v3.17.1

algo/groestl/aes_ni/groestl-intr-aes.h

@@ -67,11 +67,9 @@ static const __m128i SUBSH_MASK7 = { 0x06090c0f0205080b, 0x0e0104070a0d0003 };
  * xmm[j] will be lost
  * xmm[k] has to be all 0x1b */
 #define MUL2(i, j, k){\
-  j = _mm_xor_si128(j, j);\
-  j = _mm_cmpgt_epi8(j, i);\
+  j = _mm_cmpgt_epi8( m128_zero, i);\
   i = _mm_add_epi8(i, i);\
-  j = _mm_and_si128(j, k);\
-  i = _mm_xor_si128(i, j);\
+  i = mm128_xorand(i, j, k );\
 } 
 
  /**/
@@ -93,6 +91,96 @@ static const __m128i SUBSH_MASK7 = { 0x06090c0f0205080b, 0x0e0104070a0d0003 };
    We almost fit into 16 registers, need only 3 spills to memory.
    This implementation costs 7.7 c/b giving total speed on SNB: 10.7c/b.
    K. Matusiewicz, 2011/05/29 */
+
+#if defined(__AVX512VL__)
+
+#define MixBytes(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
+  /* t_i = a_i + a_{i+1} */\
+  b6 = a0;\
+  b7 = a1;\
+  a0 = _mm_xor_si128(a0, a1);\
+  b0 = a2;\
+  a1 = _mm_xor_si128(a1, a2);\
+  b1 = a3;\
+  TEMP2 = _mm_xor_si128(a2, a3);\
+  b2 = a4;\
+  a3 = _mm_xor_si128(a3, a4);\
+  b3 = a5;\
+  a4 = _mm_xor_si128(a4, a5);\
+  b4 = a6;\
+  a5 = _mm_xor_si128(a5, a6);\
+  b5 = a7;\
+  a6 = _mm_xor_si128(a6, a7);\
+  a7 = _mm_xor_si128(a7, b6);\
+   \
+  /* build y4 y5 y6 ... in regs xmm8, xmm9, xmm10 by adding t_i*/\
+  TEMP0 = mm128_xor3( b0, a4, a6 ); \
+  /* spill values y_4, y_5 to memory */\
+  TEMP1 = mm128_xor3( b1, a5, a7 );\
+  b2 = mm128_xor3( b2, a6, a0 ); \
+  /* save values t0, t1, t2 to xmm8, xmm9 and memory */\
+  b0 = a0;\
+  b3 = mm128_xor3( b3, a7, a1 ); \
+  b1 = a1;\
+  b6 = mm128_xor3( b6, a4, TEMP2 ); \
+  b4 = mm128_xor3( b4, a0, TEMP2 ); \
+  b7 = mm128_xor3( b7, a5, a3 ); \
+  b5 = mm128_xor3( b5, a1, a3 ); \
+  \
+  /* compute x_i = t_i + t_{i+3} */\
+  a0 = _mm_xor_si128(a0, a3);\
+  a1 = _mm_xor_si128(a1, a4);\
+  a2 = _mm_xor_si128(TEMP2, a5);\
+  a3 = _mm_xor_si128(a3, a6);\
+  a4 = _mm_xor_si128(a4, a7);\
+  a5 = _mm_xor_si128(a5, b0);\
+  a6 = _mm_xor_si128(a6, b1);\
+  a7 = _mm_xor_si128(a7, TEMP2);\
+  \
+  /* compute z_i : double x_i using temp xmm8 and 1B xmm9 */\
+  /* compute w_i : add y_{i+4} */\
+  b1 = m128_const1_64( 0x1b1b1b1b1b1b1b1b );\
+  MUL2(a0, b0, b1);\
+  a0 = _mm_xor_si128(a0, TEMP0);\
+  MUL2(a1, b0, b1);\
+  a1 = _mm_xor_si128(a1, TEMP1);\
+  MUL2(a2, b0, b1);\
+  a2 = _mm_xor_si128(a2, b2);\
+  MUL2(a3, b0, b1);\
+  a3 = _mm_xor_si128(a3, b3);\
+  MUL2(a4, b0, b1);\
+  a4 = _mm_xor_si128(a4, b4);\
+  MUL2(a5, b0, b1);\
+  a5 = _mm_xor_si128(a5, b5);\
+  MUL2(a6, b0, b1);\
+  a6 = _mm_xor_si128(a6, b6);\
+  MUL2(a7, b0, b1);\
+  a7 = _mm_xor_si128(a7, b7);\
+  \
+  /* compute v_i : double w_i      */\
+  /* add to y_4 y_5 .. v3, v4, ... */\
+  MUL2(a0, b0, b1);\
+  b5 = _mm_xor_si128(b5, a0);\
+  MUL2(a1, b0, b1);\
+  b6 = _mm_xor_si128(b6, a1);\
+  MUL2(a2, b0, b1);\
+  b7 = _mm_xor_si128(b7, a2);\
+  MUL2(a5, b0, b1);\
+  b2 = _mm_xor_si128(b2, a5);\
+  MUL2(a6, b0, b1);\
+  b3 = _mm_xor_si128(b3, a6);\
+  MUL2(a7, b0, b1);\
+  b4 = _mm_xor_si128(b4, a7);\
+  MUL2(a3, b0, b1);\
+  MUL2(a4, b0, b1);\
+  b0 = TEMP0;\
+  b1 = TEMP1;\
+  b0 = _mm_xor_si128(b0, a3);\
+  b1 = _mm_xor_si128(b1, a4);\
+}/*MixBytes*/
+
+#else
+
 #define MixBytes(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
   /* t_i = a_i + a_{i+1} */\
   b6 = a0;\
@@ -189,6 +277,8 @@ static const __m128i SUBSH_MASK7 = { 0x06090c0f0205080b, 0x0e0104070a0d0003 };
   b1 = _mm_xor_si128(b1, a4);\
 }/*MixBytes*/
 
+#endif
+
 
 /* one round
  * a0-a7 = input rows

algo/groestl/aes_ni/groestl256-intr-aes.h

@@ -58,11 +58,9 @@ static const __m128i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e };
  * xmm[j] will be lost
  * xmm[k] has to be all 0x1b */
 #define MUL2(i, j, k){\
-  j = _mm_xor_si128(j, j);\
-  j = _mm_cmpgt_epi8(j, i);\
+  j = _mm_cmpgt_epi8( m128_zero, i);\
   i = _mm_add_epi8(i, i);\
-  j = _mm_and_si128(j, k);\
-  i = _mm_xor_si128(i, j);\
+  i = mm128_xorand(i, j, k );\
 } 
 
 /* Yet another implementation of MixBytes.
@@ -82,6 +80,96 @@ static const __m128i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e };
    We almost fit into 16 registers, need only 3 spills to memory.
    This implementation costs 7.7 c/b giving total speed on SNB: 10.7c/b.
    K. Matusiewicz, 2011/05/29 */
+
+#if defined(__AVX512VL__)
+
+#define MixBytes(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
+  /* t_i = a_i + a_{i+1} */\
+  b6 = a0;\
+  b7 = a1;\
+  a0 = _mm_xor_si128(a0, a1);\
+  b0 = a2;\
+  a1 = _mm_xor_si128(a1, a2);\
+  b1 = a3;\
+  TEMP2 = _mm_xor_si128(a2, a3);\
+  b2 = a4;\
+  a3 = _mm_xor_si128(a3, a4);\
+  b3 = a5;\
+  a4 = _mm_xor_si128(a4, a5);\
+  b4 = a6;\
+  a5 = _mm_xor_si128(a5, a6);\
+  b5 = a7;\
+  a6 = _mm_xor_si128(a6, a7);\
+  a7 = _mm_xor_si128(a7, b6);\
+   \
+  /* build y4 y5 y6 ... in regs xmm8, xmm9, xmm10 by adding t_i*/\
+  TEMP0 = mm128_xor3( b0, a4, a6 ); \
+  /* spill values y_4, y_5 to memory */\
+  TEMP1 = mm128_xor3( b1, a5, a7 );\
+  b2 = mm128_xor3( b2, a6, a0 ); \
+  /* save values t0, t1, t2 to xmm8, xmm9 and memory */\
+  b0 = a0;\
+  b3 = mm128_xor3( b3, a7, a1 ); \
+  b1 = a1;\
+  b6 = mm128_xor3( b6, a4, TEMP2 ); \
+  b4 = mm128_xor3( b4, a0, TEMP2 ); \
+  b7 = mm128_xor3( b7, a5, a3 ); \
+  b5 = mm128_xor3( b5, a1, a3 ); \
+  \
+  /* compute x_i = t_i + t_{i+3} */\
+  a0 = _mm_xor_si128(a0, a3);\
+  a1 = _mm_xor_si128(a1, a4);\
+  a2 = _mm_xor_si128(TEMP2, a5);\
+  a3 = _mm_xor_si128(a3, a6);\
+  a4 = _mm_xor_si128(a4, a7);\
+  a5 = _mm_xor_si128(a5, b0);\
+  a6 = _mm_xor_si128(a6, b1);\
+  a7 = _mm_xor_si128(a7, TEMP2);\
+  \
+  /* compute z_i : double x_i using temp xmm8 and 1B xmm9 */\
+  /* compute w_i : add y_{i+4} */\
+  b1 = m128_const1_64( 0x1b1b1b1b1b1b1b1b );\
+  MUL2(a0, b0, b1);\
+  a0 = _mm_xor_si128(a0, TEMP0);\
+  MUL2(a1, b0, b1);\
+  a1 = _mm_xor_si128(a1, TEMP1);\
+  MUL2(a2, b0, b1);\
+  a2 = _mm_xor_si128(a2, b2);\
+  MUL2(a3, b0, b1);\
+  a3 = _mm_xor_si128(a3, b3);\
+  MUL2(a4, b0, b1);\
+  a4 = _mm_xor_si128(a4, b4);\
+  MUL2(a5, b0, b1);\
+  a5 = _mm_xor_si128(a5, b5);\
+  MUL2(a6, b0, b1);\
+  a6 = _mm_xor_si128(a6, b6);\
+  MUL2(a7, b0, b1);\
+  a7 = _mm_xor_si128(a7, b7);\
+  \
+  /* compute v_i : double w_i      */\
+  /* add to y_4 y_5 .. v3, v4, ... */\
+  MUL2(a0, b0, b1);\
+  b5 = _mm_xor_si128(b5, a0);\
+  MUL2(a1, b0, b1);\
+  b6 = _mm_xor_si128(b6, a1);\
+  MUL2(a2, b0, b1);\
+  b7 = _mm_xor_si128(b7, a2);\
+  MUL2(a5, b0, b1);\
+  b2 = _mm_xor_si128(b2, a5);\
+  MUL2(a6, b0, b1);\
+  b3 = _mm_xor_si128(b3, a6);\
+  MUL2(a7, b0, b1);\
+  b4 = _mm_xor_si128(b4, a7);\
+  MUL2(a3, b0, b1);\
+  MUL2(a4, b0, b1);\
+  b0 = TEMP0;\
+  b1 = TEMP1;\
+  b0 = _mm_xor_si128(b0, a3);\
+  b1 = _mm_xor_si128(b1, a4);\
+}/*MixBytes*/
+
+#else
+
 #define MixBytes(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
   /* t_i = a_i + a_{i+1} */\
   b6 = a0;\
@@ -178,6 +266,8 @@ static const __m128i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e };
   b1 = _mm_xor_si128(b1, a4);\
 }/*MixBytes*/
 
+#endif
+
 /* one round
  * i = round number
  * a0-a7 = input rows

algo/groestl/groestl256-intr-4way.h

@@ -96,11 +96,9 @@ static const __m512i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e,
  * xmm[j] will be lost
  * xmm[k] has to be all 0x1b */
 #define MUL2(i, j, k){\
-  j = _mm512_xor_si512(j, j);\
-  j = _mm512_movm_epi8( _mm512_cmpgt_epi8_mask(j, i) );\
+  j = _mm512_movm_epi8( _mm512_cmpgt_epi8_mask( m512_zero, i) );\
   i = _mm512_add_epi8(i, i);\
-  j = _mm512_and_si512(j, k);\
-  i = _mm512_xor_si512(i, j);\
+  i = mm512_xorand( i, j, k );\
 } 
 
 /* Yet another implementation of MixBytes.
@@ -120,6 +118,95 @@ static const __m512i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e,
    We almost fit into 16 registers, need only 3 spills to memory.
    This implementation costs 7.7 c/b giving total speed on SNB: 10.7c/b.
    K. Matusiewicz, 2011/05/29 */
+
+#define MixBytes( a0, a1, a2, a3, a4, a5, a6, a7, \
+                  b0, b1, b2, b3, b4, b5, b6, b7) { \
+  /* t_i = a_i + a_{i+1} */\
+  b6 = a0; \
+  b7 = a1; \
+  a0 = _mm512_xor_si512( a0, a1 ); \
+  b0 = a2; \
+  a1 = _mm512_xor_si512( a1, a2 ); \
+  b1 = a3; \
+  TEMP2 = _mm512_xor_si512( a2, a3 ); \
+  b2 = a4; \
+  a3 = _mm512_xor_si512( a3, a4 ); \
+  b3 = a5; \
+  a4 = _mm512_xor_si512( a4, a5 );\
+  b4 = a6; \
+  a5 = _mm512_xor_si512( a5, a6 ); \
+  b5 = a7; \
+  a6 = _mm512_xor_si512( a6, a7 ); \
+  a7 = _mm512_xor_si512( a7, b6 ); \
+  \
+  /* build y4 y5 y6 ... in regs xmm8, xmm9, xmm10 by adding t_i*/\
+  TEMP0 = mm512_xor3( b0, a4, a6 ); \
+  /* spill values y_4, y_5 to memory */\
+  TEMP1 = mm512_xor3( b1, a5, a7 ); \
+  b2 = mm512_xor3( b2, a6, a0 ); \
+  /* save values t0, t1, t2 to xmm8, xmm9 and memory */\
+  b0 = a0; \
+  b3 = mm512_xor3( b3, a7, a1 ); \
+  b1 = a1; \
+  b6 = mm512_xor3( b6, a4, TEMP2 ); \
+  b4 = mm512_xor3( b4, a0, TEMP2 ); \
+  b7 = mm512_xor3( b7, a5, a3 ); \
+  b5 = mm512_xor3( b5, a1, a3 ); \
+  \
+  /* compute x_i = t_i + t_{i+3} */\
+  a0 = _mm512_xor_si512( a0, a3 ); \
+  a1 = _mm512_xor_si512( a1, a4 ); \
+  a2 = _mm512_xor_si512( TEMP2, a5 ); \
+  a3 = _mm512_xor_si512( a3, a6 ); \
+  a4 = _mm512_xor_si512( a4, a7 ); \
+  a5 = _mm512_xor_si512( a5, b0 ); \
+  a6 = _mm512_xor_si512( a6, b1 ); \
+  a7 = _mm512_xor_si512( a7, TEMP2 ); \
+  \
+  /* compute z_i : double x_i using temp xmm8 and 1B xmm9 */\
+  /* compute w_i : add y_{i+4} */\
+  b1 = m512_const1_64( 0x1b1b1b1b1b1b1b1b ); \
+  MUL2( a0, b0, b1 ); \
+  a0 = _mm512_xor_si512( a0, TEMP0 ); \
+  MUL2( a1, b0, b1 ); \
+  a1 = _mm512_xor_si512( a1, TEMP1 ); \
+  MUL2( a2, b0, b1 ); \
+  a2 = _mm512_xor_si512( a2, b2 ); \
+  MUL2( a3, b0, b1 ); \
+  a3 = _mm512_xor_si512( a3, b3 ); \
+  MUL2( a4, b0, b1 ); \
+  a4 = _mm512_xor_si512( a4, b4 ); \
+  MUL2( a5, b0, b1 ); \
+  a5 = _mm512_xor_si512( a5, b5 ); \
+  MUL2( a6, b0, b1 ); \
+  a6 = _mm512_xor_si512( a6, b6 ); \
+  MUL2( a7, b0, b1 ); \
+  a7 = _mm512_xor_si512( a7, b7 ); \
+  \
+  /* compute v_i : double w_i      */\
+  /* add to y_4 y_5 .. v3, v4, ... */\
+  MUL2( a0, b0, b1 ); \
+  b5 = _mm512_xor_si512( b5, a0 ); \
+  MUL2( a1, b0, b1 ); \
+  b6 = _mm512_xor_si512( b6, a1 ); \
+  MUL2( a2, b0, b1 ); \
+  b7 = _mm512_xor_si512( b7, a2 ); \
+  MUL2( a5, b0, b1 ); \
+  b2 = _mm512_xor_si512( b2, a5 ); \
+  MUL2( a6, b0, b1 ); \
+  b3 = _mm512_xor_si512( b3, a6 ); \
+  MUL2( a7, b0, b1 ); \
+  b4 = _mm512_xor_si512( b4, a7 ); \
+  MUL2( a3, b0, b1 ); \
+  MUL2( a4, b0, b1 ); \
+  b0 = TEMP0;\
+  b1 = TEMP1;\
+  b0 = _mm512_xor_si512( b0, a3 ); \
+  b1 = _mm512_xor_si512( b1, a4 ); \
+}/*MixBytes*/
+
+
+#if 0
 #define MixBytes(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
   /* t_i = a_i + a_{i+1} */\
   b6 = a0;\
@@ -215,7 +302,7 @@ static const __m512i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e,
   b0 = _mm512_xor_si512(b0, a3);\
   b1 = _mm512_xor_si512(b1, a4);\
 }/*MixBytes*/
-
+#endif
 
 #define ROUND(i, a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
   /* AddRoundConstant */\

algo/groestl/groestl512-intr-4way.h

@@ -104,11 +104,9 @@ static const __m512i SUBSH_MASK7 = { 0x06090c0f0205080b, 0x0e0104070a0d0003,
  * xmm[j] will be lost
  * xmm[k] has to be all 0x1b */
 #define MUL2(i, j, k){\
-  j = _mm512_xor_si512(j, j);\
-  j = _mm512_movm_epi8( _mm512_cmpgt_epi8_mask(j, i) );\
+  j = _mm512_movm_epi8( _mm512_cmpgt_epi8_mask( m512_zero, i) );\
   i = _mm512_add_epi8(i, i);\
-  j = _mm512_and_si512(j, k);\
-  i = _mm512_xor_si512(i, j);\
+  i = mm512_xorand( i, j, k );\
 } 
 
  /**/
@@ -130,100 +128,90 @@ static const __m512i SUBSH_MASK7 = { 0x06090c0f0205080b, 0x0e0104070a0d0003,
    We almost fit into 16 registers, need only 3 spills to memory.
    This implementation costs 7.7 c/b giving total speed on SNB: 10.7c/b.
    K. Matusiewicz, 2011/05/29 */
-#define MixBytes(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
+#define MixBytes( a0, a1, a2, a3, a4, a5, a6, a7, \
+                  b0, b1, b2, b3, b4, b5, b6, b7) { \
   /* t_i = a_i + a_{i+1} */\
-  b6 = a0;\
-  b7 = a1;\
-  a0 = _mm512_xor_si512(a0, a1);\
-  b0 = a2;\
-  a1 = _mm512_xor_si512(a1, a2);\
-  b1 = a3;\
-  a2 = _mm512_xor_si512(a2, a3);\
-  b2 = a4;\
-  a3 = _mm512_xor_si512(a3, a4);\
-  b3 = a5;\
-  a4 = _mm512_xor_si512(a4, a5);\
-  b4 = a6;\
-  a5 = _mm512_xor_si512(a5, a6);\
-  b5 = a7;\
-  a6 = _mm512_xor_si512(a6, a7);\
-  a7 = _mm512_xor_si512(a7, b6);\
+  b6 = a0; \
+  b7 = a1; \
+  a0 = _mm512_xor_si512( a0, a1 ); \
+  b0 = a2; \
+  a1 = _mm512_xor_si512( a1, a2 ); \
+  b1 = a3; \
+  TEMP2 = _mm512_xor_si512( a2, a3 ); \
+  b2 = a4; \
+  a3 = _mm512_xor_si512( a3, a4 ); \
+  b3 = a5; \
+  a4 = _mm512_xor_si512( a4, a5 );\
+  b4 = a6; \
+  a5 = _mm512_xor_si512( a5, a6 ); \
+  b5 = a7; \
+  a6 = _mm512_xor_si512( a6, a7 ); \
+  a7 = _mm512_xor_si512( a7, b6 ); \
   \
   /* build y4 y5 y6 ... in regs xmm8, xmm9, xmm10 by adding t_i*/\
-  b0 = _mm512_xor_si512(b0, a4);\
-  b6 = _mm512_xor_si512(b6, a4);\
-  b1 = _mm512_xor_si512(b1, a5);\
-  b7 = _mm512_xor_si512(b7, a5);\
-  b2 = _mm512_xor_si512(b2, a6);\
-  b0 = _mm512_xor_si512(b0, a6);\
+  TEMP0 = mm512_xor3( b0, a4, a6 ); \
   /* spill values y_4, y_5 to memory */\
-  TEMP0 = b0;\
-  b3 = _mm512_xor_si512(b3, a7);\
-  b1 = _mm512_xor_si512(b1, a7);\
-  TEMP1 = b1;\
-  b4 = _mm512_xor_si512(b4, a0);\
-  b2 = _mm512_xor_si512(b2, a0);\
+  TEMP1 = mm512_xor3( b1, a5, a7 ); \
+  b2 = mm512_xor3( b2, a6, a0 ); \
   /* save values t0, t1, t2 to xmm8, xmm9 and memory */\
-  b0 = a0;\
-  b5 = _mm512_xor_si512(b5, a1);\
-  b3 = _mm512_xor_si512(b3, a1);\
-  b1 = a1;\
-  b6 = _mm512_xor_si512(b6, a2);\
-  b4 = _mm512_xor_si512(b4, a2);\
-  TEMP2 = a2;\
-  b7 = _mm512_xor_si512(b7, a3);\
-  b5 = _mm512_xor_si512(b5, a3);\
+  b0 = a0; \
+  b3 = mm512_xor3( b3, a7, a1 ); \
+  b1 = a1; \
+  b6 = mm512_xor3( b6, a4, TEMP2 ); \
+  b4 = mm512_xor3( b4, a0, TEMP2 ); \
+  b7 = mm512_xor3( b7, a5, a3 ); \
+  b5 = mm512_xor3( b5, a1, a3 ); \
   \
   /* compute x_i = t_i + t_{i+3} */\
-  a0 = _mm512_xor_si512(a0, a3);\
-  a1 = _mm512_xor_si512(a1, a4);\
-  a2 = _mm512_xor_si512(a2, a5);\
-  a3 = _mm512_xor_si512(a3, a6);\
-  a4 = _mm512_xor_si512(a4, a7);\
-  a5 = _mm512_xor_si512(a5, b0);\
-  a6 = _mm512_xor_si512(a6, b1);\
-  a7 = _mm512_xor_si512(a7, TEMP2);\
+  a0 = _mm512_xor_si512( a0, a3 ); \
+  a1 = _mm512_xor_si512( a1, a4 ); \
+  a2 = _mm512_xor_si512( TEMP2, a5 ); \
+  a3 = _mm512_xor_si512( a3, a6 ); \
+  a4 = _mm512_xor_si512( a4, a7 ); \
+  a5 = _mm512_xor_si512( a5, b0 ); \
+  a6 = _mm512_xor_si512( a6, b1 ); \
+  a7 = _mm512_xor_si512( a7, TEMP2 ); \
   \
   /* compute z_i : double x_i using temp xmm8 and 1B xmm9 */\
   /* compute w_i : add y_{i+4} */\
-  b1 = m512_const1_64( 0x1b1b1b1b1b1b1b1b );\
-  MUL2(a0, b0, b1);\
-  a0 = _mm512_xor_si512(a0, TEMP0);\
-  MUL2(a1, b0, b1);\
-  a1 = _mm512_xor_si512(a1, TEMP1);\
-  MUL2(a2, b0, b1);\
-  a2 = _mm512_xor_si512(a2, b2);\
-  MUL2(a3, b0, b1);\
-  a3 = _mm512_xor_si512(a3, b3);\
-  MUL2(a4, b0, b1);\
-  a4 = _mm512_xor_si512(a4, b4);\
-  MUL2(a5, b0, b1);\
-  a5 = _mm512_xor_si512(a5, b5);\
-  MUL2(a6, b0, b1);\
-  a6 = _mm512_xor_si512(a6, b6);\
-  MUL2(a7, b0, b1);\
-  a7 = _mm512_xor_si512(a7, b7);\
+  b1 = m512_const1_64( 0x1b1b1b1b1b1b1b1b ); \
+  MUL2( a0, b0, b1 ); \
+  a0 = _mm512_xor_si512( a0, TEMP0 ); \
+  MUL2( a1, b0, b1 ); \
+  a1 = _mm512_xor_si512( a1, TEMP1 ); \
+  MUL2( a2, b0, b1 ); \
+  a2 = _mm512_xor_si512( a2, b2 ); \
+  MUL2( a3, b0, b1 ); \
+  a3 = _mm512_xor_si512( a3, b3 ); \
+  MUL2( a4, b0, b1 ); \
+  a4 = _mm512_xor_si512( a4, b4 ); \
+  MUL2( a5, b0, b1 ); \
+  a5 = _mm512_xor_si512( a5, b5 ); \
+  MUL2( a6, b0, b1 ); \
+  a6 = _mm512_xor_si512( a6, b6 ); \
+  MUL2( a7, b0, b1 ); \
+  a7 = _mm512_xor_si512( a7, b7 ); \
   \
   /* compute v_i : double w_i      */\
   /* add to y_4 y_5 .. v3, v4, ... */\
-  MUL2(a0, b0, b1);\
-  b5 = _mm512_xor_si512(b5, a0);\
-  MUL2(a1, b0, b1);\
-  b6 = _mm512_xor_si512(b6, a1);\
-  MUL2(a2, b0, b1);\
-  b7 = _mm512_xor_si512(b7, a2);\
-  MUL2(a5, b0, b1);\
-  b2 = _mm512_xor_si512(b2, a5);\
-  MUL2(a6, b0, b1);\
-  b3 = _mm512_xor_si512(b3, a6);\
-  MUL2(a7, b0, b1);\
-  b4 = _mm512_xor_si512(b4, a7);\
-  MUL2(a3, b0, b1);\
-  MUL2(a4, b0, b1);\
+  MUL2( a0, b0, b1 ); \
+  b5 = _mm512_xor_si512( b5, a0 ); \
+  MUL2( a1, b0, b1 ); \
+  b6 = _mm512_xor_si512( b6, a1 ); \
+  MUL2( a2, b0, b1 ); \
+  b7 = _mm512_xor_si512( b7, a2 ); \
+  MUL2( a5, b0, b1 ); \
+  b2 = _mm512_xor_si512( b2, a5 ); \
+  MUL2( a6, b0, b1 ); \
+  b3 = _mm512_xor_si512( b3, a6 ); \
+  MUL2( a7, b0, b1 ); \
+  b4 = _mm512_xor_si512( b4, a7 ); \
+  MUL2( a3, b0, b1 ); \
+  MUL2( a4, b0, b1 ); \
   b0 = TEMP0;\
   b1 = TEMP1;\
-  b0 = _mm512_xor_si512(b0, a3);\
-  b1 = _mm512_xor_si512(b1, a4);\
+  b0 = _mm512_xor_si512( b0, a3 ); \
+  b1 = _mm512_xor_si512( b1, a4 ); \
 }/*MixBytes*/
 
 /* one round
@@ -709,11 +697,9 @@ static const __m256i SUBSH_MASK7_2WAY =
  * xmm[j] will be lost
  * xmm[k] has to be all 0x1b */
 #define MUL2_2WAY(i, j, k){\
-  j = _mm256_xor_si256(j, j);\
-  j = _mm256_cmpgt_epi8(j, i );\
+  j = _mm256_cmpgt_epi8( m256_zero, i );\
   i = _mm256_add_epi8(i, i);\
-  j = _mm256_and_si256(j, k);\
-  i = _mm256_xor_si256(i, j);\
+  i = mm256_xorand( i, j, k );\
 }
 
 #define MixBytes_2way(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\

algo/groestl/myrgr-4way.c

@@ -44,6 +44,7 @@ void myriad_8way_hash( void *output, const void *input )
 
      rintrlv_8x64_4x128( vhashA, vhashB, input, 640 );
      groestl512_4way_update_close( &ctx.groestl, vhashA, vhashA, 640 );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(groestl512_4way_context) );
      groestl512_4way_update_close( &ctx.groestl, vhashB, vhashB, 640 );
 
      uint32_t hash0[20] __attribute__ ((aligned (64)));
@@ -58,8 +59,6 @@ void myriad_8way_hash( void *output, const void *input )
 //     rintrlv_4x128_8x32( vhash, vhashA, vhashB, 512 );
      dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhashA );
      dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhashB );
-     intrlv_8x32_512( vhash, hash0, hash1, hash2, hash3, hash4, hash5,
-                       hash6, hash7 );
 
 #else
 
@@ -76,27 +75,27 @@ void myriad_8way_hash( void *output, const void *input )
                    hash4, hash5, hash6, hash7, input, 640 );
 
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
      update_and_final_groestl( &ctx.groestl, (char*)hash1, (char*)hash1, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
      update_and_final_groestl( &ctx.groestl, (char*)hash2, (char*)hash2, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
      update_and_final_groestl( &ctx.groestl, (char*)hash4, (char*)hash4, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
      update_and_final_groestl( &ctx.groestl, (char*)hash5, (char*)hash5, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
      update_and_final_groestl( &ctx.groestl, (char*)hash6, (char*)hash6, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
      update_and_final_groestl( &ctx.groestl, (char*)hash7, (char*)hash7, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
-
-     intrlv_8x32( vhash, hash0, hash1, hash2, hash3,
-                         hash4, hash5, hash6, hash7, 512 );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
 
 #endif
 
+     intrlv_8x32_512( vhash, hash0, hash1, hash2, hash3, hash4, hash5,
+                       hash6, hash7 );
+     
      sha256_8way_update( &ctx.sha, vhash, 64 );
      sha256_8way_close( &ctx.sha, output );
 }