nonce,
-	                   uint64_t *hashes_done, struct thr_info *mythr )
+                           uint64_t *hashes_done, struct thr_info *mythr )
 {
    uint32_t vdata[20*8] __attribute__ ((aligned (64)));
    uint32_t hash[8*8] __attribute__ ((aligned (32)));
@@ -71,39 +201,31 @@ int scanhash_sha256t_8way( int thr_id, struct work *work, uint32_t max_nonce,
       do
       {
         *noncev = mm256_bswap_32(
-		 _mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n ) );
+                 _mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n ) );
-
+         pdata[19] = n;
-	 pdata[19] = n;
 
          sha256t_8way_hash( hash, vdata );
 
-         uint32_t *hash7 = &(hash[7<<3]); 
+         uint32_t *hash7 = &(hash[7<<3]);
-	 
+
          for ( int lane = 0; lane < 8; lane++ )
          if ( !( hash7[ lane ] & mask ) )
-         { 
+         {
             // deinterleave hash for lane
-	    uint32_t lane_hash[8] __attribute__ ((aligned (64)));
+            uint32_t lane_hash[8] __attribute__ ((aligned (64)));
-	    mm256_extract_lane_8x32( lane_hash, hash, lane, 256 );
+            mm256_extract_lane_8x32( lane_hash, hash, lane, 256 );
 
-	    if ( fulltest( lane_hash, ptarget ) )
+            if ( fulltest( lane_hash, ptarget ) )
             {
-	      pdata[19] = n + lane;
+              pdata[19] = n + lane;
-              work_set_target_ratio( work, lane_hash );
+              submit_solution( work, lane_hash, mythr, lane );
-              if ( submit_work( mythr, work ) )
-                applog( LOG_NOTICE, "Share %d submitted by thread %d, lane %d.",
-                             accepted_share_count + rejected_share_count + 1,
-                             thr_id, lane );
-              else
-                applog( LOG_WARNING, "Failed to submit share." );
 	    }
-	 }
+         }
          n += 8;
 
       } while ( (n < max_nonce-10) && !work_restart[thr_id].restart );
       break;
    }
-    
    *hashes_done = n - first_nonce + 1;
    return 0;
 }
@@ -189,22 +311,14 @@ int scanhash_sha256t_4way( int thr_id, struct work *work, uint32_t max_nonce,
             if ( fulltest( lane_hash, ptarget ) )
             {
               pdata[19] = n + lane;
-              work_set_target_ratio( work, lane_hash );
+              submit_solution( work, lane_hash, mythr, lane );
-              if ( submit_work( mythr, work ) )
+	    }
-                applog( LOG_NOTICE, "Share %d submitted by thread %d, lane %d.",
-                             accepted_share_count + rejected_share_count + 1,
-                             thr_id, lane );
-              else
-                applog( LOG_WARNING, "Failed to submit share." );
-            }
          }
-
 	 n += 4;
 
       } while ( (n < max_nonce - 4) && !work_restart[thr_id].restart );
       break;
    }
-
    *hashes_done = n - first_nonce + 1;
    return 0;
 }

algo/sha/sha256t-gate.c

@@ -2,7 +2,11 @@
 
 bool register_sha256t_algo( algo_gate_t* gate )
 {
-#if defined(SHA256T_8WAY)
+#if defined(SHA256T_11WAY)
+    gate->optimizations = SSE2_OPT | AVX2_OPT | SHA_OPT;
+    gate->scanhash   = (void*)&scanhash_sha256t_11way;
+    gate->hash       = (void*)&sha256t_11way_hash;
+#elif defined(SHA256T_8WAY)
     gate->optimizations = SSE2_OPT | AVX2_OPT | SHA_OPT;
     gate->scanhash   = (void*)&scanhash_sha256t_8way;
     gate->hash       = (void*)&sha256t_8way_hash;

algo/sha/sha256t-gate.h

@@ -6,18 +6,29 @@
 
 // Override multi way on ryzen, SHA is better.
 #if !defined(RYZEN_)
-//#if defined(__SSE4_2__)
 #if defined(__SSE2__)
   #define SHA256T_4WAY
 #endif
 #if defined(__AVX2__)
   #define SHA256T_8WAY
+//  #define SHA256T_11WAY
 #endif
 #endif
 
 bool register_sha256t_algo( algo_gate_t* gate );
 bool register_sha256q_algo( algo_gate_t* gate );
 
+#if defined(SHA256T_11WAY)
+
+void sha256t_11way_hash( void *outx, void *outy, void *outz, const void *inpx,
+	                 const void *inpy, const void *inpz );
+int scanhash_sha256t_11way( int thr_id, struct work *work, uint32_t max_nonce,
+                            uint64_t *hashes_done, struct thr_info *mythr );
+//void sha256q_8way_hash( void *output, const void *input );
+//int scanhash_sha256q_11way( int thr_id, struct work *work, uint32_t max_nonce,
+//                            uint64_t *hashes_done, struct thr_info *mythr );
+#endif
+
 #if defined(SHA256T_8WAY)
 
 void sha256t_8way_hash( void *output, const void *input );
@@ -26,8 +37,9 @@ int scanhash_sha256t_8way( int thr_id, struct work *work, uint32_t max_nonce,
 void sha256q_8way_hash( void *output, const void *input );
 int scanhash_sha256q_8way( int thr_id, struct work *work, uint32_t max_nonce,
                            uint64_t *hashes_done, struct thr_info *mythr );
+#endif
 
-#elif defined(SHA256T_4WAY)
+#if defined(SHA256T_4WAY)
 
 void sha256t_4way_hash( void *output, const void *input );
 int scanhash_sha256t_4way( int thr_id, struct work *work, uint32_t max_nonce,
@@ -35,7 +47,7 @@ int scanhash_sha256t_4way( int thr_id, struct work *work, uint32_t max_nonce,
 void sha256q_4way_hash( void *output, const void *input );
 int scanhash_sha256q_4way( int thr_id, struct work *work, uint32_t max_nonce,
                            uint64_t *hashes_done, struct thr_info *mythr );
-#else
+#endif
 
 void sha256t_hash( void *output, const void *input );
 int scanhash_sha256t( int thr_id, struct work *work, uint32_t max_nonce,
@@ -46,5 +58,3 @@ int scanhash_sha256q( int thr_id, struct work *work, uint32_t max_nonce,
 
 #endif
 
-#endif
-

avxdefs.h

@@ -101,14 +101,14 @@
 
 // First some integer stuff that mirrors the SIMD utilities
 
-#define ROR_64( x, c ) ((x)>>(c) | ((x)<<(64-(c))))
+#define ror_64( x, c ) (((x)>>(c)) | ((x)<<(64-(c))))
-#define ROL_64( x, c ) ((x)<<(c) | ((x)>>(64-(c))))
+#define rol_64( x, c ) (((x)<<(c)) | ((x)>>(64-(c))))
-#define ROR_32( x, c ) ((x)>>(c) | ((x)<<(32-(c))))
+#define ror_32( x, c ) (((x)>>(c)) | ((x)<<(32-(c))))
-#define ROL_32( x, c ) ((x)<<(c) | ((x)>>(32-(c))))
+#define rol_32( x, c ) (((x)<<(c)) | ((x)>>(32-(c))))
-#define BSWAP_64( x )  __builtin_bswap64(x)
+#define bswap_64( x )  __builtin_bswap64(x)
-#define BSWAP_32( x )  __builtin_bswap32(x)
+#define bswap_32( x )  __builtin_bswap32(x)
 
-// __int128
+// 128 bit integer
 
 typedef unsigned __int128 uint128_t;
 
@@ -123,11 +123,7 @@ typedef unsigned __int128 uint128_t;
 // There are rumours MMX wil be removed. Although casting with int64
 // works there is likely some overhead to move the data to An MMX register
 // and back.
-// Byte swap and rotation may be more efficient using an MMX shuffle
+// Byte swap and rotation may be more efficient using an MMX shuffle.
-// except that it won't compile due to a "target specific option mismatch"
-// with "inlining failed in call to always inline". MMX was designed for
-// 32 bit CPUs and might not work on 64 bit CPUs where the CPU has full
-// support for 64 bit operations without vectoring.  
 //
 // Universal 64 bit overlay
 union _m64v
@@ -1939,7 +1935,7 @@ do { \
 
 #endif   // AVX512F
 
-#if 1
+#if 0
 //////////////////////////////////////////////////
 //
 //   Compile test.

configure

@@ -1,6 +1,6 @@
 #! /bin/sh
 # Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.69 for cpuminer-opt 3.9.2.2.
+# Generated by GNU Autoconf 2.69 for cpuminer-opt 3.9.2.3.
 #
 #
 # Copyright (C) 1992-1996, 1998-2012 Free Software Foundation, Inc.
@@ -577,8 +577,8 @@ MAKEFLAGS=
 # Identity of this package.
 PACKAGE_NAME='cpuminer-opt'
 PACKAGE_TARNAME='cpuminer-opt'
-PACKAGE_VERSION='3.9.2.2'
+PACKAGE_VERSION='3.9.2.3'
-PACKAGE_STRING='cpuminer-opt 3.9.2.2'
+PACKAGE_STRING='cpuminer-opt 3.9.2.3'
 PACKAGE_BUGREPORT=''
 PACKAGE_URL=''
 
@@ -1332,7 +1332,7 @@ if test "$ac_init_help" = "long"; then
   # Omit some internal or obsolete options to make the list less imposing.
   # This message is too long to be a string in the A/UX 3.1 sh.
   cat <<_ACEOF
-\`configure' configures cpuminer-opt 3.9.2.2 to adapt to many kinds of systems.
+\`configure' configures cpuminer-opt 3.9.2.3 to adapt to many kinds of systems.
 
 Usage: $0 [OPTION]... [VAR=VALUE]...
 
@@ -1404,7 +1404,7 @@ fi
 
 if test -n "$ac_init_help"; then
   case $ac_init_help in
-     short | recursive ) echo "Configuration of cpuminer-opt 3.9.2.2:";;
+     short | recursive ) echo "Configuration of cpuminer-opt 3.9.2.3:";;
    esac
   cat <<\_ACEOF
 
@@ -1509,7 +1509,7 @@ fi
 test -n "$ac_init_help" && exit $ac_status
 if $ac_init_version; then
   cat <<\_ACEOF
-cpuminer-opt configure 3.9.2.2
+cpuminer-opt configure 3.9.2.3
 generated by GNU Autoconf 2.69
 
 Copyright (C) 2012 Free Software Foundation, Inc.
@@ -2012,7 +2012,7 @@ cat >config.log <<_ACEOF
 This file contains any messages produced by compilers while
 running configure, to aid debugging if configure makes a mistake.
 
-It was created by cpuminer-opt $as_me 3.9.2.2, which was
+It was created by cpuminer-opt $as_me 3.9.2.3, which was
 generated by GNU Autoconf 2.69.  Invocation command line was
 
   $ $0 $@
@@ -2993,7 +2993,7 @@ fi
 
 # Define the identity of the package.
  PACKAGE='cpuminer-opt'
- VERSION='3.9.2.2'
+ VERSION='3.9.2.3'
 
 
 cat >>confdefs.h <<_ACEOF
@@ -6690,7 +6690,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
 # report actual input values of CONFIG_FILES etc. instead of their
 # values after options handling.
 ac_log="
-This file was extended by cpuminer-opt $as_me 3.9.2.2, which was
+This file was extended by cpuminer-opt $as_me 3.9.2.3, which was
 generated by GNU Autoconf 2.69.  Invocation command line was
 
   CONFIG_FILES    = $CONFIG_FILES
@@ -6756,7 +6756,7 @@ _ACEOF
 cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
 ac_cs_config="`$as_echo "$ac_configure_args" | sed 's/^ //; s/[\\""\`\$]/\\\\&/g'`"
 ac_cs_version="\\
-cpuminer-opt config.status 3.9.2.2
+cpuminer-opt config.status 3.9.2.3
 configured by $0, generated by GNU Autoconf 2.69,
   with options \\"\$ac_cs_config\\"
 

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([cpuminer-opt], [3.9.2.2])
+AC_INIT([cpuminer-opt], [3.9.2.3])
 
 AC_PREREQ([2.59c])
 AC_CANONICAL_SYSTEM

cpu-miner.c

@@ -2926,11 +2926,13 @@ void parse_arg(int key, char *arg )
 //		if ( ul > ( 1ULL << num_cpus ) - 1ULL )
 //			ul = -1LL;
 #if AFFINITY_USES_UINT128
-// replicate the low 64 bits to make a full 128 bit mask
+// replicate the low 64 bits to make a full 128 bit maski if there are more
-		opt_affinity = (uint128_t)(ul);
+// than 64 CPUs, otherwise zero extend the upper half.
-                opt_affinity = (opt_affinity << 64 ) | (uint128_t)ul;
+                opt_affinity = (uint128_t)ul;
+                if ( num_cpus > 64 )
+                   opt_affinity = (opt_affinity << 64 ) | (uint128_t)ul;
 #else
-                opt_affinity = ul;
+                   opt_affinity = ul;
 #endif
 		break;
 	case 1021:

interleave.h

@@ -43,8 +43,127 @@
 //
 // AVX512: 4x128, 8x64, 16x32
 //
-// Interleaving and deinterleaving is done in blocks of 16*16, 32*32,
+// Interleaving and deinterleaving is done in blocks of 8*8, 16*16, 32*32,
-// or 64*64 bytes for SSE2, AVX2 and AVX512 vectors respectively.
+// or 64*64 bytes for MMX, SSE2, AVX2 and AVX512 vectors respectively.
+
+//////////////////////////////////////////////////////
+// 
+//          MMX 64 bit vectors
+
+#define mm64_put_32( s0, s1 ) \
+  _mm_set_pi32( *((const uint32_t*)(s1)), *((const uint32_t*)(s0)) )
+
+#define mm64_get_32( s, i0, i1 ) \
+  _mm_set_pi32( ((const uint32_t*)(s))[i1], ((const uint32_t*)(s))[i0] )
+
+// 1 MMX block, 8 bytes * 2 lanes
+static inline void mm64_interleave_2x32( void *d, const void *s0,
+                                         const void *s1, int len )
+{
+  casti_m64( d, 0 ) = mm64_put_32( s0    , s1     );
+  casti_m64( d, 1 ) = mm64_put_32( s0+  4, s1+  4 );
+  casti_m64( d, 2 ) = mm64_put_32( s0+  8, s1+  8 );
+  casti_m64( d, 3 ) = mm64_put_32( s0+ 12, s1+ 12 );
+  casti_m64( d, 4 ) = mm64_put_32( s0+ 16, s1+ 16 );
+  casti_m64( d, 5 ) = mm64_put_32( s0+ 20, s1+ 20 );
+  casti_m64( d, 6 ) = mm64_put_32( s0+ 24, s1+ 24 );
+  casti_m64( d, 7 ) = mm64_put_32( s0+ 28, s1+ 28 );
+
+  if ( len <= 256 ) return;
+
+  casti_m64( d, 8 ) = mm64_put_32( s0+ 32, s1+ 32 );
+  casti_m64( d, 9 ) = mm64_put_32( s0+ 36, s1+ 36 );
+  casti_m64( d,10 ) = mm64_put_32( s0+ 40, s1+ 40 );
+  casti_m64( d,11 ) = mm64_put_32( s0+ 44, s1+ 44 );
+  casti_m64( d,12 ) = mm64_put_32( s0+ 48, s1+ 48 );
+  casti_m64( d,13 ) = mm64_put_32( s0+ 52, s1+ 52 );
+  casti_m64( d,14 ) = mm64_put_32( s0+ 56, s1+ 56 );
+  casti_m64( d,15 ) = mm64_put_32( s0+ 60, s1+ 60 );
+
+  if ( len <= 512 ) return;
+
+  casti_m64( d,16 ) = mm64_put_32( s0+ 64, s1+ 64 );
+  casti_m64( d,17 ) = mm64_put_32( s0+ 68, s1+ 68 );
+  casti_m64( d,18 ) = mm64_put_32( s0+ 72, s1+ 72 );
+  casti_m64( d,19 ) = mm64_put_32( s0+ 76, s1+ 76 );
+
+  if ( len <= 640 ) return;
+
+  casti_m64( d,20 ) = mm64_put_32( s0+ 80, s1+ 80 );
+  casti_m64( d,21 ) = mm64_put_32( s0+ 84, s1+ 84 );
+  casti_m64( d,22 ) = mm64_put_32( s0+ 88, s1+ 88 );
+  casti_m64( d,23 ) = mm64_put_32( s0+ 92, s1+ 92 );
+  casti_m64( d,24 ) = mm64_put_32( s0+ 96, s1+ 96 );
+  casti_m64( d,25 ) = mm64_put_32( s0+100, s1+100 );
+  casti_m64( d,26 ) = mm64_put_32( s0+104, s1+104 );
+  casti_m64( d,27 ) = mm64_put_32( s0+108, s1+108 );
+  casti_m64( d,28 ) = mm64_put_32( s0+112, s1+112 );
+  casti_m64( d,29 ) = mm64_put_32( s0+116, s1+116 );
+  casti_m64( d,30 ) = mm64_put_32( s0+120, s1+120 );
+  casti_m64( d,31 ) = mm64_put_32( s0+124, s1+124 );
+}
+
+static inline void mm64_deinterleave_2x32( void *d00, void *d01,
+                                    const int n, const void *s, int len )
+{
+   casti_m64( d00,0 ) = mm64_get_32( s,  0,  2 );
+   casti_m64( d01,0 ) = mm64_get_32( s,  1,  3 );
+   casti_m64( d00,1 ) = mm64_get_32( s,  4,  6 );
+   casti_m64( d01,1 ) = mm64_get_32( s,  5,  7 );
+   casti_m64( d00,2 ) = mm64_get_32( s,  8, 10 );
+   casti_m64( d01,2 ) = mm64_get_32( s,  9, 11 );
+   casti_m64( d00,3 ) = mm64_get_32( s, 12, 14 );
+   casti_m64( d01,3 ) = mm64_get_32( s, 13, 15 );
+
+   if ( len <= 256 ) return;
+
+   casti_m64( d00,4 ) = mm64_get_32( s, 16, 18 );
+   casti_m64( d01,4 ) = mm64_get_32( s, 17, 19 );
+   casti_m64( d00,5 ) = mm64_get_32( s, 20, 22 );
+   casti_m64( d01,5 ) = mm64_get_32( s, 21, 23 );
+   casti_m64( d00,6 ) = mm64_get_32( s, 24, 26 );
+   casti_m64( d01,6 ) = mm64_get_32( s, 25, 27 );
+   casti_m64( d00,7 ) = mm64_get_32( s, 28, 30 );
+   casti_m64( d01,7 ) = mm64_get_32( s, 29, 31 );
+
+   if ( len <= 512 ) return;
+
+   casti_m64( d00,8 ) = mm64_get_32( s, 32, 34 );
+   casti_m64( d01,8 ) = mm64_get_32( s, 33, 35 );
+   casti_m64( d00,9 ) = mm64_get_32( s, 36, 38 );
+   casti_m64( d01,9 ) = mm64_get_32( s, 37, 39 );
+
+   if ( len <= 640 ) return;
+
+   casti_m64( d00,10 ) = mm64_get_32( s, 40, 42 );
+   casti_m64( d01,10 ) = mm64_get_32( s, 41, 43 );
+   casti_m64( d00,11 ) = mm64_get_32( s, 44, 46 );
+   casti_m64( d01,11 ) = mm64_get_32( s, 45, 47 );
+   casti_m64( d00,12 ) = mm64_get_32( s, 48, 50 );
+   casti_m64( d01,12 ) = mm64_get_32( s, 49, 51 );
+   casti_m64( d00,13 ) = mm64_get_32( s, 52, 54 );
+   casti_m64( d01,13 ) = mm64_get_32( s, 53, 55 );
+   casti_m64( d00,14 ) = mm64_get_32( s, 56, 58 );
+   casti_m64( d01,14 ) = mm64_get_32( s, 57, 59 );
+   casti_m64( d00,15 ) = mm64_get_32( s, 60, 62 );
+   casti_m64( d01,15 ) = mm64_get_32( s, 61, 63 );
+}
+
+static inline void mm64_extract_lane_2x32( void *d, const void *s,
+                                         const int lane, const int bit_len )
+{
+  casti_m64( d, 0 ) = mm64_get_32( s, lane   , lane+ 4 );
+  casti_m64( d, 1 ) = mm64_get_32( s, lane+ 8, lane+12 );
+  casti_m64( d, 2 ) = mm64_get_32( s, lane+16, lane+20 );
+  casti_m64( d, 3 ) = mm64_get_32( s, lane+24, lane+28 );
+
+  if ( bit_len <= 256 ) return;
+  casti_m64( d, 4 ) = mm64_get_32( s, lane+32, lane+36 );
+  casti_m64( d, 5 ) = mm64_get_32( s, lane+40, lane+44 );
+  casti_m64( d, 6 ) = mm64_get_32( s, lane+48, lane+52 );
+  casti_m64( d, 7 ) = mm64_get_32( s, lane+56, lane+60 );
+  // bit_len == 512
+}
 
 
 ///////////////////////////////////////////////////////////////