v3.9.5.4

RELEASE_NOTES

@@ -38,6 +38,21 @@ supported.
 Change Log
 ----------
 
+v3.9.5.4
+
+Fixed sha256q AVX2 poor performance.
+Fixed skein2 buffer overflow and restored bswap-interleave optimization.
+More restructuring.
+
+v3.9.5.3
+
+Fix crash mining hodl with aes-sse42.
+More restructuring and share report tweaks.
+
+v3.9.5.2
+
+Revert bswap-interleave optimization for causing crashes on Windows.
+
 v3.9.5.1
 
 Fixed skein2 crash on Windows.

algo/blake/blake-4way.c

@@ -27,25 +27,19 @@ int scanhash_blake_4way( struct work *work, uint32_t max_nonce,
    uint32_t *ptarget = work->target;
    const uint32_t first_nonce = pdata[19];
    uint32_t HTarget = ptarget[7];
-   uint32_t _ALIGN(32) edata[20];
+   __m128i  *noncev = (__m128i*)vdata + 19;   // aligned
    uint32_t n = first_nonce;
    int thr_id = mythr->id;  // thr_id arg is deprecated
 
    if (opt_benchmark)
       HTarget = 0x7f;
 
-   // we need big endian data...
-   swab32_array( edata, pdata, 20 );
-   mm128_intrlv_4x32( vdata, edata, edata, edata, edata, 640 );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    blake256r14_4way_init( &blake_4w_ctx );
    blake256r14_4way( &blake_4w_ctx, vdata, 64 );
 
-   uint32_t *noncep = vdata + 76;   // 19*4
    do {
-      be32enc( noncep,    n   );
-      be32enc( noncep +1, n+1 );
-      be32enc( noncep +2, n+2 );
-      be32enc( noncep +3, n+3 );
+      *noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
 
       blakehash_4way( hash, vdata );
 
@@ -76,9 +70,9 @@ void blakehash_8way( void *state, const void *input )
      memcpy( &ctx, &blake_8w_ctx, sizeof ctx );
      blake256r14_8way( &ctx, input + (64<<3), 16 );
      blake256r14_8way_close( &ctx, vhash );
-     mm256_dintrlv_8x32( state,     state+ 32, state+ 64, state+ 96,
-                              state+128, state+160, state+192, state+224,
-                              vhash, 256 );
+     _dintrlv_8x32( state,     state+ 32, state+ 64, state+ 96,
+                    state+128, state+160, state+192, state+224,
+                    vhash, 256 );
 }
 
 int scanhash_blake_8way( struct work *work, uint32_t max_nonce,
@@ -90,32 +84,21 @@ int scanhash_blake_8way( struct work *work, uint32_t max_nonce,
    uint32_t *ptarget = work->target;
    const uint32_t first_nonce = pdata[19];
    uint32_t HTarget = ptarget[7];
-   uint32_t _ALIGN(32) edata[20];
    uint32_t n = first_nonce;
+   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
    int thr_id = mythr->id;  // thr_id arg is deprecated
 
    if (opt_benchmark)
       HTarget = 0x7f;
 
-   // we need big endian data...
-   swab32_array( edata, pdata, 20 );
-
-   mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
-                                 edata, edata, edata, edata, 640 );
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
 
    blake256r14_8way_init( &blake_8w_ctx );
    blake256r14_8way( &blake_8w_ctx, vdata, 64 );
 
-   uint32_t *noncep = vdata + 152;   // 19*8
    do {
-      be32enc( noncep,    n   );
-      be32enc( noncep +1, n+1 );
-      be32enc( noncep +2, n+2 );
-      be32enc( noncep +3, n+3 );
-      be32enc( noncep +4, n+4 );
-      be32enc( noncep +5, n+5 );
-      be32enc( noncep +6, n+6 );
-      be32enc( noncep +7, n+7 );
+      *noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
+                                                  n+3, n+2, n+1, n ) );
       pdata[19] = n;
 
       blakehash_8way( hash, vdata );

algo/blake/blake256-hash-4way.c

@@ -412,34 +412,16 @@ do { \
 	V5 = H5; \
 	V6 = H6; \
 	V7 = H7; \
-        V8 = _mm_xor_si128( S0, _mm_set_epi32( CS0, CS0, CS0, CS0 ) ); \
-        V9 = _mm_xor_si128( S1, _mm_set_epi32( CS1, CS1, CS1, CS1 ) ); \
-        VA = _mm_xor_si128( S2, _mm_set_epi32( CS2, CS2, CS2, CS2 ) ); \
-        VB = _mm_xor_si128( S3, _mm_set_epi32( CS3, CS3, CS3, CS3 ) ); \
-        VC = _mm_xor_si128( _mm_set_epi32( T0, T0, T0, T0 ), \
-                            _mm_set_epi32( CS4, CS4, CS4, CS4 ) ); \
-        VD = _mm_xor_si128( _mm_set_epi32( T0, T0, T0, T0 ), \
-                            _mm_set_epi32( CS5, CS5, CS5, CS5 ) ); \
-        VE = _mm_xor_si128( _mm_set_epi32( T1, T1, T1, T1 ) \
-                          , _mm_set_epi32( CS6, CS6, CS6, CS6 ) ); \
-        VF = _mm_xor_si128( _mm_set_epi32( T1, T1, T1, T1 ), \
-                            _mm_set_epi32( CS7, CS7, CS7, CS7 ) ); \
-	M[0x0] = mm128_bswap_32( *(buf +  0) ); \
-	M[0x1] = mm128_bswap_32( *(buf +  1) ); \
-	M[0x2] = mm128_bswap_32( *(buf +  2) ); \
-	M[0x3] = mm128_bswap_32( *(buf +  3) ); \
-	M[0x4] = mm128_bswap_32( *(buf +  4) ); \
-	M[0x5] = mm128_bswap_32( *(buf +  5) ); \
-	M[0x6] = mm128_bswap_32( *(buf +  6) ); \
-	M[0x7] = mm128_bswap_32( *(buf +  7) ); \
-	M[0x8] = mm128_bswap_32( *(buf +  8) ); \
-	M[0x9] = mm128_bswap_32( *(buf +  9) ); \
-	M[0xA] = mm128_bswap_32( *(buf + 10) ); \
-	M[0xB] = mm128_bswap_32( *(buf + 11) ); \
-	M[0xC] = mm128_bswap_32( *(buf + 12) ); \
-	M[0xD] = mm128_bswap_32( *(buf + 13) ); \
-	M[0xE] = mm128_bswap_32( *(buf + 14) ); \
-	M[0xF] = mm128_bswap_32( *(buf + 15) ); \
+   V8 = _mm_xor_si128( S0, _mm_set1_epi32( CS0 ) ); \
+   V9 = _mm_xor_si128( S1, _mm_set1_epi32( CS1 ) ); \
+   VA = _mm_xor_si128( S2, _mm_set1_epi32( CS2 ) ); \
+   VB = _mm_xor_si128( S3, _mm_set1_epi32( CS3 ) ); \
+   VC = _mm_xor_si128( _mm_set1_epi32( T0 ), _mm_set1_epi32( CS4 ) ); \
+   VD = _mm_xor_si128( _mm_set1_epi32( T0 ), _mm_set1_epi32( CS5 ) ); \
+   VE = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS6 ) ); \
+   VF = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS7 ) ); \
+   mm128_block_bswap_32( M, buf ); \
+   mm128_block_bswap_32( M+8, buf+8 ); \
 	for (r = 0; r < rounds; r ++) \
 		ROUND_S_4WAY(r); \
         H0 = _mm_xor_si128( _mm_xor_si128( \
@@ -464,6 +446,54 @@ do { \
 
 // current impl
 
+#if defined(__SSSE3__)
+
+#define BLAKE256_4WAY_BLOCK_BSWAP32 do \
+{ \
+   __m128i shuf_bswap32 = _mm_set_epi64x( 0x0c0d0e0f08090a0b, \
+                                          0x0405060700010203 ); \
+   M0 = _mm_shuffle_epi8( buf[ 0], shuf_bswap32 ); \
+   M1 = _mm_shuffle_epi8( buf[ 1], shuf_bswap32 ); \
+   M2 = _mm_shuffle_epi8( buf[ 2], shuf_bswap32 ); \
+   M3 = _mm_shuffle_epi8( buf[ 3], shuf_bswap32 ); \
+   M4 = _mm_shuffle_epi8( buf[ 4], shuf_bswap32 ); \
+   M5 = _mm_shuffle_epi8( buf[ 5], shuf_bswap32 ); \
+   M6 = _mm_shuffle_epi8( buf[ 6], shuf_bswap32 ); \
+   M7 = _mm_shuffle_epi8( buf[ 7], shuf_bswap32 ); \
+   M8 = _mm_shuffle_epi8( buf[ 8], shuf_bswap32 ); \
+   M9 = _mm_shuffle_epi8( buf[ 9], shuf_bswap32 ); \
+   MA = _mm_shuffle_epi8( buf[10], shuf_bswap32 ); \
+   MB = _mm_shuffle_epi8( buf[11], shuf_bswap32 ); \
+   MC = _mm_shuffle_epi8( buf[12], shuf_bswap32 ); \
+   MD = _mm_shuffle_epi8( buf[13], shuf_bswap32 ); \
+   ME = _mm_shuffle_epi8( buf[14], shuf_bswap32 ); \
+   MF = _mm_shuffle_epi8( buf[15], shuf_bswap32 ); \
+} while(0)
+
+#else  // SSE2
+
+#define BLAKE256_4WAY_BLOCK_BSWAP32 do \
+{ \
+   M0 = mm128_bswap_32( buf[0] ); \
+   M1 = mm128_bswap_32( buf[1] ); \
+   M2 = mm128_bswap_32( buf[2] ); \
+   M3 = mm128_bswap_32( buf[3] ); \
+   M4 = mm128_bswap_32( buf[4] ); \
+   M5 = mm128_bswap_32( buf[5] ); \
+   M6 = mm128_bswap_32( buf[6] ); \
+   M7 = mm128_bswap_32( buf[7] ); \
+   M8 = mm128_bswap_32( buf[8] ); \
+   M9 = mm128_bswap_32( buf[9] ); \
+   MA = mm128_bswap_32( buf[10] ); \
+   MB = mm128_bswap_32( buf[11] ); \
+   MC = mm128_bswap_32( buf[12] ); \
+   MD = mm128_bswap_32( buf[13] ); \
+   ME = mm128_bswap_32( buf[14] ); \
+   MF = mm128_bswap_32( buf[15] ); \
+} while(0)
+
+#endif  // SSSE3 else SSE2
+
 #define COMPRESS32_4WAY( rounds ) \
 do { \
    __m128i M0, M1, M2, M3, M4, M5, M6, M7; \
@@ -486,22 +516,7 @@ do { \
    VD = _mm_xor_si128( _mm_set1_epi32( T0 ), _mm_set1_epi32( CS5 ) ); \
    VE = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS6 ) ); \
    VF = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS7 ) ); \
-   M0 = mm128_bswap_32( buf[ 0] ); \
-   M1 = mm128_bswap_32( buf[ 1] ); \
-   M2 = mm128_bswap_32( buf[ 2] ); \
-   M3 = mm128_bswap_32( buf[ 3] ); \
-   M4 = mm128_bswap_32( buf[ 4] ); \
-   M5 = mm128_bswap_32( buf[ 5] ); \
-   M6 = mm128_bswap_32( buf[ 6] ); \
-   M7 = mm128_bswap_32( buf[ 7] ); \
-   M8 = mm128_bswap_32( buf[ 8] ); \
-   M9 = mm128_bswap_32( buf[ 9] ); \
-   MA = mm128_bswap_32( buf[10] ); \
-   MB = mm128_bswap_32( buf[11] ); \
-   MC = mm128_bswap_32( buf[12] ); \
-   MD = mm128_bswap_32( buf[13] ); \
-   ME = mm128_bswap_32( buf[14] ); \
-   MF = mm128_bswap_32( buf[15] ); \
+   BLAKE256_4WAY_BLOCK_BSWAP32; \
    ROUND_S_4WAY(0); \
    ROUND_S_4WAY(1); \
    ROUND_S_4WAY(2); \
@@ -519,14 +534,14 @@ do { \
       ROUND_S_4WAY(2); \
       ROUND_S_4WAY(3); \
    } \
-   H0 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( V8, V0 ), S0 ), H0 ); \
-   H1 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( V9, V1 ), S1 ), H1 ); \
-   H2 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VA, V2 ), S2 ), H2 ); \
-   H3 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VB, V3 ), S3 ), H3 ); \
-   H4 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VC, V4 ), S0 ), H4 ); \
-   H5 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VD, V5 ), S1 ), H5 ); \
-   H6 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VE, V6 ), S2 ), H6 ); \
-   H7 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VF, V7 ), S3 ), H7 ); \
+   H0 = mm128_xor4( V8, V0, S0, H0 ); \
+   H1 = mm128_xor4( V9, V1, S1, H1 ); \
+   H2 = mm128_xor4( VA, V2, S2, H2 ); \
+   H3 = mm128_xor4( VB, V3, S3, H3 ); \
+   H4 = mm128_xor4( VC, V4, S0, H4 ); \
+   H5 = mm128_xor4( VD, V5, S1, H5 ); \
+   H6 = mm128_xor4( VE, V6, S2, H6 ); \
+   H7 = mm128_xor4( VF, V7, S3, H7 ); \
 } while (0)
 
 #endif
@@ -607,6 +622,7 @@ do { \
    __m256i M8, M9, MA, MB, MC, MD, ME, MF; \
    __m256i V0, V1, V2, V3, V4, V5, V6, V7; \
    __m256i V8, V9, VA, VB, VC, VD, VE, VF; \
+   __m256i shuf_bswap32; \
    V0 = H0; \
    V1 = H1; \
    V2 = H2; \
@@ -623,22 +639,24 @@ do { \
    VD = _mm256_xor_si256( _mm256_set1_epi32( T0 ), _mm256_set1_epi32( CS5 ) ); \
    VE = _mm256_xor_si256( _mm256_set1_epi32( T1 ), _mm256_set1_epi32( CS6 ) ); \
    VF = _mm256_xor_si256( _mm256_set1_epi32( T1 ), _mm256_set1_epi32( CS7 ) ); \
-   M0 = mm256_bswap_32( * buf ); \
-   M1 = mm256_bswap_32( *(buf+1) ); \
-   M2 = mm256_bswap_32( *(buf+2) ); \
-   M3 = mm256_bswap_32( *(buf+3) ); \
-   M4 = mm256_bswap_32( *(buf+4) ); \
-   M5 = mm256_bswap_32( *(buf+5) ); \
-   M6 = mm256_bswap_32( *(buf+6) ); \
-   M7 = mm256_bswap_32( *(buf+7) ); \
-   M8 = mm256_bswap_32( *(buf+8) ); \
-   M9 = mm256_bswap_32( *(buf+9) ); \
-   MA = mm256_bswap_32( *(buf+10) ); \
-   MB = mm256_bswap_32( *(buf+11) ); \
-   MC = mm256_bswap_32( *(buf+12) ); \
-   MD = mm256_bswap_32( *(buf+13) ); \
-   ME = mm256_bswap_32( *(buf+14) ); \
-   MF = mm256_bswap_32( *(buf+15) ); \
+   shuf_bswap32 = _mm256_set_epi64x( 0x0c0d0e0f08090a0b, 0x0405060700010203, \
+                                     0x0c0d0e0f08090a0b, 0x0405060700010203 ); \
+   M0 = _mm256_shuffle_epi8( * buf    , shuf_bswap32 ); \
+   M1 = _mm256_shuffle_epi8( *(buf+ 1), shuf_bswap32 ); \
+   M2 = _mm256_shuffle_epi8( *(buf+ 2), shuf_bswap32 ); \
+   M3 = _mm256_shuffle_epi8( *(buf+ 3), shuf_bswap32 ); \
+   M4 = _mm256_shuffle_epi8( *(buf+ 4), shuf_bswap32 ); \
+   M5 = _mm256_shuffle_epi8( *(buf+ 5), shuf_bswap32 ); \
+   M6 = _mm256_shuffle_epi8( *(buf+ 6), shuf_bswap32 ); \
+   M7 = _mm256_shuffle_epi8( *(buf+ 7), shuf_bswap32 ); \
+   M8 = _mm256_shuffle_epi8( *(buf+ 8), shuf_bswap32 ); \
+   M9 = _mm256_shuffle_epi8( *(buf+ 9), shuf_bswap32 ); \
+   MA = _mm256_shuffle_epi8( *(buf+10), shuf_bswap32 ); \
+   MB = _mm256_shuffle_epi8( *(buf+11), shuf_bswap32 ); \
+   MC = _mm256_shuffle_epi8( *(buf+12), shuf_bswap32 ); \
+   MD = _mm256_shuffle_epi8( *(buf+13), shuf_bswap32 ); \
+   ME = _mm256_shuffle_epi8( *(buf+14), shuf_bswap32 ); \
+   MF = _mm256_shuffle_epi8( *(buf+15), shuf_bswap32 ); \
    ROUND_S_8WAY(0); \
    ROUND_S_8WAY(1); \
    ROUND_S_8WAY(2); \
@@ -656,22 +674,14 @@ do { \
       ROUND_S_8WAY(2); \
       ROUND_S_8WAY(3); \
    } \
-   H0 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( V8, V0 ), \
-                                                              S0 ), H0 ); \
-   H1 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( V9, V1 ), \
-                                                              S1 ), H1 ); \
-   H2 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VA, V2 ), \
-                                                              S2 ), H2 ); \
-   H3 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VB, V3 ), \
-                                                              S3 ), H3 ); \
-   H4 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VC, V4 ), \
-                                                              S0 ), H4 ); \
-   H5 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VD, V5 ), \
-                                                              S1 ), H5 ); \
-   H6 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VE, V6 ), \
-                                                              S2 ), H6 ); \
-   H7 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VF, V7 ), \
-                                                              S3 ), H7 ); \
+   H0 = mm256_xor4( V8, V0, S0, H0 ); \
+   H1 = mm256_xor4( V9, V1, S1, H1 ); \
+   H2 = mm256_xor4( VA, V2, S2, H2 ); \
+   H3 = mm256_xor4( VB, V3, S3, H3 ); \
+   H4 = mm256_xor4( VC, V4, S0, H4 ); \
+   H5 = mm256_xor4( VD, V5, S1, H5 ); \
+   H6 = mm256_xor4( VE, V6, S2, H6 ); \
+   H7 = mm256_xor4( VF, V7, S3, H7 ); \
 } while (0)
 
 
@@ -685,6 +695,7 @@ static void
 blake32_4way_init( blake_4way_small_context *ctx, const uint32_t *iv,
                    const uint32_t *salt, int rounds )
 {
+   __m128i zero = m128_zero;
    casti_m128i( ctx->H, 0 ) = _mm_set1_epi32( iv[0] );
    casti_m128i( ctx->H, 1 ) = _mm_set1_epi32( iv[1] );
    casti_m128i( ctx->H, 2 ) = _mm_set1_epi32( iv[2] );
@@ -694,16 +705,10 @@ blake32_4way_init( blake_4way_small_context *ctx, const uint32_t *iv,
    casti_m128i( ctx->H, 6 ) = _mm_set1_epi32( iv[6] );
    casti_m128i( ctx->H, 7 ) = _mm_set1_epi32( iv[7] );
 
-   casti_m128i( ctx->S, 0 ) = m128_zero;
-   casti_m128i( ctx->S, 1 ) = m128_zero;
-   casti_m128i( ctx->S, 2 ) = m128_zero;
-   casti_m128i( ctx->S, 3 ) = m128_zero;
-/*
-   sc->S[0] = _mm_set1_epi32( salt[0] );
-   sc->S[1] = _mm_set1_epi32( salt[1] );
-   sc->S[2] = _mm_set1_epi32( salt[2] );
-   sc->S[3] = _mm_set1_epi32( salt[3] );
-*/
+   casti_m128i( ctx->S, 0 ) = zero;
+   casti_m128i( ctx->S, 1 ) = zero;
+   casti_m128i( ctx->S, 2 ) = zero;
+   casti_m128i( ctx->S, 3 ) = zero;
    ctx->T0 = ctx->T1 = 0;
    ctx->ptr = 0;
    ctx->rounds = rounds;
@@ -796,14 +801,7 @@ blake32_4way_close( blake_4way_small_context *ctx, unsigned ub, unsigned n,
       blake32_4way( ctx, buf, 64 );
    }
 
-   casti_m128i( dst, 0 ) = mm128_bswap_32( casti_m128i( ctx->H, 0 ) );
-   casti_m128i( dst, 1 ) = mm128_bswap_32( casti_m128i( ctx->H, 1 ) );
-   casti_m128i( dst, 2 ) = mm128_bswap_32( casti_m128i( ctx->H, 2 ) );
-   casti_m128i( dst, 3 ) = mm128_bswap_32( casti_m128i( ctx->H, 3 ) );
-   casti_m128i( dst, 4 ) = mm128_bswap_32( casti_m128i( ctx->H, 4 ) );
-   casti_m128i( dst, 5 ) = mm128_bswap_32( casti_m128i( ctx->H, 5 ) );
-   casti_m128i( dst, 6 ) = mm128_bswap_32( casti_m128i( ctx->H, 6 ) );
-   casti_m128i( dst, 7 ) = mm128_bswap_32( casti_m128i( ctx->H, 7 ) );
+   mm128_block_bswap_32( (__m128i*)dst, (__m128i*)ctx->H );
 }
 
 #if defined (__AVX2__)
@@ -816,11 +814,21 @@ static void
 blake32_8way_init( blake_8way_small_context *sc, const sph_u32 *iv,
                    const sph_u32 *salt, int rounds )
 {
-   int i;
-   for ( i = 0; i < 8; i++ )
-      sc->H[i] = _mm256_set1_epi32( iv[i] );
-   for ( i = 0; i < 4; i++ )
-      sc->S[i] = _mm256_set1_epi32( salt[i] );
+   __m256i zero = m256_zero;
+   casti_m256i( sc->H, 0 ) = _mm256_set1_epi32( iv[0] );
+   casti_m256i( sc->H, 1 ) = _mm256_set1_epi32( iv[1] );
+   casti_m256i( sc->H, 2 ) = _mm256_set1_epi32( iv[2] );
+   casti_m256i( sc->H, 3 ) = _mm256_set1_epi32( iv[3] );
+   casti_m256i( sc->H, 4 ) = _mm256_set1_epi32( iv[4] );
+   casti_m256i( sc->H, 5 ) = _mm256_set1_epi32( iv[5] );
+   casti_m256i( sc->H, 6 ) = _mm256_set1_epi32( iv[6] );
+   casti_m256i( sc->H, 7 ) = _mm256_set1_epi32( iv[7] );
+
+   casti_m256i( sc->S, 0 ) = zero;
+   casti_m256i( sc->S, 1 ) = zero;
+   casti_m256i( sc->S, 2 ) = zero;
+   casti_m256i( sc->S, 3 ) = zero;
+
    sc->T0 = sc->T1 = 0;
    sc->ptr = 0;
    sc->rounds = rounds;
@@ -872,14 +880,10 @@ static void
 blake32_8way_close( blake_8way_small_context *sc, unsigned ub, unsigned n,
                     void *dst, size_t out_size_w32 )
 {
-//   union {
-        __m256i buf[16];
-//        sph_u32 dummy;
-//   } u;
-   size_t ptr, k;
+   __m256i buf[16];
+   size_t ptr;
    unsigned bit_len;
    sph_u32 th, tl;
-   __m256i *out;
 
    ptr = sc->ptr;
    bit_len = ((unsigned)ptr << 3);
@@ -923,9 +927,7 @@ blake32_8way_close( blake_8way_small_context *sc, unsigned ub, unsigned n,
         *(buf+(60>>2)) = mm256_bswap_32( _mm256_set1_epi32( tl ) );
         blake32_8way( sc, buf, 64 );
    }
-   out = (__m256i*)dst;
-   for ( k = 0; k < out_size_w32; k++ )
-        out[k] = mm256_bswap_32( sc->H[k] );
+   mm256_block_bswap_32( (__m256i*)dst, (__m256i*)sc->H );
 }
 
 #endif

algo/blake/blake2s-4way.c

@@ -16,9 +16,9 @@ void blake2s_8way_hash( void *output, const void *input )
    blake2s_8way_update( &ctx, input + (64<<3), 16 );
    blake2s_8way_final( &ctx, vhash, BLAKE2S_OUTBYTES );
 
-   mm256_dintrlv_8x32( output,     output+ 32, output+ 64, output+ 96,
-                            output+128, output+160, output+192, output+224,
-                            vhash, 256 );
+   dintrlv_8x32( output,     output+ 32, output+ 64, output+ 96,
+                 output+128, output+160, output+192, output+224,
+                 vhash, 256 );
 }
 
 int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
@@ -28,28 +28,19 @@ int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
    uint32_t hash[8*8] __attribute__ ((aligned (32)));
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
-   uint32_t _ALIGN(64) edata[20];
    const uint32_t Htarg = ptarget[7];
    const uint32_t first_nonce = pdata[19];
+   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
    uint32_t n = first_nonce;
-   uint32_t *noncep = vdata + 152;   // 19*8
    int thr_id = mythr->id;  // thr_id arg is deprecated
 
-   swab32_array( edata, pdata, 20 );
-   mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
-                                 edata, edata, edata, edata, 640 );
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
    blake2s_8way_init( &blake2s_8w_ctx, BLAKE2S_OUTBYTES );
    blake2s_8way_update( &blake2s_8w_ctx, vdata, 64 );
 
    do {
-      be32enc( noncep,    n   );
-      be32enc( noncep +1, n+1 );
-      be32enc( noncep +2, n+2 );
-      be32enc( noncep +3, n+3 );
-      be32enc( noncep +4, n+4 );
-      be32enc( noncep +5, n+5 );
-      be32enc( noncep +6, n+6 );
-      be32enc( noncep +7, n+7 );
+      *noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
+                                                  n+3, n+2, n+1, n ) );
       pdata[19] = n;
 
       blake2s_8way_hash( hash, vdata );
@@ -94,23 +85,18 @@ int scanhash_blake2s_4way( struct work *work, uint32_t max_nonce,
    uint32_t hash[8*4] __attribute__ ((aligned (32)));
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
-   uint32_t _ALIGN(64) edata[20];
    const uint32_t Htarg = ptarget[7];
    const uint32_t first_nonce = pdata[19];
+   __m128i  *noncev = (__m128i*)vdata + 19;   // aligned
    uint32_t n = first_nonce;
-   uint32_t *noncep = vdata + 76;   // 19*4
    int thr_id = mythr->id;  // thr_id arg is deprecated
 
-   swab32_array( edata, pdata, 20 );
-   mm128_intrlv_4x32( vdata, edata, edata, edata, edata, 640 );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    blake2s_4way_init( &blake2s_4w_ctx, BLAKE2S_OUTBYTES );
    blake2s_4way_update( &blake2s_4w_ctx, vdata, 64 );
 
    do {
-      be32enc( noncep,    n   );
-      be32enc( noncep +1, n+1 );
-      be32enc( noncep +2, n+2 );
-      be32enc( noncep +3, n+3 );
+      *noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
       pdata[19] = n;
 
       blake2s_4way_hash( hash, vdata );

algo/blake/blake512-hash-4way.c

@@ -412,18 +412,18 @@ static const sph_u64 CB[16] = {
 	V5 = H5; \
 	V6 = H6; \
 	V7 = H7; \
-        V8 = _mm256_xor_si256( S0, _mm256_set_epi64x( CB0, CB0, CB0, CB0 ) ); \
-        V9 = _mm256_xor_si256( S1, _mm256_set_epi64x( CB1, CB1, CB1, CB1 ) ); \
-        VA = _mm256_xor_si256( S2, _mm256_set_epi64x( CB2, CB2, CB2, CB2 ) ); \
-        VB = _mm256_xor_si256( S3, _mm256_set_epi64x( CB3, CB3, CB3, CB3 ) ); \
-        VC = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
-                               _mm256_set_epi64x( CB4, CB4, CB4, CB4 ) ); \
-        VD = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
-                               _mm256_set_epi64x( CB5, CB5, CB5, CB5 ) ); \
-        VE = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
-                               _mm256_set_epi64x( CB6, CB6, CB6, CB6 ) ); \
-        VF = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
-                               _mm256_set_epi64x( CB7, CB7, CB7, CB7 ) ); \
+   V8 = _mm256_xor_si256( S0, _mm256_set_epi64x( CB0, CB0, CB0, CB0 ) ); \
+   V9 = _mm256_xor_si256( S1, _mm256_set_epi64x( CB1, CB1, CB1, CB1 ) ); \
+   VA = _mm256_xor_si256( S2, _mm256_set_epi64x( CB2, CB2, CB2, CB2 ) ); \
+   VB = _mm256_xor_si256( S3, _mm256_set_epi64x( CB3, CB3, CB3, CB3 ) ); \
+   VC = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
+                          _mm256_set_epi64x( CB4, CB4, CB4, CB4 ) ); \
+   VD = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
+                          _mm256_set_epi64x( CB5, CB5, CB5, CB5 ) ); \
+   VE = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
+                          _mm256_set_epi64x( CB6, CB6, CB6, CB6 ) ); \
+   VF = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
+                          _mm256_set_epi64x( CB7, CB7, CB7, CB7 ) ); \
 	M[0x0] = mm256_bswap_64( *(buf+0) ); \
 	M[0x1] = mm256_bswap_64( *(buf+1) ); \
 	M[0x2] = mm256_bswap_64( *(buf+2) ); \
@@ -464,80 +464,76 @@ static const sph_u64 CB[16] = {
 
 //current impl
 
-#define COMPRESS64_4WAY   do { \
-     __m256i M0, M1, M2, M3, M4, M5, M6, M7; \
-     __m256i M8, M9, MA, MB, MC, MD, ME, MF; \
-     __m256i V0, V1, V2, V3, V4, V5, V6, V7; \
-     __m256i V8, V9, VA, VB, VC, VD, VE, VF; \
-     V0 = H0; \
-     V1 = H1; \
-     V2 = H2; \
-     V3 = H3; \
-     V4 = H4; \
-     V5 = H5; \
-     V6 = H6; \
-     V7 = H7; \
-     V8 = _mm256_xor_si256( S0, _mm256_set_epi64x( CB0, CB0, CB0, CB0 ) );  \
-     V9 = _mm256_xor_si256( S1, _mm256_set_epi64x( CB1, CB1, CB1, CB1 ) );  \
-     VA = _mm256_xor_si256( S2, _mm256_set_epi64x( CB2, CB2, CB2, CB2 ) );  \
-     VB = _mm256_xor_si256( S3, _mm256_set_epi64x( CB3, CB3, CB3, CB3 ) );  \
-     VC = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
-                            _mm256_set_epi64x( CB4, CB4, CB4, CB4 ) );  \
-     VD = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
-                            _mm256_set_epi64x( CB5, CB5, CB5, CB5 ) );  \
-     VE = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
-                            _mm256_set_epi64x( CB6, CB6, CB6, CB6 ) );  \
-     VF = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
-                            _mm256_set_epi64x( CB7, CB7, CB7, CB7 ) );  \
-     M0 = mm256_bswap_64( *(buf + 0) ); \
-     M1 = mm256_bswap_64( *(buf + 1) ); \
-     M2 = mm256_bswap_64( *(buf + 2) ); \
-     M3 = mm256_bswap_64( *(buf + 3) ); \
-     M4 = mm256_bswap_64( *(buf + 4) ); \
-     M5 = mm256_bswap_64( *(buf + 5) ); \
-     M6 = mm256_bswap_64( *(buf + 6) ); \
-     M7 = mm256_bswap_64( *(buf + 7) ); \
-     M8 = mm256_bswap_64( *(buf + 8) ); \
-     M9 = mm256_bswap_64( *(buf + 9) ); \
-     MA = mm256_bswap_64( *(buf + 10) ); \
-     MB = mm256_bswap_64( *(buf + 11) ); \
-     MC = mm256_bswap_64( *(buf + 12) ); \
-     MD = mm256_bswap_64( *(buf + 13) ); \
-     ME = mm256_bswap_64( *(buf + 14) ); \
-     MF = mm256_bswap_64( *(buf + 15) ); \
-     ROUND_B_4WAY(0); \
-     ROUND_B_4WAY(1); \
-     ROUND_B_4WAY(2); \
-     ROUND_B_4WAY(3); \
-     ROUND_B_4WAY(4); \
-     ROUND_B_4WAY(5); \
-     ROUND_B_4WAY(6); \
-     ROUND_B_4WAY(7); \
-     ROUND_B_4WAY(8); \
-     ROUND_B_4WAY(9); \
-     ROUND_B_4WAY(0); \
-     ROUND_B_4WAY(1); \
-     ROUND_B_4WAY(2); \
-     ROUND_B_4WAY(3); \
-     ROUND_B_4WAY(4); \
-     ROUND_B_4WAY(5); \
-     H0 = _mm256_xor_si256( _mm256_xor_si256( \
-                            _mm256_xor_si256( S0, V0 ), V8 ), H0 ); \
-     H1 = _mm256_xor_si256( _mm256_xor_si256( \
-                            _mm256_xor_si256( S1, V1 ), V9 ), H1 ); \
-     H2 = _mm256_xor_si256( _mm256_xor_si256( \
-                            _mm256_xor_si256( S2, V2 ), VA ), H2 ); \
-     H3 = _mm256_xor_si256( _mm256_xor_si256( \
-                            _mm256_xor_si256( S3, V3 ), VB ), H3 ); \
-     H4 = _mm256_xor_si256( _mm256_xor_si256( \
-                            _mm256_xor_si256( S0, V4 ), VC ), H4 ); \
-     H5 = _mm256_xor_si256( _mm256_xor_si256( \
-                            _mm256_xor_si256( S1, V5 ), VD ), H5 ); \
-     H6 = _mm256_xor_si256( _mm256_xor_si256( \
-                            _mm256_xor_si256( S2, V6 ), VE ), H6 ); \
-     H7 = _mm256_xor_si256( _mm256_xor_si256( \
-                            _mm256_xor_si256( S3, V7 ), VF ), H7 ); \
-	} while (0)
+#define COMPRESS64_4WAY   do \
+{ \
+  __m256i M0, M1, M2, M3, M4, M5, M6, M7; \
+  __m256i M8, M9, MA, MB, MC, MD, ME, MF; \
+  __m256i V0, V1, V2, V3, V4, V5, V6, V7; \
+  __m256i V8, V9, VA, VB, VC, VD, VE, VF; \
+  __m256i shuf_bswap64; \
+  V0 = H0; \
+  V1 = H1; \
+  V2 = H2; \
+  V3 = H3; \
+  V4 = H4; \
+  V5 = H5; \
+  V6 = H6; \
+  V7 = H7; \
+  V8 = _mm256_xor_si256( S0, _mm256_set1_epi64x( CB0 ) );  \
+  V9 = _mm256_xor_si256( S1, _mm256_set1_epi64x( CB1 ) );  \
+  VA = _mm256_xor_si256( S2, _mm256_set1_epi64x( CB2 ) );  \
+  VB = _mm256_xor_si256( S3, _mm256_set1_epi64x( CB3 ) );  \
+  VC = _mm256_xor_si256( _mm256_set1_epi64x( T0 ), \
+                         _mm256_set1_epi64x( CB4 ) );  \
+  VD = _mm256_xor_si256( _mm256_set1_epi64x( T0 ), \
+                         _mm256_set1_epi64x( CB5 ) );  \
+  VE = _mm256_xor_si256( _mm256_set1_epi64x( T1 ), \
+                         _mm256_set1_epi64x( CB6 ) );  \
+  VF = _mm256_xor_si256( _mm256_set1_epi64x( T1 ), \
+                         _mm256_set1_epi64x( CB7 ) );  \
+  shuf_bswap64 = _mm256_set_epi64x( 0x08090a0b0c0d0e0f, 0x0001020304050607, \
+                                    0x08090a0b0c0d0e0f, 0x0001020304050607 ); \
+  M0 = _mm256_shuffle_epi8( *(buf+ 0), shuf_bswap64 ); \
+  M1 = _mm256_shuffle_epi8( *(buf+ 1), shuf_bswap64 ); \
+  M2 = _mm256_shuffle_epi8( *(buf+ 2), shuf_bswap64 ); \
+  M3 = _mm256_shuffle_epi8( *(buf+ 3), shuf_bswap64 ); \
+  M4 = _mm256_shuffle_epi8( *(buf+ 4), shuf_bswap64 ); \
+  M5 = _mm256_shuffle_epi8( *(buf+ 5), shuf_bswap64 ); \
+  M6 = _mm256_shuffle_epi8( *(buf+ 6), shuf_bswap64 ); \
+  M7 = _mm256_shuffle_epi8( *(buf+ 7), shuf_bswap64 ); \
+  M8 = _mm256_shuffle_epi8( *(buf+ 8), shuf_bswap64 ); \
+  M9 = _mm256_shuffle_epi8( *(buf+ 9), shuf_bswap64 ); \
+  MA = _mm256_shuffle_epi8( *(buf+10), shuf_bswap64 ); \
+  MB = _mm256_shuffle_epi8( *(buf+11), shuf_bswap64 ); \
+  MC = _mm256_shuffle_epi8( *(buf+12), shuf_bswap64 ); \
+  MD = _mm256_shuffle_epi8( *(buf+13), shuf_bswap64 ); \
+  ME = _mm256_shuffle_epi8( *(buf+14), shuf_bswap64 ); \
+  MF = _mm256_shuffle_epi8( *(buf+15), shuf_bswap64 ); \
+  ROUND_B_4WAY(0); \
+  ROUND_B_4WAY(1); \
+  ROUND_B_4WAY(2); \
+  ROUND_B_4WAY(3); \
+  ROUND_B_4WAY(4); \
+  ROUND_B_4WAY(5); \
+  ROUND_B_4WAY(6); \
+  ROUND_B_4WAY(7); \
+  ROUND_B_4WAY(8); \
+  ROUND_B_4WAY(9); \
+  ROUND_B_4WAY(0); \
+  ROUND_B_4WAY(1); \
+  ROUND_B_4WAY(2); \
+  ROUND_B_4WAY(3); \
+  ROUND_B_4WAY(4); \
+  ROUND_B_4WAY(5); \
+  H0 = mm256_xor4( V8, V0, S0, H0 ); \
+  H1 = mm256_xor4( V9, V1, S1, H1 ); \
+  H2 = mm256_xor4( VA, V2, S2, H2 ); \
+  H3 = mm256_xor4( VB, V3, S3, H3 ); \
+  H4 = mm256_xor4( VC, V4, S0, H4 ); \
+  H5 = mm256_xor4( VD, V5, S1, H5 ); \
+  H6 = mm256_xor4( VE, V6, S2, H6 ); \
+  H7 = mm256_xor4( VF, V7, S3, H7 ); \
+} while (0)
 
 #endif
 
@@ -547,13 +543,23 @@ static void
 blake64_4way_init( blake_4way_big_context *sc, const sph_u64 *iv,
               const sph_u64 *salt )
 {
-        int i;
-        for ( i = 0; i < 8; i++ )
-           sc->H[i] = _mm256_set1_epi64x( iv[i] );
-        for ( i = 0; i < 4; i++ )
-           sc->S[i] = _mm256_set1_epi64x( salt[i] );
-        sc->T0 = sc->T1 = 0;
-        sc->ptr = 0;
+   __m256i zero = m256_zero;
+   casti_m256i( sc->H, 0 ) = _mm256_set1_epi64x( iv[0] );
+   casti_m256i( sc->H, 1 ) = _mm256_set1_epi64x( iv[1] );
+   casti_m256i( sc->H, 2 ) = _mm256_set1_epi64x( iv[2] );
+   casti_m256i( sc->H, 3 ) = _mm256_set1_epi64x( iv[3] );
+   casti_m256i( sc->H, 4 ) = _mm256_set1_epi64x( iv[4] );
+   casti_m256i( sc->H, 5 ) = _mm256_set1_epi64x( iv[5] );
+   casti_m256i( sc->H, 6 ) = _mm256_set1_epi64x( iv[6] );
+   casti_m256i( sc->H, 7 ) = _mm256_set1_epi64x( iv[7] );
+
+   casti_m256i( sc->S, 0 ) = zero;
+   casti_m256i( sc->S, 1 ) = zero;
+   casti_m256i( sc->S, 2 ) = zero;
+   casti_m256i( sc->S, 3 ) = zero;
+
+   sc->T0 = sc->T1 = 0;
+   sc->ptr = 0;
 }
 
 static void
@@ -604,15 +610,11 @@ static void
 blake64_4way_close( blake_4way_big_context *sc,
 	unsigned ub, unsigned n, void *dst, size_t out_size_w64)
 {
-//   union {
-      __m256i buf[16];
-//      sph_u64 dummy;
-//   } u;
-   size_t ptr, k;
+   __m256i buf[16];
+   size_t ptr;
    unsigned bit_len;
    uint64_t z, zz;
    sph_u64 th, tl;
-   __m256i *out;
 
    ptr = sc->ptr;
    bit_len = ((unsigned)ptr << 3);
@@ -665,9 +667,7 @@ blake64_4way_close( blake_4way_big_context *sc,
 
        blake64_4way( sc, buf, 128 );
    }
-   out = (__m256i*)dst;
-   for ( k = 0; k < out_size_w64; k++ )
-       out[k] = mm256_bswap_64( sc->H[k] );
+   mm256_block_bswap_64( (__m256i*)dst, sc->H );
 }
 
 void

algo/blake/blakecoin-4way.c

@@ -29,23 +29,18 @@ int scanhash_blakecoin_4way( struct work *work, uint32_t max_nonce,
    uint32_t *ptarget = work->target;
    const uint32_t first_nonce = pdata[19];
    uint32_t HTarget = ptarget[7];
-   uint32_t _ALIGN(32) edata[20];
    uint32_t n = first_nonce;
+   __m128i  *noncev = (__m128i*)vdata + 19;   // aligned
    int thr_id = mythr->id;  // thr_id arg is deprecated
    if ( opt_benchmark )
       HTarget = 0x7f;
 
-   swab32_array( edata, pdata, 20 );
-   mm128_intrlv_4x32( vdata, edata, edata, edata, edata, 640 );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    blake256r8_4way_init( &blakecoin_4w_ctx );
    blake256r8_4way( &blakecoin_4w_ctx, vdata, 64 );
 
-   uint32_t *noncep = vdata + 76;   // 19*4
    do {
-      be32enc( noncep,    n   );
-      be32enc( noncep +1, n+1 );
-      be32enc( noncep +2, n+2 );
-      be32enc( noncep +3, n+3 );
+      *noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
       pdata[19] = n;
       blakecoin_4way_hash( hash, vdata );
 
@@ -79,9 +74,8 @@ void blakecoin_8way_hash( void *state, const void *input )
      blake256r8_8way( &ctx, input + (64<<3), 16 );
      blake256r8_8way_close( &ctx, vhash );
 
-     mm256_dintrlv_8x32( state,     state+ 32, state+ 64, state+ 96,
-                              state+128, state+160, state+192, state+224,
-                              vhash, 256 );
+     dintrlv_8x32( state,     state+ 32, state+ 64, state+ 96, state+128,
+                   state+160, state+192, state+224, vhash, 256 );
 }
 
 int scanhash_blakecoin_8way( struct work *work, uint32_t max_nonce,
@@ -93,29 +87,19 @@ int scanhash_blakecoin_8way( struct work *work, uint32_t max_nonce,
    uint32_t *ptarget = work->target;
    const uint32_t first_nonce = pdata[19];
    uint32_t HTarget = ptarget[7];
-   uint32_t _ALIGN(32) edata[20];
    uint32_t n = first_nonce;
-   uint32_t *noncep = vdata + 152;   // 19*8
+   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
    int thr_id = mythr->id;  // thr_id arg is deprecated
    if ( opt_benchmark )
       HTarget = 0x7f;
 
-   // we need big endian data...
-   swab32_array( edata, pdata, 20 );
-   mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
-                                 edata, edata, edata, edata, 640 );
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
    blake256r8_8way_init( &blakecoin_8w_ctx );
    blake256r8_8way( &blakecoin_8w_ctx, vdata, 64 );
 
    do {
-      be32enc( noncep,    n   );
-      be32enc( noncep +1, n+1 );
-      be32enc( noncep +2, n+2 );
-      be32enc( noncep +3, n+3 );
-      be32enc( noncep +4, n+4 );
-      be32enc( noncep +5, n+5 );
-      be32enc( noncep +6, n+6 );
-      be32enc( noncep +7, n+7 );
+      *noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
+                                                  n+3, n+2, n+1, n ) );
       pdata[19] = n;
       blakecoin_8way_hash( hash, vdata );
 

algo/blake/pentablake-4way.c

@@ -10,13 +10,8 @@
 #include "blake-hash-4way.h"
 #include "sph_blake.h"
 
-//#define DEBUG_ALGO
-
 extern void pentablakehash_4way( void *output, const void *input )
 {
-//	unsigned char _ALIGN(32) hash[128];
-//	// same as uint32_t hashA[16], hashB[16];
-//	#define hashB hash+64
 
      uint64_t hash0[8] __attribute__ ((aligned (64)));
      uint64_t hash1[8] __attribute__ ((aligned (64)));
@@ -29,22 +24,7 @@ extern void pentablakehash_4way( void *output, const void *input )
      blake512_4way_init( &ctx );
      blake512_4way( &ctx, input, 80 );
      blake512_4way_close( &ctx, vhash );
-/*
-uint64_t sin0[10], sin1[10], sin2[10], sin3[10];
-mm256_deinterleave_4x64( sin0, sin1, sin2, sin3, input, 640 );
-sph_blake512_context ctx2_blake;
-sph_blake512_init(&ctx2_blake);
-sph_blake512(&ctx2_blake, sin0, 80);
-sph_blake512_close(&ctx2_blake, (void*) hash);
 
-mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
-uint64_t* hash64 = (uint64_t*)hash;
-for( int i = 0; i < 8; i++ )
-{
-   if ( hash0[i] != hash64[i] )
-      printf("hash mismatch %u\n",i);
-}
-*/
      blake512_4way_init( &ctx );
      blake512_4way( &ctx, vhash, 64 );
      blake512_4way_close( &ctx, vhash );
@@ -61,42 +41,10 @@ for( int i = 0; i < 8; i++ )
      blake512_4way( &ctx, vhash, 64 );
      blake512_4way_close( &ctx, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
      memcpy( output,    hash0, 32 );
      memcpy( output+32, hash1, 32 );
      memcpy( output+64, hash2, 32 );
      memcpy( output+96, hash3, 32 );
-
-/*
-     uint64_t sin0[10] __attribute__ ((aligned (64)));
-     uint64_t sin1[10] __attribute__ ((aligned (64)));
-     uint64_t sin2[10] __attribute__ ((aligned (64)));
-     uint64_t sin3[10] __attribute__ ((aligned (64)));
-
-	sph_blake512_context     ctx_blake;
-
-	sph_blake512_init(&ctx_blake);
-	sph_blake512(&ctx_blake, input, 80);
-	sph_blake512_close(&ctx_blake, hash);
-
-        sph_blake512_init(&ctx_blake);
-	sph_blake512(&ctx_blake, hash, 64);
-	sph_blake512_close(&ctx_blake, hash);
-
-        sph_blake512_init(&ctx_blake);
-	sph_blake512(&ctx_blake, hash, 64);
-	sph_blake512_close(&ctx_blake, hash);
-
-        sph_blake512_init(&ctx_blake);
-	sph_blake512(&ctx_blake, hash, 64);
-	sph_blake512_close(&ctx_blake, hash);
-
-        sph_blake512_init(&ctx_blake);
-	sph_blake512(&ctx_blake, hash, 64);
-	sph_blake512_close(&ctx_blake, hash);
-
-	memcpy(output, hash, 32);
-*/
 }
 
 int scanhash_pentablake_4way( struct work *work,
@@ -137,7 +85,7 @@ int scanhash_pentablake_4way( struct work *work,
     swab32_array( endiandata, pdata, 20 );
 
     uint64_t *edata = (uint64_t*)endiandata;
-    mm256_intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+    intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
 
     for ( int m=0; m < 6; m++ )
     {

algo/bmw/bmw256-hash-4way.c

@@ -113,50 +113,27 @@ static const uint32_t IV256[] = {
 
 
 #define expand1s( qt, M, H, i ) \
-   _mm_add_epi32( \
-      _mm_add_epi32( \
-         _mm_add_epi32( \
-             _mm_add_epi32( \
-                _mm_add_epi32( ss1( qt[ (i)-16 ] ), \
-                               ss2( qt[ (i)-15 ] ) ), \
-                _mm_add_epi32( ss3( qt[ (i)-14 ] ), \
-                               ss0( qt[ (i)-13 ] ) ) ), \
-             _mm_add_epi32( \
-                _mm_add_epi32( ss1( qt[ (i)-12 ] ), \
-                               ss2( qt[ (i)-11 ] ) ), \
-                _mm_add_epi32( ss3( qt[ (i)-10 ] ), \
-                               ss0( qt[ (i)- 9 ] ) ) ) ), \
-         _mm_add_epi32( \
-             _mm_add_epi32( \
-                _mm_add_epi32( ss1( qt[ (i)- 8 ] ), \
-                               ss2( qt[ (i)- 7 ] ) ), \
-                _mm_add_epi32( ss3( qt[ (i)- 6 ] ), \
-                               ss0( qt[ (i)- 5 ] ) ) ), \
-             _mm_add_epi32( \
-                _mm_add_epi32( ss1( qt[ (i)- 4 ] ), \
-                               ss2( qt[ (i)- 3 ] ) ), \
-                _mm_add_epi32( ss3( qt[ (i)- 2 ] ), \
-                               ss0( qt[ (i)- 1 ] ) ) ) ) ), \
+   _mm_add_epi32(  mm128_add4_32( \
+            mm128_add4_32( ss1( qt[ (i)-16 ] ), ss2( qt[ (i)-15 ] ), \
+                           ss3( qt[ (i)-14 ] ), ss0( qt[ (i)-13 ] ) ), \
+            mm128_add4_32( ss1( qt[ (i)-12 ] ), ss2( qt[ (i)-11 ] ), \
+                           ss3( qt[ (i)-10 ] ), ss0( qt[ (i)- 9 ] ) ), \
+            mm128_add4_32( ss1( qt[ (i)- 8 ] ), ss2( qt[ (i)- 7 ] ), \
+                           ss3( qt[ (i)- 6 ] ), ss0( qt[ (i)- 5 ] ) ),  \
+            mm128_add4_32( ss1( qt[ (i)- 4 ] ), ss2( qt[ (i)- 3 ] ), \
+                           ss3( qt[ (i)- 2 ] ), ss0( qt[ (i)- 1 ] ) ) ), \
       add_elt_s( M, H, (i)-16 ) )
 
 #define expand2s( qt, M, H, i) \
-   _mm_add_epi32( \
-      _mm_add_epi32( \
-         _mm_add_epi32( \
-             _mm_add_epi32( \
-                _mm_add_epi32( qt[ (i)-16 ], rs1( qt[ (i)-15 ] ) ), \
-                _mm_add_epi32( qt[ (i)-14 ], rs2( qt[ (i)-13 ] ) ) ), \
-             _mm_add_epi32( \
-                _mm_add_epi32( qt[ (i)-12 ], rs3( qt[ (i)-11 ] ) ), \
-                _mm_add_epi32( qt[ (i)-10 ], rs4( qt[ (i)- 9 ] ) ) ) ), \
-         _mm_add_epi32( \
-             _mm_add_epi32( \
-                _mm_add_epi32( qt[ (i)- 8 ], rs5( qt[ (i)- 7 ] ) ), \
-                _mm_add_epi32( qt[ (i)- 6 ], rs6( qt[ (i)- 5 ] ) ) ), \
-             _mm_add_epi32( \
-                _mm_add_epi32( qt[ (i)- 4 ], rs7( qt[ (i)- 3 ] ) ), \
-                _mm_add_epi32( ss4( qt[ (i)- 2 ] ), \
-                               ss5( qt[ (i)- 1 ] ) ) ) ) ), \
+   _mm_add_epi32( mm128_add4_32( \
+            mm128_add4_32( qt[ (i)-16 ], rs1( qt[ (i)-15 ] ), \
+                           qt[ (i)-14 ], rs2( qt[ (i)-13 ] ) ), \
+            mm128_add4_32( qt[ (i)-12 ], rs3( qt[ (i)-11 ] ), \
+                           qt[ (i)-10 ], rs4( qt[ (i)- 9 ] ) ), \
+            mm128_add4_32( qt[ (i)- 8 ], rs5( qt[ (i)- 7 ] ), \
+                           qt[ (i)- 6 ], rs6( qt[ (i)- 5 ] ) ), \
+            mm128_add4_32( qt[ (i)- 4 ], rs7( qt[ (i)- 3 ] ), \
+                           ss4( qt[ (i)- 2 ] ), ss5( qt[ (i)- 1 ] ) ) ), \
       add_elt_s( M, H, (i)-16 ) )
 
 #define Ws0 \
@@ -357,17 +334,11 @@ void compress_small( const __m128i *M, const __m128i H[16], __m128i dH[16] )
    qt[30] = expand2s( qt, M, H, 30 );
    qt[31] = expand2s( qt, M, H, 31 );
 
-   xl = _mm_xor_si128(
-              _mm_xor_si128( _mm_xor_si128( qt[16], qt[17] ),
-                             _mm_xor_si128( qt[18], qt[19] ) ),
-              _mm_xor_si128( _mm_xor_si128( qt[20], qt[21] ),
-                             _mm_xor_si128( qt[22], qt[23] ) ) );
-   xh = _mm_xor_si128( xl,
-             _mm_xor_si128(
-                 _mm_xor_si128( _mm_xor_si128( qt[24], qt[25] ),
-                                   _mm_xor_si128( qt[26], qt[27] ) ),
-                 _mm_xor_si128( _mm_xor_si128( qt[28], qt[29] ),
-                                   _mm_xor_si128( qt[30], qt[31] ) )));
+   xl = _mm_xor_si128( mm128_xor4( qt[16], qt[17], qt[18], qt[19] ),
+                       mm128_xor4( qt[20], qt[21], qt[22], qt[23] ) );
+   xh = _mm_xor_si128( xl, _mm_xor_si128(
+                             mm128_xor4( qt[24], qt[25], qt[26], qt[27] ),
+                             mm128_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
 
    dH[ 0] = _mm_add_epi32(
                  _mm_xor_si128( M[0],
@@ -695,22 +666,15 @@ bmw256_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
 
 #define expand2s8( qt, M, H, i) \
    _mm256_add_epi32( \
-      _mm256_add_epi32( \
-         _mm256_add_epi32( \
-             _mm256_add_epi32( \
-                _mm256_add_epi32( qt[ (i)-16 ], r8s1( qt[ (i)-15 ] ) ), \
-                _mm256_add_epi32( qt[ (i)-14 ], r8s2( qt[ (i)-13 ] ) ) ), \
-             _mm256_add_epi32( \
-                _mm256_add_epi32( qt[ (i)-12 ], r8s3( qt[ (i)-11 ] ) ), \
-                _mm256_add_epi32( qt[ (i)-10 ], r8s4( qt[ (i)- 9 ] ) ) ) ), \
-         _mm256_add_epi32( \
-             _mm256_add_epi32( \
-                _mm256_add_epi32( qt[ (i)- 8 ], r8s5( qt[ (i)- 7 ] ) ), \
-                _mm256_add_epi32( qt[ (i)- 6 ], r8s6( qt[ (i)- 5 ] ) ) ), \
-             _mm256_add_epi32( \
-                _mm256_add_epi32( qt[ (i)- 4 ], r8s7( qt[ (i)- 3 ] ) ), \
-                _mm256_add_epi32( s8s4( qt[ (i)- 2 ] ), \
-                                  s8s5( qt[ (i)- 1 ] ) ) ) ) ), \
+      mm256_add4_32( \
+          mm256_add4_32( qt[ (i)-16 ], r8s1( qt[ (i)-15 ] ), \
+                         qt[ (i)-14 ], r8s2( qt[ (i)-13 ] ) ), \
+          mm256_add4_32( qt[ (i)-12 ], r8s3( qt[ (i)-11 ] ), \
+                         qt[ (i)-10 ], r8s4( qt[ (i)- 9 ] ) ), \
+          mm256_add4_32( qt[ (i)- 8 ], r8s5( qt[ (i)- 7 ] ), \
+                         qt[ (i)- 6 ], r8s6( qt[ (i)- 5 ] ) ), \
+          mm256_add4_32( qt[ (i)- 4 ], r8s7( qt[ (i)- 3 ] ), \
+                         s8s4( qt[ (i)- 2 ] ), s8s5( qt[ (i)- 1 ] ) ) ), \
       add_elt_s8( M, H, (i)-16 ) )
 
 
@@ -913,16 +877,11 @@ void compress_small_8way( const __m256i *M, const __m256i H[16],
    qt[31] = expand2s8( qt, M, H, 31 );
 
    xl = _mm256_xor_si256(
-              _mm256_xor_si256( _mm256_xor_si256( qt[16], qt[17] ),
-                                _mm256_xor_si256( qt[18], qt[19] ) ),
-              _mm256_xor_si256( _mm256_xor_si256( qt[20], qt[21] ),
-                                _mm256_xor_si256( qt[22], qt[23] ) ) );
-   xh = _mm256_xor_si256( xl,
-             _mm256_xor_si256(
-                 _mm256_xor_si256( _mm256_xor_si256( qt[24], qt[25] ),
-                                   _mm256_xor_si256( qt[26], qt[27] ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( qt[28], qt[29] ),
-                                   _mm256_xor_si256( qt[30], qt[31] ) )));
+              mm256_xor4( qt[16], qt[17], qt[18], qt[19] ),
+              mm256_xor4( qt[20], qt[21], qt[22], qt[23] ) );
+   xh = _mm256_xor_si256( xl,  _mm256_xor_si256(
+                 mm256_xor4( qt[24], qt[25], qt[26], qt[27] ),
+                 mm256_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
 
    dH[ 0] = _mm256_add_epi32(
                  _mm256_xor_si256( M[0],

algo/bmw/bmw512-hash-4way.c

@@ -569,28 +569,20 @@ void bmw512_2way_close( bmw_2way_big_context *ctx, void *dst )
 
 
 #define sb0(x) \
-   _mm256_xor_si256( _mm256_xor_si256( _mm256_srli_epi64( (x), 1), \
-                                       _mm256_slli_epi64( (x), 3) ), \
-                     _mm256_xor_si256( mm256_rol_64( (x),  4), \
-                                       mm256_rol_64( (x), 37) ) )
+   mm256_xor4( _mm256_srli_epi64( (x), 1), _mm256_slli_epi64( (x), 3), \
+                mm256_rol_64(     (x), 4),  mm256_rol_64(     (x),37) )
 
 #define sb1(x) \
-   _mm256_xor_si256( _mm256_xor_si256( _mm256_srli_epi64( (x), 1), \
-                                       _mm256_slli_epi64( (x), 2) ), \
-                     _mm256_xor_si256( mm256_rol_64( (x), 13), \
-                                       mm256_rol_64( (x), 43) ) )
+   mm256_xor4( _mm256_srli_epi64( (x), 1), _mm256_slli_epi64( (x), 2), \
+                mm256_rol_64(     (x),13),  mm256_rol_64(     (x),43) )
 
 #define sb2(x) \
-   _mm256_xor_si256( _mm256_xor_si256( _mm256_srli_epi64( (x), 2), \
-                                       _mm256_slli_epi64( (x), 1) ), \
-                     _mm256_xor_si256( mm256_rol_64( (x), 19), \
-                                       mm256_rol_64( (x), 53) ) )
+   mm256_xor4( _mm256_srli_epi64( (x), 2), _mm256_slli_epi64( (x), 1), \
+                mm256_rol_64(     (x),19),  mm256_rol_64(     (x),53) )
 
 #define sb3(x) \
-   _mm256_xor_si256( _mm256_xor_si256( _mm256_srli_epi64( (x), 2), \
-                                       _mm256_slli_epi64( (x), 2) ), \
-                     _mm256_xor_si256( mm256_rol_64( (x), 28), \
-                                       mm256_rol_64( (x), 59) ) )
+   mm256_xor4( _mm256_srli_epi64( (x), 2), _mm256_slli_epi64( (x), 2), \
+                mm256_rol_64(     (x),28),  mm256_rol_64(     (x),59) )
 
 #define sb4(x) \
   _mm256_xor_si256( (x), _mm256_srli_epi64( (x), 1 ) )
@@ -618,55 +610,32 @@ void bmw512_2way_close( bmw_2way_big_context *ctx, void *dst )
                              rol_off_64( M, j, 10 ) ), \
             _mm256_set1_epi64x( ( (j) + 16 ) * 0x0555555555555555ULL ) ), \
        H[ ( (j)+7 ) & 0xF ] )
-          
+
+
 #define expand1b( qt, M, H, i ) \
-   _mm256_add_epi64( \
-      _mm256_add_epi64( \
-         _mm256_add_epi64( \
-             _mm256_add_epi64( \
-                _mm256_add_epi64( sb1( qt[ (i)-16 ] ), \
-                                  sb2( qt[ (i)-15 ] ) ), \
-                _mm256_add_epi64( sb3( qt[ (i)-14 ] ), \
-                                  sb0( qt[ (i)-13 ] ) ) ), \
-             _mm256_add_epi64( \
-                _mm256_add_epi64( sb1( qt[ (i)-12 ] ), \
-                                  sb2( qt[ (i)-11 ] ) ), \
-                _mm256_add_epi64( sb3( qt[ (i)-10 ] ), \
-                                  sb0( qt[ (i)- 9 ] ) ) ) ), \
-         _mm256_add_epi64( \
-             _mm256_add_epi64( \
-                _mm256_add_epi64( sb1( qt[ (i)- 8 ] ), \
-                                  sb2( qt[ (i)- 7 ] ) ), \
-                _mm256_add_epi64( sb3( qt[ (i)- 6 ] ), \
-                                  sb0( qt[ (i)- 5 ] ) ) ), \
-             _mm256_add_epi64( \
-                _mm256_add_epi64( sb1( qt[ (i)- 4 ] ), \
-                                  sb2( qt[ (i)- 3 ] ) ), \
-                _mm256_add_epi64( sb3( qt[ (i)- 2 ] ), \
-                                  sb0( qt[ (i)- 1 ] ) ) ) ) ), \
+   _mm256_add_epi64( mm256_add4_64( \
+      mm256_add4_64( sb1( qt[ (i)-16 ] ), sb2( qt[ (i)-15 ] ), \
+                     sb3( qt[ (i)-14 ] ), sb0( qt[ (i)-13 ] )), \
+      mm256_add4_64( sb1( qt[ (i)-12 ] ), sb2( qt[ (i)-11 ] ), \
+                     sb3( qt[ (i)-10 ] ), sb0( qt[ (i)- 9 ] )), \
+      mm256_add4_64( sb1( qt[ (i)- 8 ] ), sb2( qt[ (i)- 7 ] ), \
+                     sb3( qt[ (i)- 6 ] ), sb0( qt[ (i)- 5 ] )), \
+      mm256_add4_64( sb1( qt[ (i)- 4 ] ), sb2( qt[ (i)- 3 ] ), \
+                     sb3( qt[ (i)- 2 ] ), sb0( qt[ (i)- 1 ] ) ) ), \
       add_elt_b( M, H, (i)-16 ) )
 
 #define expand2b( qt, M, H, i) \
-   _mm256_add_epi64( \
-      _mm256_add_epi64( \
-         _mm256_add_epi64( \
-             _mm256_add_epi64( \
-                _mm256_add_epi64( qt[ (i)-16 ], rb1( qt[ (i)-15 ] ) ), \
-                _mm256_add_epi64( qt[ (i)-14 ], rb2( qt[ (i)-13 ] ) ) ), \
-             _mm256_add_epi64( \
-                _mm256_add_epi64( qt[ (i)-12 ], rb3( qt[ (i)-11 ] ) ), \
-                _mm256_add_epi64( qt[ (i)-10 ], rb4( qt[ (i)- 9 ] ) ) ) ), \
-         _mm256_add_epi64( \
-             _mm256_add_epi64( \
-                _mm256_add_epi64( qt[ (i)- 8 ], rb5( qt[ (i)- 7 ] ) ), \
-                _mm256_add_epi64( qt[ (i)- 6 ], rb6( qt[ (i)- 5 ] ) ) ), \
-             _mm256_add_epi64( \
-                _mm256_add_epi64( qt[ (i)- 4 ], rb7( qt[ (i)- 3 ] ) ), \
-                _mm256_add_epi64( sb4( qt[ (i)- 2 ] ), \
-                                  sb5( qt[ (i)- 1 ] ) ) ) ) ), \
+   _mm256_add_epi64( mm256_add4_64( \
+      mm256_add4_64( qt[ (i)-16 ], rb1( qt[ (i)-15 ] ), \
+                     qt[ (i)-14 ], rb2( qt[ (i)-13 ] ) ), \
+      mm256_add4_64( qt[ (i)-12 ], rb3( qt[ (i)-11 ] ), \
+                     qt[ (i)-10 ], rb4( qt[ (i)- 9 ] ) ), \
+      mm256_add4_64( qt[ (i)- 8 ], rb5( qt[ (i)- 7 ] ), \
+                     qt[ (i)- 6 ], rb6( qt[ (i)- 5 ] ) ), \
+      mm256_add4_64( qt[ (i)- 4 ], rb7( qt[ (i)- 3 ] ), \
+                     sb4( qt[ (i)- 2 ] ), sb5( qt[ (i)- 1 ] ) ) ), \
       add_elt_b( M, H, (i)-16 ) )
 
-
 #define Wb0 \
    _mm256_add_epi64( \
        _mm256_add_epi64( \
@@ -864,95 +833,90 @@ void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
    qt[30] = expand2b( qt, M, H, 30 ); 
    qt[31] = expand2b( qt, M, H, 31 ); 
 
-   xl = _mm256_xor_si256( 
-              _mm256_xor_si256( _mm256_xor_si256( qt[16], qt[17] ), 
-                                _mm256_xor_si256( qt[18], qt[19] ) ), 
-              _mm256_xor_si256( _mm256_xor_si256( qt[20], qt[21] ), 
-                                _mm256_xor_si256( qt[22], qt[23] ) ) ); 
-   xh = _mm256_xor_si256( xl, 
-             _mm256_xor_si256( 
-                 _mm256_xor_si256( _mm256_xor_si256( qt[24], qt[25] ),
-                                   _mm256_xor_si256( qt[26], qt[27] ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( qt[28], qt[29] ),
-                                   _mm256_xor_si256( qt[30], qt[31] ) )));
+   xl = _mm256_xor_si256(
+           mm256_xor4( qt[16], qt[17], qt[18], qt[19] ), 
+           mm256_xor4( qt[20], qt[21], qt[22], qt[23] ) ); 
+   xh = _mm256_xor_si256( xl, _mm256_xor_si256( 
+           mm256_xor4( qt[24], qt[25], qt[26], qt[27] ),
+           mm256_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
 
    dH[ 0] = _mm256_add_epi64(
-                 _mm256_xor_si256( M[0],
-                      _mm256_xor_si256( _mm256_slli_epi64( xh, 5 ),
-                                        _mm256_srli_epi64( qt[16], 5 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[24] ), qt[ 0] ));
+               _mm256_xor_si256( M[0],
+                  _mm256_xor_si256( _mm256_slli_epi64( xh, 5 ),
+                                    _mm256_srli_epi64( qt[16], 5 ) ) ),
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[24] ), qt[ 0] ) );
    dH[ 1] = _mm256_add_epi64(
-                 _mm256_xor_si256( M[1],
-                      _mm256_xor_si256( _mm256_srli_epi64( xh, 7 ),
-                                        _mm256_slli_epi64( qt[17], 8 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[25] ), qt[ 1] ));
+               _mm256_xor_si256( M[1],
+                  _mm256_xor_si256( _mm256_srli_epi64( xh, 7 ),
+                                    _mm256_slli_epi64( qt[17], 8 ) ) ),
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[25] ), qt[ 1] ) );
    dH[ 2] = _mm256_add_epi64(
-                 _mm256_xor_si256( M[2],
-                      _mm256_xor_si256( _mm256_srli_epi64( xh, 5 ),
-                                        _mm256_slli_epi64( qt[18], 5 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[26] ), qt[ 2] ));
+               _mm256_xor_si256( M[2],
+                  _mm256_xor_si256( _mm256_srli_epi64( xh, 5 ),
+                                    _mm256_slli_epi64( qt[18], 5 ) ) ),
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[26] ), qt[ 2] ) );
    dH[ 3] = _mm256_add_epi64(
-                 _mm256_xor_si256( M[3],
-                      _mm256_xor_si256( _mm256_srli_epi64( xh, 1 ),
-                                        _mm256_slli_epi64( qt[19], 5 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[27] ), qt[ 3] ));
+               _mm256_xor_si256( M[3],
+                  _mm256_xor_si256( _mm256_srli_epi64( xh, 1 ),
+                                    _mm256_slli_epi64( qt[19], 5 ) ) ),
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[27] ), qt[ 3] ) );
    dH[ 4] = _mm256_add_epi64(
-                 _mm256_xor_si256( M[4],
-                      _mm256_xor_si256( _mm256_srli_epi64( xh, 3 ),
-                                        _mm256_slli_epi64( qt[20], 0 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[28] ), qt[ 4] ));
+               _mm256_xor_si256( M[4],
+                  _mm256_xor_si256( _mm256_srli_epi64( xh, 3 ),
+                                    _mm256_slli_epi64( qt[20], 0 ) ) ),
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[28] ), qt[ 4] ) );
    dH[ 5] = _mm256_add_epi64(
-                 _mm256_xor_si256( M[5],
-                      _mm256_xor_si256( _mm256_slli_epi64( xh, 6 ),
-                                        _mm256_srli_epi64( qt[21], 6 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[29] ), qt[ 5] ));
+               _mm256_xor_si256( M[5],
+                  _mm256_xor_si256( _mm256_slli_epi64( xh, 6 ),
+                                    _mm256_srli_epi64( qt[21], 6 ) ) ),
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[29] ), qt[ 5] ) );
    dH[ 6] = _mm256_add_epi64(
-                 _mm256_xor_si256( M[6],
-                      _mm256_xor_si256( _mm256_srli_epi64( xh, 4 ),
-                                        _mm256_slli_epi64( qt[22], 6 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[30] ), qt[ 6] ));
+               _mm256_xor_si256( M[6],
+                  _mm256_xor_si256( _mm256_srli_epi64( xh, 4 ),
+                                    _mm256_slli_epi64( qt[22], 6 ) ) ),
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[30] ), qt[ 6] ) );
    dH[ 7] = _mm256_add_epi64(
-                 _mm256_xor_si256( M[7],
-                      _mm256_xor_si256( _mm256_srli_epi64( xh, 11 ),
-                                        _mm256_slli_epi64( qt[23], 2 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[31] ), qt[ 7] ));
+               _mm256_xor_si256( M[7],
+                  _mm256_xor_si256( _mm256_srli_epi64( xh, 11 ),
+                                    _mm256_slli_epi64( qt[23], 2 ) ) ),
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[31] ), qt[ 7] ) );
    dH[ 8] = _mm256_add_epi64( _mm256_add_epi64(
-                 mm256_rol_64( dH[4], 9 ),
+              mm256_rol_64( dH[4], 9 ),
                  _mm256_xor_si256( _mm256_xor_si256( xh, qt[24] ), M[ 8] )),
                  _mm256_xor_si256( _mm256_slli_epi64( xl, 8 ),
                                    _mm256_xor_si256( qt[23], qt[ 8] ) ) );
    dH[ 9] = _mm256_add_epi64( _mm256_add_epi64(
-                 mm256_rol_64( dH[5], 10 ),
+              mm256_rol_64( dH[5], 10 ),
                  _mm256_xor_si256( _mm256_xor_si256( xh, qt[25] ), M[ 9] )),
                  _mm256_xor_si256( _mm256_srli_epi64( xl, 6 ),
                                    _mm256_xor_si256( qt[16], qt[ 9] ) ) );
    dH[10] = _mm256_add_epi64( _mm256_add_epi64(
-                 mm256_rol_64( dH[6], 11 ),
+              mm256_rol_64( dH[6], 11 ),
                  _mm256_xor_si256( _mm256_xor_si256( xh, qt[26] ), M[10] )),
                  _mm256_xor_si256( _mm256_slli_epi64( xl, 6 ),
                                    _mm256_xor_si256( qt[17], qt[10] ) ) );
    dH[11] = _mm256_add_epi64( _mm256_add_epi64(
-                 mm256_rol_64( dH[7], 12 ),
+              mm256_rol_64( dH[7], 12 ),
                  _mm256_xor_si256( _mm256_xor_si256( xh, qt[27] ), M[11] )),
                  _mm256_xor_si256( _mm256_slli_epi64( xl, 4 ),
                                    _mm256_xor_si256( qt[18], qt[11] ) ) );
    dH[12] = _mm256_add_epi64( _mm256_add_epi64(
-                 mm256_rol_64( dH[0], 13 ),
+              mm256_rol_64( dH[0], 13 ),
                  _mm256_xor_si256( _mm256_xor_si256( xh, qt[28] ), M[12] )),
                  _mm256_xor_si256( _mm256_srli_epi64( xl, 3 ),
                                    _mm256_xor_si256( qt[19], qt[12] ) ) );
    dH[13] = _mm256_add_epi64( _mm256_add_epi64(
-                 mm256_rol_64( dH[1], 14 ),
+              mm256_rol_64( dH[1], 14 ),
                  _mm256_xor_si256( _mm256_xor_si256( xh, qt[29] ), M[13] )),
                  _mm256_xor_si256( _mm256_srli_epi64( xl, 4 ),
                                    _mm256_xor_si256( qt[20], qt[13] ) ) );
    dH[14] = _mm256_add_epi64( _mm256_add_epi64(
-                 mm256_rol_64( dH[2], 15 ),
+              mm256_rol_64( dH[2], 15 ),
                  _mm256_xor_si256( _mm256_xor_si256( xh, qt[30] ), M[14] )),
                  _mm256_xor_si256( _mm256_srli_epi64( xl, 7 ),
                                    _mm256_xor_si256( qt[21], qt[14] ) ) );
    dH[15] = _mm256_add_epi64( _mm256_add_epi64(
-                 mm256_rol_64( dH[3], 16 ),
+              mm256_rol_64( dH[3], 16 ),
                  _mm256_xor_si256( _mm256_xor_si256( xh, qt[31] ), M[15] )),
                  _mm256_xor_si256( _mm256_srli_epi64( xl, 2 ),
                                    _mm256_xor_si256( qt[22], qt[15] ) ) );

algo/cryptonight/cryptolight.c

@@ -242,6 +242,8 @@ void cryptolight_hash(void* output, const void* input, int len) {
 	free(ctx);
 }
 
+#if defined(__AES__)
+
 static void cryptolight_hash_ctx_aes_ni(void* output, const void* input,
                        int len, struct cryptonight_ctx* ctx)
 {
@@ -312,6 +314,8 @@ static void cryptolight_hash_ctx_aes_ni(void* output, const void* input,
 	oaes_free((OAES_CTX **) &ctx->aes_ctx);
 }
 
+#endif
+
 int scanhash_cryptolight( struct work *work,
 		uint32_t max_nonce, uint64_t *hashes_done, struct thr_info *mythr)
 {

algo/groestl/myrgr-4way.c

@@ -47,10 +47,6 @@ void myriad_4way_hash( void *output, const void *input )
 
      sha256_4way( &ctx.sha, vhash, 64 );
      sha256_4way_close( &ctx.sha, output );
-
-//     sha256_4way_close( &ctx.sha, vhash );
-//     mm128_dintrlv_4x32( output, output+32, output+64, output+96,
-//                           vhash, 256 );
 }
 
 int scanhash_myriad_4way( struct work *work, uint32_t max_nonce,
@@ -68,18 +64,10 @@ int scanhash_myriad_4way( struct work *work, uint32_t max_nonce,
    __m128i  *noncev = (__m128i*)vdata + 19;   // aligned
    int thr_id = mythr->id;  // thr_id arg is deprecated
 
-/*
-        uint32_t *pdata = work->data;
-        uint32_t *ptarget = work->target;
-
-	uint32_t _ALIGN(64) endiandata[20];
-	const uint32_t first_nonce = pdata[19];
-	uint32_t nonce = first_nonce;
-*/
    if ( opt_benchmark )
       ( (uint32_t*)ptarget )[7] = 0x0000ff;
 
-   mm128_bswap_intrlv80_4x32( vdata, pdata );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    do {
       *noncev = mm128_bswap_32( _mm_set_epi32( n+3,n+2,n+1,n ) );
 

algo/hamsi/hamsi-hash-4way.c

@@ -531,16 +531,17 @@ static const sph_u32 T512[64][16] = {
 
 #define INPUT_BIG \
 do { \
+  const __m256i zero = _mm256_setzero_si256(); \
   __m256i db = *buf; \
   const sph_u32 *tp = &T512[0][0]; \
-  m0 = m256_zero; \
-  m1 = m256_zero; \
-  m2 = m256_zero; \
-  m3 = m256_zero; \
-  m4 = m256_zero; \
-  m5 = m256_zero; \
-  m6 = m256_zero; \
-  m7 = m256_zero; \
+  m0 = zero; \
+  m1 = zero; \
+  m2 = zero; \
+  m3 = zero; \
+  m4 = zero; \
+  m5 = zero; \
+  m6 = zero; \
+  m7 = zero; \
   for ( int u = 0; u < 64; u++ ) \
   { \
      __m256i dm = _mm256_and_si256( db, m256_one_64 ) ; \
@@ -913,9 +914,7 @@ void hamsi512_4way( hamsi_4way_big_context *sc, const void *data, size_t len )
 
 void hamsi512_4way_close( hamsi_4way_big_context *sc, void *dst )
 {
-   __m256i *out = (__m256i*)dst;
    __m256i pad[1];
-   size_t u;
    int ch, cl;
 
    sph_enc32be( &ch, sc->count_high );
@@ -925,8 +924,8 @@ void hamsi512_4way_close( hamsi_4way_big_context *sc, void *dst )
                                   0UL, 0x80UL, 0UL, 0x80UL );
    hamsi_big( sc, sc->buf, 1 );
    hamsi_big_final( sc, pad );
-   for ( u = 0; u < 8; u ++ )
-      out[u] = mm256_bswap_32( sc->h[u] );
+
+   mm256_block_bswap_32( (__m256i*)dst, sc->h );
 }
 
 #ifdef __cplusplus

algo/hodl/aes.c

@@ -83,7 +83,7 @@ void ExpandAESKey256(__m128i *keys, const __m128i *KeyBuf)
     keys[14] = tmp1;
 }
 
-#ifdef __SSE4_2__
+#if defined(__SSE4_2__)
 //#ifdef __AVX__
 
 #define AESENC(i,j) \
@@ -151,7 +151,7 @@ void AES256CBC(__m128i** data, const __m128i** next, __m128i ExpandedKey[][16],
     }
 }
 
-#else    // NO SSE4.2
+#else    // NO AVX
 
 static inline __m128i AES256Core(__m128i State, const __m128i *ExpandedKey)
 {

algo/hodl/hodl-gate.c

@@ -166,7 +166,7 @@ bool register_hodl_algo( algo_gate_t* gate )
 //     return false;
 //  }
   pthread_barrier_init( &hodl_barrier, NULL, opt_n_threads );
-  gate->optimizations         = AES_OPT | SSE42_OPT | AVX2_OPT;
+  gate->optimizations         = AES_OPT | AVX_OPT | AVX2_OPT;
   gate->scanhash              = (void*)&hodl_scanhash;
   gate->get_new_work          = (void*)&hodl_get_new_work;
   gate->longpoll_rpc_call     = (void*)&hodl_longpoll_rpc_call;

algo/hodl/hodl-wolf.c

@@ -17,7 +17,7 @@ void GenerateGarbageCore( CacheEntry *Garbage, int ThreadID, int ThreadCount,
     const uint32_t StartChunk = ThreadID * Chunk;
     const uint32_t EndChunk   = StartChunk + Chunk;
 
-#ifdef __SSE4_2__
+#if defined(__SSE4_2__)
 //#ifdef __AVX__
     uint64_t* TempBufs[ SHA512_PARALLEL_N ] ;
     uint64_t* desination[ SHA512_PARALLEL_N ];
@@ -64,7 +64,7 @@ void Rev256(uint32_t *Dest, const uint32_t *Src)
 int scanhash_hodl_wolf( struct work* work, uint32_t max_nonce,
                         uint64_t *hashes_done, struct thr_info *mythr )
 {
-#ifdef __SSE4_2__
+#if defined(__SSE4_2__)
 //#ifdef __AVX__
     uint32_t *pdata = work->data;
     uint32_t *ptarget = work->target;
@@ -140,7 +140,7 @@ int scanhash_hodl_wolf( struct work* work, uint32_t max_nonce,
     return(0);
 
 
-#else  // no SSE4.2
+#else  // no AVX
 
     uint32_t *pdata = work->data;
     uint32_t *ptarget = work->target;
@@ -148,6 +148,7 @@ int scanhash_hodl_wolf( struct work* work, uint32_t max_nonce,
     CacheEntry *Garbage = (CacheEntry*)hodl_scratchbuf;
     CacheEntry Cache;
     uint32_t CollisionCount = 0;
+    int threadNumber = mythr->id;
 
     swab32_array( BlockHdr, pdata, 20 );
         // Search for pattern in psuedorandom data      
@@ -205,7 +206,7 @@ int scanhash_hodl_wolf( struct work* work, uint32_t max_nonce,
     *hashes_done = CollisionCount;
     return(0);
 
-#endif  // SSE4.2 else
+#endif  // AVX else
 
 }
 

algo/hodl/sha512-avx.h

@@ -23,6 +23,7 @@ typedef struct
    __m256i h[8];
    __m256i w[80];
 #elif defined(__SSE4_2__)
+//#elif defined(__AVX__)
    __m128i h[8];
    __m128i w[80];
 #else
@@ -32,7 +33,8 @@ typedef struct
 
 #ifdef __AVX2__
 #define SHA512_PARALLEL_N 8
-#elif defined(__SSE$_2__)
+#elif defined(__SSE4_2__)
+//#elif defined(__AVX__)
 #define SHA512_PARALLEL_N 4
 #else
 #define SHA512_PARALLEL_N 1   // dummy value

algo/hodl/sha512_avx.c

@@ -1,6 +1,6 @@
 #ifndef __AVX2__
 
-#ifdef __SSE4_2__
+#if defined(__SSE4_2__)
 //#ifdef __AVX__
 
 //Dependencies

algo/hodl/wolf-aes.h

@@ -6,7 +6,7 @@
 
 void ExpandAESKey256(__m128i *keys, const __m128i *KeyBuf);
 
-#ifdef __SSE4_2__
+#if defined(__SSE4_2__)
 //#ifdef __AVX__
 
 #define AES_PARALLEL_N 8

algo/jh/jha-4way.c

@@ -3,7 +3,6 @@
 #include <stdint.h>
 #include <string.h>
 #include <stdio.h>
-//#include "avxdefs.h"
 
 #if defined(JHA_4WAY)
 
@@ -13,9 +12,6 @@
 #include "algo/keccak/keccak-hash-4way.h"
 #include "algo/groestl/aes_ni/hash-groestl.h"
 
-//static __thread keccak512_4way_context jha_kec_mid
-//                                   __attribute__ ((aligned (64)));
-
 void jha_hash_4way( void *out, const void *input )
 {
     uint64_t hash0[8] __attribute__ ((aligned (64)));
@@ -46,7 +42,7 @@ void jha_hash_4way( void *out, const void *input )
        vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256(
                vh[0], _mm256_set1_epi64x( 1 ) ), m256_zero );
 
-       mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+       dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
        init_groestl( &ctx_groestl, 64 );
        update_and_final_groestl( &ctx_groestl, (char*)hash0,
                                                (char*)hash0, 512 );
@@ -59,7 +55,7 @@ void jha_hash_4way( void *out, const void *input )
        init_groestl( &ctx_groestl, 64 );
        update_and_final_groestl( &ctx_groestl, (char*)hash3,
                                                (char*)hash3, 512 );
-       mm256_intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
+       intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
 
        skein512_4way_init( &ctx_skein );
        skein512_4way( &ctx_skein, vhash, 64 );
@@ -79,8 +75,6 @@ void jha_hash_4way( void *out, const void *input )
        for ( int i = 0; i < 8; i++ )
           casti_m256i( out, i ) = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );
     }
-
-//    mm256_dintrlv_4x64( out, out+32, out+64, out+96, vhash, 256 );
 }
 
 int scanhash_jha_4way( struct work *work, uint32_t max_nonce,
@@ -88,7 +82,6 @@ int scanhash_jha_4way( struct work *work, uint32_t max_nonce,
 {
    uint32_t hash[8*4] __attribute__ ((aligned (64)));
    uint32_t vdata[20*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t *hash7 = &(hash[25]);
    uint32_t lane_hash[8] __attribute__ ((aligned (32)));
    uint32_t *pdata = work->data;
@@ -116,11 +109,7 @@ int scanhash_jha_4way( struct work *work, uint32_t max_nonce,
 		0
 	};
 
-   for ( int i=0; i < 19; i++ )
-      be32enc( &edata[i], pdata[i] );
-
-   mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_4x64( vdata, pdata );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
    for ( int m = 0; m < 6; m++ )
    {
@@ -130,26 +119,17 @@ int scanhash_jha_4way( struct work *work, uint32_t max_nonce,
          do {
               *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
                 _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
-//              be32enc( noncep,   n   );
-//              be32enc( noncep+2, n+1 );
-//              be32enc( noncep+4, n+2 );
-//              be32enc( noncep+6, n+3 );
 
               jha_hash_4way( hash, vdata );
               pdata[19] = n;
 
-//              for ( int i = 0; i < 4; i++ )
-//              if ( ( !( (hash+(i<<3))[7] & mask ) == 0 )
-//                  && fulltest( hash+(i<<3), ptarget ) )
               for ( int i = 0; i < 4; i++ ) if ( !( (hash7[i] & mask ) == 0 ) )
               {
-                 mm256_extr_lane_4x64( lane_hash, hash, i, 256 );
+                 extr_lane_4x64( lane_hash, hash, i, 256 );
                  if ( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark )
                  {
                     pdata[19] = n+i;
                     submit_lane_solution( work, lane_hash, mythr, i );
-//                 nonces[ num_found++ ] = n+i;
-//                 work_set_target_ratio( work, hash+(i<<3) );
                  }
               }
               n += 4;

algo/keccak/keccak-4way.c

@@ -20,8 +20,7 @@ int scanhash_keccak_4way( struct work *work, uint32_t max_nonce,
                           uint64_t *hashes_done, struct thr_info *mythr )
 {
    uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-   uint32_t hash[8*4] __attribute__ ((aligned (32)));
-   uint32_t edata[20] __attribute__ ((aligned (64)));
+   uint32_t hash[16*4] __attribute__ ((aligned (32)));
    uint32_t lane_hash[8] __attribute__ ((aligned (32)));
    uint32_t *hash7 = &(hash[25]);   // 3*8+1
    uint32_t *pdata = work->data;
@@ -32,9 +31,7 @@ int scanhash_keccak_4way( struct work *work, uint32_t max_nonce,
 //   const uint32_t Htarg = ptarget[7];
     int thr_id = mythr->id;  // thr_id arg is deprecated
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_4x64( vdata, pdata );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
    do {
        *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
                 _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
@@ -44,7 +41,7 @@ int scanhash_keccak_4way( struct work *work, uint32_t max_nonce,
       for ( int lane = 0; lane < 4; lane++ )
       if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
       {
-          mm256_extr_lane_4x64( lane_hash, hash, lane, 256 );
+          extr_lane_4x64( lane_hash, hash, lane, 256 );
           if ( fulltest( lane_hash, ptarget ) )
           {
               pdata[19] = n + lane;

algo/luffa/sph_luffa.c

@@ -77,6 +77,24 @@ static const sph_u32 V_INIT[5][8] = {
 	}
 };
 
+#if SPH_LUFFA_PARALLEL
+
+static const sph_u64 RCW010[8] = {
+   SPH_C64(0xb6de10ed303994a6), SPH_C64(0x70f47aaec0e65299),
+   SPH_C64(0x0707a3d46cc33a12), SPH_C64(0x1c1e8f51dc56983e),
+   SPH_C64(0x707a3d451e00108f), SPH_C64(0xaeb285627800423d),
+   SPH_C64(0xbaca15898f5b7882), SPH_C64(0x40a46f3e96e1db12)
+};
+
+static const sph_u64 RCW014[8] = {
+   SPH_C64(0x01685f3de0337818), SPH_C64(0x05a17cf4441ba90d),
+   SPH_C64(0xbd09caca7f34d442), SPH_C64(0xf4272b289389217f),
+   SPH_C64(0x144ae5cce5a8bce6), SPH_C64(0xfaa7ae2b5274baf4),
+   SPH_C64(0x2e48f1c126889ba7), SPH_C64(0xb923c7049a226e9d)
+};
+
+#else
+
 static const sph_u32 RC00[8] = {
 	SPH_C32(0x303994a6), SPH_C32(0xc0e65299),
 	SPH_C32(0x6cc33a12), SPH_C32(0xdc56983e),
@@ -105,20 +123,18 @@ static const sph_u32 RC14[8] = {
 	SPH_C32(0x2e48f1c1), SPH_C32(0xb923c704)
 };
 
-#if SPH_LUFFA_PARALLEL
-
-static const sph_u64 RCW010[8] = {
-	SPH_C64(0xb6de10ed303994a6), SPH_C64(0x70f47aaec0e65299),
-	SPH_C64(0x0707a3d46cc33a12), SPH_C64(0x1c1e8f51dc56983e),
-	SPH_C64(0x707a3d451e00108f), SPH_C64(0xaeb285627800423d),
-	SPH_C64(0xbaca15898f5b7882), SPH_C64(0x40a46f3e96e1db12)
+static const sph_u32 RC30[8] = {
+   SPH_C32(0xb213afa5), SPH_C32(0xc84ebe95),
+   SPH_C32(0x4e608a22), SPH_C32(0x56d858fe),
+   SPH_C32(0x343b138f), SPH_C32(0xd0ec4e3d),
+   SPH_C32(0x2ceb4882), SPH_C32(0xb3ad2208)
 };
 
-static const sph_u64 RCW014[8] = {
-	SPH_C64(0x01685f3de0337818), SPH_C64(0x05a17cf4441ba90d),
-	SPH_C64(0xbd09caca7f34d442), SPH_C64(0xf4272b289389217f),
-	SPH_C64(0x144ae5cce5a8bce6), SPH_C64(0xfaa7ae2b5274baf4),
-	SPH_C64(0x2e48f1c126889ba7), SPH_C64(0xb923c7049a226e9d)
+static const sph_u32 RC34[8] = {
+   SPH_C32(0xe028c9bf), SPH_C32(0x44756f91),
+   SPH_C32(0x7e8fce32), SPH_C32(0x956548be),
+   SPH_C32(0xfe191be2), SPH_C32(0x3cb226e5),
+   SPH_C32(0x5944a28e), SPH_C32(0xa1c4c355)
 };
 
 #endif
@@ -137,19 +153,6 @@ static const sph_u32 RC24[8] = {
 	SPH_C32(0x36eda57f), SPH_C32(0x703aace7)
 };
 
-static const sph_u32 RC30[8] = {
-	SPH_C32(0xb213afa5), SPH_C32(0xc84ebe95),
-	SPH_C32(0x4e608a22), SPH_C32(0x56d858fe),
-	SPH_C32(0x343b138f), SPH_C32(0xd0ec4e3d),
-	SPH_C32(0x2ceb4882), SPH_C32(0xb3ad2208)
-};
-
-static const sph_u32 RC34[8] = {
-	SPH_C32(0xe028c9bf), SPH_C32(0x44756f91),
-	SPH_C32(0x7e8fce32), SPH_C32(0x956548be),
-	SPH_C32(0xfe191be2), SPH_C32(0x3cb226e5),
-	SPH_C32(0x5944a28e), SPH_C32(0xa1c4c355)
-};
 
 #if SPH_LUFFA_PARALLEL
 

algo/lyra2/allium-4way.c

@@ -44,11 +44,11 @@ void allium_4way_hash( void *state, const void *input )
    blake256_4way( &ctx.blake, input + (64<<2), 16 );
    blake256_4way_close( &ctx.blake, vhash32 );
 
-   mm256_rintrlv_4x32_4x64( vhash64, vhash32, 256 );
+   rintrlv_4x32_4x64( vhash64, vhash32, 256 );
    keccak256_4way( &ctx.keccak, vhash64, 32 );
    keccak256_4way_close( &ctx.keccak, vhash64 );
 
-   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+   dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
 
    LYRA2RE( hash0, 32, hash0, 32, hash0, 32, 1, 8, 8 );
    LYRA2RE( hash1, 32, hash1, 32, hash1, 32, 1, 8, 8 );
@@ -68,12 +68,12 @@ void allium_4way_hash( void *state, const void *input )
    LYRA2RE( hash2, 32, hash2, 32, hash2, 32, 1, 8, 8 );
    LYRA2RE( hash3, 32, hash3, 32, hash3, 32, 1, 8, 8 );
 
-   mm256_intrlv_4x64( vhash64, hash0, hash1, hash2, hash3, 256 );
+   intrlv_4x64( vhash64, hash0, hash1, hash2, hash3, 256 );
 
    skein256_4way( &ctx.skein, vhash64, 32 );
    skein256_4way_close( &ctx.skein, vhash64 );
 
-   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+   dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
 
    update_and_final_groestl256( &ctx.groestl, state, hash0, 256 );
    memcpy( &ctx.groestl, &allium_4way_ctx.groestl,
@@ -103,7 +103,7 @@ int scanhash_allium_4way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ( (uint32_t*)ptarget )[7] = 0x0000ff;
 
-   mm128_bswap_intrlv80_4x32( vdata, pdata );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    blake256_4way_init( &allium_4way_ctx.blake );
    blake256_4way( &allium_4way_ctx.blake, vdata, 64 );
 

algo/lyra2/lyra2h-4way.c

@@ -5,7 +5,7 @@
 #include <memory.h>
 #include <mm_malloc.h>
 #include "lyra2.h"
-#include "algo/blake/sph_blake.h"
+//#include "algo/blake/sph_blake.h"
 #include "algo/blake/blake-hash-4way.h"
 
 __thread uint64_t* lyra2h_4way_matrix;
@@ -64,7 +64,7 @@ int scanhash_lyra2h_4way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ptarget[7] = 0x0000ff;
 
-   mm128_bswap_intrlv80_4x32( vdata, pdata );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    lyra2h_4way_midstate( vdata );
 
    do {

algo/lyra2/lyra2rev2-4way.c

@@ -42,12 +42,12 @@ void lyra2rev2_4way_hash( void *state, const void *input )
    blake256_4way( &ctx.blake, input + (64<<2), 16 );
    blake256_4way_close( &ctx.blake, vhash );
 
-   mm256_rintrlv_4x32_4x64( vhash64, vhash, 256 );
+   rintrlv_4x32_4x64( vhash64, vhash, 256 );
 
    keccak256_4way( &ctx.keccak, vhash64, 32 );
    keccak256_4way_close( &ctx.keccak, vhash64 );
 
-   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+   dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
 
    cubehashUpdateDigest( &ctx.cube, (byte*) hash0, (const byte*) hash0, 32 );
    cubehashInit( &ctx.cube, 256, 16, 32 );
@@ -62,12 +62,12 @@ void lyra2rev2_4way_hash( void *state, const void *input )
    LYRA2REV2( l2v2_wholeMatrix, hash2, 32, hash2, 32, hash2, 32, 1, 4, 4 );
    LYRA2REV2( l2v2_wholeMatrix, hash3, 32, hash3, 32, hash3, 32, 1, 4, 4 );
 
-   mm256_intrlv_4x64( vhash64, hash0, hash1, hash2, hash3, 256 );
+   intrlv_4x64( vhash64, hash0, hash1, hash2, hash3, 256 );
 
    skein256_4way( &ctx.skein, vhash64, 32 );
    skein256_4way_close( &ctx.skein, vhash64 );
 
-   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+   dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
 
    cubehashInit( &ctx.cube, 256, 16, 32 );
    cubehashUpdateDigest( &ctx.cube, (byte*) hash0, (const byte*) hash0, 32 );
@@ -102,7 +102,7 @@ int scanhash_lyra2rev2_4way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ( (uint32_t*)ptarget )[7] = 0x0000ff;
 
-   mm128_bswap_intrlv80_4x32( vdata, pdata );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
 
    blake256_4way_init( &l2v2_4way_ctx.blake );
    blake256_4way( &l2v2_4way_ctx.blake, vdata, 64 );

algo/lyra2/lyra2rev3-4way.c

@@ -41,7 +41,7 @@ void lyra2rev3_8way_hash( void *state, const void *input )
    blake256_8way( &ctx.blake, input, 80 );
    blake256_8way_close( &ctx.blake, vhash );
 
-   mm256_dintrlv_8x32( hash0, hash1, hash2, hash3,
+   dintrlv_8x32( hash0, hash1, hash2, hash3,
                        hash4, hash5, hash6, hash7, vhash, 256 );
 
    LYRA2REV3( l2v3_wholeMatrix, hash0, 32, hash0, 32, hash0, 32, 1, 4, 4 );
@@ -78,7 +78,7 @@ void lyra2rev3_8way_hash( void *state, const void *input )
    LYRA2REV3( l2v3_wholeMatrix, hash6, 32, hash6, 32, hash6, 32, 1, 4, 4 );
    LYRA2REV3( l2v3_wholeMatrix, hash7, 32, hash7, 32, hash7, 32, 1, 4, 4 );
 
-   mm256_intrlv_8x32( vhash, hash0, hash1, hash2, hash3,
+   intrlv_8x32( vhash, hash0, hash1, hash2, hash3,
                              hash4, hash5, hash6, hash7, 256 );
 
    bmw256_8way( &ctx.bmw, vhash, 32 );
@@ -91,7 +91,6 @@ int scanhash_lyra2rev3_8way( struct work *work, uint32_t max_nonce,
 {
    uint32_t hash[8*8] __attribute__ ((aligned (64)));
    uint32_t vdata[20*8] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t *hash7 = &(hash[7<<3]);
    uint32_t lane_hash[8] __attribute__ ((aligned (32)));
    uint32_t *pdata = work->data;
@@ -105,10 +104,7 @@ int scanhash_lyra2rev3_8way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ( (uint32_t*)ptarget )[7] = 0x0000ff;
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
-                              edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_8x32( vdata, pdata );
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
    do
    {
       *noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
@@ -119,7 +115,7 @@ int scanhash_lyra2rev3_8way( struct work *work, uint32_t max_nonce,
 
       for ( int lane = 0; lane < 8; lane++ ) if ( hash7[lane] <= Htarg )
       {
-         mm256_extr_lane_8x32( lane_hash, hash, lane, 256 );
+         extr_lane_8x32( lane_hash, hash, lane, 256 );
          if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
          {
               pdata[19] = n + lane;
@@ -208,7 +204,7 @@ int scanhash_lyra2rev3_4way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ( (uint32_t*)ptarget )[7] = 0x0000ff;
 
-   mm128_bswap_intrlv80_4x32( vdata, pdata );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    do
    {
       *noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );

algo/lyra2/lyra2z-4way.c

@@ -60,7 +60,7 @@ int scanhash_lyra2z_4way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ptarget[7] = 0x0000ff;
 
-   mm128_bswap_intrlv80_4x32( vdata, pdata );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    lyra2z_4way_midstate( vdata );
 
    do {
@@ -119,8 +119,8 @@ void lyra2z_8way_hash( void *state, const void *input )
      blake256_8way( &ctx_blake, input + (64*8), 16 );
      blake256_8way_close( &ctx_blake, vhash );
 
-     mm256_dintrlv_8x32( hash0, hash1, hash2, hash3,
-                         hash4, hash5, hash6, hash7, vhash, 256 );
+     dintrlv_8x32( hash0, hash1, hash2, hash3,
+                   hash4, hash5, hash6, hash7, vhash, 256 );
 
      LYRA2Z( lyra2z_8way_matrix, hash0, 32, hash0, 32, hash0, 32, 8, 8, 8 );
      LYRA2Z( lyra2z_8way_matrix, hash1, 32, hash1, 32, hash1, 32, 8, 8, 8 );
@@ -146,7 +146,6 @@ int scanhash_lyra2z_8way( struct work *work, uint32_t max_nonce,
 {
    uint32_t hash[8*8] __attribute__ ((aligned (64)));
    uint32_t vdata[20*8] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
    const uint32_t Htarg = ptarget[7];
@@ -158,10 +157,7 @@ int scanhash_lyra2z_8way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ptarget[7] = 0x0000ff;
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
-                              edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_8x32( vdata, pdata );
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
    lyra2z_8way_midstate( vdata );
 
    do {

algo/lyra2/phi2-4way.c

@@ -69,13 +69,13 @@ void phi2_hash_4way( void *state, const void *input )
    LYRA2RE( &hashA[3][0], 32, &hashB[3][0], 32, &hashB[3][0], 32, 1, 8, 8 );
    LYRA2RE( &hashA[3][8], 32, &hashB[3][8], 32, &hashB[3][8], 32, 1, 8, 8 );
 
-   mm256_intrlv_4x64( vhash, hashA[0], hashA[1], hashA[2], hashA[3], 512 );
+   intrlv_4x64( vhash, hashA[0], hashA[1], hashA[2], hashA[3], 512 );
 
    jh512_4way_init( &ctx.jh );
    jh512_4way( &ctx.jh, vhash, 64 );
    jh512_4way_close( &ctx.jh, vhash );
 
-   mm256_dintrlv_4x64( hash[0], hash[1], hash[2], hash[3], vhash, 512 );
+   dintrlv_4x64( hash[0], hash[1], hash[2], hash[3], vhash, 512 );
 
    if ( hash[0][0] & 1 )
   	{
@@ -141,7 +141,7 @@ void phi2_hash_4way( void *state, const void *input )
                           (const BitSequence *)hash[3], 512 );
    }
 
-   mm256_intrlv_4x64( vhash, hash[0], hash[1], hash[2], hash[3], 512 );
+   intrlv_4x64( vhash, hash[0], hash[1], hash[2], hash[3], 512 );
    
    skein512_4way_init( &ctx.skein );
 	skein512_4way( &ctx.skein, vhash, 64 );
@@ -217,7 +217,7 @@ int scanhash_phi2_4way( struct work *work, uint32_t max_nonce,
 
       for ( int lane = 0; lane < 4; lane++ ) if (  hash7[ lane<<1 ] < Htarg )
       {
-          mm256_extr_lane_4x64( lane_hash, hash, lane, 256 );
+          extr_lane_4x64( lane_hash, hash, lane, 256 );
           if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
           {
               pdata[19] = n + lane;

algo/nist5/nist5-4way.c

@@ -12,9 +12,6 @@
 #include "algo/keccak/keccak-hash-4way.h"
 #include "algo/groestl/aes_ni/hash-groestl.h"
 
-// no improvement with midstate
-//static __thread blake512_4way_context ctx_mid;
-
 void nist5hash_4way( void *out, const void *input )
 {
      uint64_t hash0[8] __attribute__ ((aligned (64)));
@@ -28,14 +25,11 @@ void nist5hash_4way( void *out, const void *input )
      skein512_4way_context  ctx_skein;
      keccak512_4way_context ctx_keccak;
 
-//     memcpy( &ctx_blake, &ctx_mid, sizeof(ctx_mid) );
-//     blake512_4way( &ctx_blake, input + (64<<2), 16 );
-
      blake512_4way_init( &ctx_blake );
      blake512_4way( &ctx_blake, input, 80 );
      blake512_4way_close( &ctx_blake, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_groestl( &ctx_groestl, 64 );
      update_and_final_groestl( &ctx_groestl, (char*)hash0,
@@ -50,7 +44,7 @@ void nist5hash_4way( void *out, const void *input )
      update_and_final_groestl( &ctx_groestl, (char*)hash3,
                                (const char*)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      jh512_4way_init( &ctx_jh );
      jh512_4way( &ctx_jh, vhash, 64 );
@@ -72,13 +66,12 @@ int scanhash_nist5_4way( struct work *work, uint32_t max_nonce,
      uint32_t *hash7 = &(hash[25]);
      uint32_t lane_hash[8] __attribute__ ((aligned (32)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-     uint32_t endiandata[20] __attribute__((aligned(64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      uint32_t n = pdata[19];
      const uint32_t first_nonce = pdata[19];
      const uint32_t Htarg = ptarget[7];
-     uint32_t *noncep = vdata + 73;   // 9*8 + 1
+     __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
      int thr_id = mythr->id;  // thr_id arg is deprecated
 
      uint64_t htmax[] = {          0,
@@ -95,15 +88,7 @@ int scanhash_nist5_4way( struct work *work, uint32_t max_nonce,
                           0xFFFF0000,
                                    0 };
 
-     // we need bigendian data...
-     swab32_array( endiandata, pdata, 20 );
-
-     uint64_t *edata = (uint64_t*)endiandata;
-     mm256_intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
-
-     // precalc midstate
-//     blake512_4way_init( &ctx_mid );
-//     blake512_4way( &ctx_mid, vdata, 64 );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      for ( int m=0; m < 6; m++ )
      {
@@ -112,17 +97,15 @@ int scanhash_nist5_4way( struct work *work, uint32_t max_nonce,
            uint32_t mask = masks[m];
 
            do {
-              be32enc( noncep,   n   );
-              be32enc( noncep+2, n+1 );
-              be32enc( noncep+4, n+2 );
-              be32enc( noncep+6, n+3 );
+              *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
+                 _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
 
               nist5hash_4way( hash, vdata );
 
               for ( int lane = 0; lane < 4; lane++ )
               if ( ( hash7[ lane ] & mask ) == 0 )
               {
-                 mm256_extr_lane_4x64( lane_hash, hash, lane, 256 );
+                 extr_lane_4x64( lane_hash, hash, lane, 256 );
                  if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
                  {
                     pdata[19] = n + lane;

algo/quark/anime-4way.c

@@ -50,6 +50,7 @@ void anime_4way_hash( void *state, const void *input )
     __m256i vh_mask;
     const uint32_t mask = 8;
     const __m256i bit3_mask = _mm256_set1_epi64x( 8 );
+    const __m256i zero = _mm256_setzero_si256();
     anime_4way_ctx_holder ctx;
     memcpy( &ctx, &anime_4way_ctx, sizeof(anime_4way_ctx) );
 
@@ -59,10 +60,9 @@ void anime_4way_hash( void *state, const void *input )
     blake512_4way( &ctx.blake, vhash, 64 );
     blake512_4way_close( &ctx.blake, vhash );
 
-    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
-                                  m256_zero );
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ), zero );
 
-    mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+    dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
     if ( hash0[0] & mask )
     {
@@ -88,7 +88,7 @@ void anime_4way_hash( void *state, const void *input )
                                                (char*)hash3, 512 );
     }
 
-    mm256_intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
+    intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
 
     if ( mm256_anybits0( vh_mask ) )
     {
@@ -98,7 +98,7 @@ void anime_4way_hash( void *state, const void *input )
 
     mm256_blend_hash_4x64( vh, vhA, vhB, vh_mask );
 
-    mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+    dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
     reinit_groestl( &ctx.groestl );
     update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -109,13 +109,12 @@ void anime_4way_hash( void *state, const void *input )
     reinit_groestl( &ctx.groestl );
     update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-    mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+    intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
     jh512_4way( &ctx.jh, vhash, 64 );
     jh512_4way_close( &ctx.jh, vhash );
 
-    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
-                                  m256_zero );
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ), zero );
 
     if ( mm256_anybits1( vh_mask ) )
     {
@@ -139,8 +138,7 @@ void anime_4way_hash( void *state, const void *input )
     skein512_4way( &ctx.skein, vhash, 64 );
     skein512_4way_close( &ctx.skein, vhash );
 
-    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
-                                  m256_zero );
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ), zero );
 
     if ( mm256_anybits1( vh_mask ) )
     {
@@ -157,7 +155,7 @@ void anime_4way_hash( void *state, const void *input )
 
     mm256_blend_hash_4x64( vh, vhA, vhB, vh_mask );
 
-    mm256_dintrlv_4x64( state, state+32, state+64, state+96, vhash, 256 );
+    dintrlv_4x64( state, state+32, state+64, state+96, vhash, 256 );
 }
 
 int scanhash_anime_4way( struct work *work, uint32_t max_nonce,
@@ -165,7 +163,6 @@ int scanhash_anime_4way( struct work *work, uint32_t max_nonce,
 {
     uint32_t hash[4*8] __attribute__ ((aligned (64)));
     uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
     uint32_t *pdata = work->data;
     uint32_t *ptarget = work->target;
     uint32_t n = pdata[19];
@@ -190,9 +187,7 @@ int scanhash_anime_4way( struct work *work, uint32_t max_nonce,
                 0
         };
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//    mm256_bswap_intrlv80_4x64( vdata, pdata );
+    mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
     for (int m=0; m < 6; m++)
        if (Htarg <= htmax[m])

algo/quark/hmq1725-4way.c

@@ -67,7 +67,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
      bmw512_4way( &ctx.bmw, input, 80 );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_whirlpool_init( &ctx.whirlpool );
      sph_whirlpool( &ctx.whirlpool, hash0, 64 );
@@ -84,7 +84,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
 
 // first fork, A is groestl serial, B is skein parallel.
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], vmask ),
                                    m256_zero );
@@ -116,7 +116,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
                                                (char*)hash3, 512 );
 //     }
 
-     mm256_intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
 
 // B
 
@@ -158,7 +158,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
 
      mm256_blend_hash_4x64( vh, vhA, vhB, vh_mask );
     
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_luffa( &ctx.luffa, 512 );
      update_and_final_luffa( &ctx.luffa, (BitSequence*)hash0,
@@ -186,7 +186,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
      cubehashUpdateDigest( &ctx.cube, (BitSequence *)hash3,
                                 (const BitSequence *)hash3, 64 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
 // A= keccak parallel, B= jh parallel
     
@@ -209,7 +209,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
 
      mm256_blend_hash_4x64( vh, vhA, vhB, vh_mask );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_shavite512_init( &ctx.shavite );
      sph_shavite512 ( &ctx.shavite, hash0, 64 );
@@ -240,7 +240,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
 // A is whirlpool serial, B is haval parallel.
     
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], vmask ),
                                    m256_zero );
@@ -271,7 +271,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
         sph_whirlpool_close( &ctx.whirlpool, hash3 );
 //     }
 
-     mm256_intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
 
 // B
 
@@ -285,7 +285,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
 
      mm256_blend_hash_4x64( vh, vhA, vhB, vh_mask );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
     
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -300,13 +300,13 @@ extern void hmq1725_4way_hash(void *state, const void *input)
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                              (const BitSequence *)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
      
      blake512_4way_init( &ctx.blake );
      blake512_4way( &ctx.blake, vhash, 64 );
      blake512_4way_close( &ctx.blake, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
 // shavite & luffa, both serial, select individually.
 
@@ -362,13 +362,13 @@ extern void hmq1725_4way_hash(void *state, const void *input)
                                     (const BitSequence *)hash3, 64 );
    }
 
-   mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+   intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
    hamsi512_4way_init( &ctx.hamsi );
    hamsi512_4way( &ctx.hamsi, vhash, 64 );
    hamsi512_4way_close( &ctx.hamsi, vhash );
 
-   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+   dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
    sph_fugue512_init( &ctx.fugue );
    sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -438,13 +438,13 @@ extern void hmq1725_4way_hash(void *state, const void *input)
                              (const BitSequence *)hash3, 512 );
    }
 
-   mm128_intrlv_4x32( vhash, hash0, hash1, hash2, hash3, 512 );
+   intrlv_4x32( vhash, hash0, hash1, hash2, hash3, 512 );
 
    shabal512_4way_init( &ctx.shabal );
    shabal512_4way( &ctx.shabal, vhash, 64 );
    shabal512_4way_close( &ctx.shabal, vhash );
 
-   mm128_dintrlv_4x32( hash0, hash1, hash2, hash3, vhash, 512 );
+   dintrlv_4x32( hash0, hash1, hash2, hash3, vhash, 512 );
 
    sph_whirlpool_init( &ctx.whirlpool );
    sph_whirlpool( &ctx.whirlpool, hash0, 64 );
@@ -461,7 +461,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
 
 // A = fugue serial, B = sha512 prarallel
    
-   mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+   intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], vmask ),
                                  m256_zero );
@@ -491,7 +491,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
       sph_fugue512_close( &ctx.fugue, hash3 );
 //   }
 
-   mm256_intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
+   intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
 
 //   if ( mm256_any_clr_256( vh_mask ) )
 //   {
@@ -502,7 +502,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
 
    mm256_blend_hash_4x64( vh, vhA, vhB, vh_mask );
 
-   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+   dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
    init_groestl( &ctx.groestl, 64 );
    update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -513,7 +513,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
    init_groestl( &ctx.groestl, 64 );
    update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-   mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+   intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
    sha512_4way_init( &ctx.sha512 ); 
    sha512_4way( &ctx.sha512, vhash, 64 );
@@ -524,7 +524,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], vmask ),
                                  m256_zero );
 
-   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+   dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
      
 //   if ( mm256_any_set_256( vh_mask ) ) //4
 //   {
@@ -559,7 +559,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
       sph_whirlpool_close( &ctx.whirlpool, hash3 );
 //   }
 
-   mm256_intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, 512 );
+   intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, 512 );
 
    mm256_blend_hash_4x64( vh, vhA, vhB, vh_mask );
 
@@ -589,7 +589,7 @@ int scanhash_hmq1725_4way( struct work *work, uint32_t max_nonce,
    uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
                         0xFFFFF000, 0xFFFF0000,          0  };
 
-   mm256_bswap_intrlv80_4x64( vdata, pdata );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
    for ( int m = 0; m < 6; m++ ) if ( Htarg <= htmax[m] )
    {
       uint32_t mask = masks[ m ];

algo/quark/quark-4way.c

@@ -51,6 +51,7 @@ void quark_4way_hash( void *state, const void *input )
     quark_4way_ctx_holder ctx;
     const __m256i bit3_mask = _mm256_set1_epi64x( 8 );
     const uint32_t mask = 8;
+    const __m256i zero = _mm256_setzero_si256();
 
     memcpy( &ctx, &quark_4way_ctx, sizeof(quark_4way_ctx) );
 
@@ -60,10 +61,9 @@ void quark_4way_hash( void *state, const void *input )
     bmw512_4way( &ctx.bmw, vhash, 64 );
     bmw512_4way_close( &ctx.bmw, vhash );
 
-    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
-                                  m256_zero );
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ), zero );
 
-    mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+    dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
     if ( hash0[0] & mask )
     {
@@ -89,7 +89,7 @@ void quark_4way_hash( void *state, const void *input )
                                                (char*)hash3, 512 );
     }
 
-    mm256_intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
+    intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
 
     if ( mm256_anybits0( vh_mask ) )   
     {
@@ -99,7 +99,7 @@ void quark_4way_hash( void *state, const void *input )
 
     mm256_blend_hash_4x64( vh, vhA, vhB, vh_mask );
 
-    mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+    dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
     reinit_groestl( &ctx.groestl );
     update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -110,13 +110,12 @@ void quark_4way_hash( void *state, const void *input )
     reinit_groestl( &ctx.groestl );
     update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-    mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+    intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
     jh512_4way( &ctx.jh, vhash, 64 );
     jh512_4way_close( &ctx.jh, vhash );
 
-    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
-                                  m256_zero );
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ), zero );
 
     if ( mm256_anybits1( vh_mask ) )
     {
@@ -141,8 +140,7 @@ void quark_4way_hash( void *state, const void *input )
     skein512_4way( &ctx.skein, vhash, 64 );
     skein512_4way_close( &ctx.skein, vhash );
 
-    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
-                                  m256_zero );
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ), zero );
 
     if ( mm256_anybits1( vh_mask ) )
     {
@@ -170,7 +168,6 @@ int scanhash_quark_4way( struct work *work, uint32_t max_nonce,
 {
     uint32_t hash[4*8] __attribute__ ((aligned (64)));
     uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
     uint32_t lane_hash[8] __attribute__ ((aligned (64)));
     uint32_t *hash7 = &(hash[25]);
     uint32_t *pdata = work->data;
@@ -180,9 +177,7 @@ int scanhash_quark_4way( struct work *work, uint32_t max_nonce,
     __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
     int thr_id = mythr->id;  // thr_id arg is deprecated
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//    mm256_bswap_intrlv80_4x64( vdata, pdata );
+    mm256_bswap32_intrlv80_4x64( vdata, pdata );
     do
     {
        *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
@@ -194,7 +189,7 @@ int scanhash_quark_4way( struct work *work, uint32_t max_nonce,
        for ( int i = 0; i < 4; i++ )
        if ( ( hash7[ i<<1 ] & 0xFFFFFF00 ) == 0 )
        {
-          mm256_extr_lane_4x64( lane_hash, hash, i, 256 );
+          extr_lane_4x64( lane_hash, hash, i, 256 );
           if ( fulltest( lane_hash, ptarget ) && !opt_benchmark  )
           {
             pdata[19] = n+i;

algo/qubit/deep-2way.c

@@ -39,7 +39,7 @@ void deep_2way_hash( void *output, const void *input )
      memcpy( &ctx, &deep_2way_ctx, sizeof(deep_2way_ctx) );
      luffa_2way_update( &ctx.luffa, input + (64<<1), 16 );
      luffa_2way_close( &ctx.luffa, vhash );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
 
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
                            (const byte*) hash0, 64 );
@@ -86,7 +86,7 @@ int scanhash_deep_2way( struct work *work,uint32_t max_nonce,
      casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
 
      uint64_t *edata = (uint64_t*)endiandata;
-     mm256_intrlv_2x128( (uint64_t*)vdata, edata, edata, 640 );
+     intrlv_2x128( (uint64_t*)vdata, edata, edata, 640 );
 
      luffa_2way_init( &deep_2way_ctx.luffa, 512 );
      luffa_2way_update( &deep_2way_ctx.luffa, vdata, 64 );

algo/qubit/qubit-2way.c

@@ -41,7 +41,7 @@ void qubit_2way_hash( void *output, const void *input )
      memcpy( &ctx, &qubit_2way_ctx, sizeof(qubit_2way_ctx) );
      luffa_2way_update( &ctx.luffa, input + (64<<1), 16 );
      luffa_2way_close( &ctx.luffa, vhash );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
 
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
                            (const byte*) hash0, 64 );
@@ -55,9 +55,9 @@ void qubit_2way_hash( void *output, const void *input )
      sph_shavite512( &ctx.shavite, hash1, 64 );
      sph_shavite512_close( &ctx.shavite, hash1 );
 
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
 
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
                        (const BitSequence *) hash0, 512 );
@@ -92,7 +92,7 @@ int scanhash_qubit_2way( struct work *work,uint32_t max_nonce,
      casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
 
      uint64_t *edata = (uint64_t*)endiandata;
-     mm256_intrlv_2x128( (uint64_t*)vdata, edata, edata, 640 );
+     intrlv_2x128( (uint64_t*)vdata, edata, edata, 640 );
 
      luffa_2way_init( &qubit_2way_ctx.luffa, 512 );
      luffa_2way_update( &qubit_2way_ctx.luffa, vdata, 64 );

algo/ripemd/lbry-4way.c

@@ -40,9 +40,9 @@ void lbry_8way_hash( void* output, const void* input )
    sha256_8way_close( &ctx_sha256, vhashA );
 
    // reinterleave to do sha512 4-way 64 bit twice.
-   mm256_dintrlv_8x32( h0, h1, h2, h3, h4, h5, h6, h7, vhashA, 256 );
-   mm256_intrlv_4x64( vhashA, h0, h1, h2, h3, 256 );
-   mm256_intrlv_4x64( vhashB, h4, h5, h6, h7, 256 );
+   dintrlv_8x32( h0, h1, h2, h3, h4, h5, h6, h7, vhashA, 256 );
+   intrlv_4x64( vhashA, h0, h1, h2, h3, 256 );
+   intrlv_4x64( vhashB, h4, h5, h6, h7, 256 );
 
    sha512_4way_init( &ctx_sha512 );
    sha512_4way( &ctx_sha512, vhashA, 32 );
@@ -53,9 +53,9 @@ void lbry_8way_hash( void* output, const void* input )
    sha512_4way_close( &ctx_sha512, vhashB );
 
    // back to 8-way 32 bit
-   mm256_dintrlv_4x64( h0, h1, h2, h3, vhashA, 512 );
-   mm256_dintrlv_4x64( h4, h5, h6, h7, vhashB, 512 );
-   mm256_intrlv_8x32( vhashA, h0, h1, h2, h3, h4, h5, h6, h7, 512 );
+   dintrlv_4x64( h0, h1, h2, h3, vhashA, 512 );
+   dintrlv_4x64( h4, h5, h6, h7, vhashB, 512 );
+   intrlv_8x32( vhashA, h0, h1, h2, h3, h4, h5, h6, h7, 512 );
 
    ripemd160_8way_init( &ctx_ripemd );
    ripemd160_8way( &ctx_ripemd, vhashA, 32 );
@@ -97,11 +97,15 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
                         0xFFFFF000, 0xFFFF0000,          0 };
 
    // we need bigendian data...
-   casti_m256i( edata, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
-   casti_m256i( edata, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
-   casti_m256i( edata, 2 ) = mm256_bswap_32( casti_m256i( pdata, 2 ) );
-   casti_m256i( edata, 3 ) = mm256_bswap_32( casti_m256i( pdata, 3 ) );
-   mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
+   casti_m128i( edata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
+   casti_m128i( edata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
+   casti_m128i( edata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
+   casti_m128i( edata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
+   casti_m128i( edata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
+   casti_m128i( edata, 5 ) = mm128_bswap_32( casti_m128i( pdata, 5 ) );
+   casti_m128i( edata, 6 ) = mm128_bswap_32( casti_m128i( pdata, 6 ) );
+   casti_m128i( edata, 7 ) = mm128_bswap_32( casti_m128i( pdata, 7 ) );
+   intrlv_8x32( vdata, edata, edata, edata, edata,
                              edata, edata, edata, edata, 1024 );
    sha256_8way_init( &sha256_8w_mid );
    sha256_8way( &sha256_8w_mid, vdata, LBRY_MIDSTATE );
@@ -118,7 +122,7 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
          for ( int i = 0; i < 8; i++ )  if ( !( hash7[ i ] & mask ) )
          {
             // deinterleave hash for lane
-            mm256_extr_lane_8x32( lane_hash, hash, i, 256 );
+            extr_lane_8x32( lane_hash, hash, i, 256 );
             if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
             {
               pdata[27] = n + i;

algo/sha/md-helper-4way.c

@@ -196,9 +196,9 @@ SPH_XCAT( HASH, _addbits_and_close )(void *cc, 	unsigned ub, unsigned n,
     ptr = (unsigned)sc->count & (SPH_BLEN - 1U);
 
 #ifdef PW01
-    sc->buf[ptr>>3] = _mm256_set1_epi64x( 0x100 >> 8 );
+    sc->buf[ptr>>3] = m256_const1_64( 0x100 >> 8 );
 #else
-    sc->buf[ptr>>3] = _mm256_set1_epi64x( 0x80 );
+    sc->buf[ptr>>3] = m256_const1_64( 0x80 );
 #endif
     ptr += 8;
 

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

@@ -86,8 +86,7 @@ static const sph_u32 K256[64] = {
 // SHA-256 4 way
 
 #define SHA2s_MEXP( a, b, c, d ) \
-     _mm_add_epi32( _mm_add_epi32( _mm_add_epi32( \
-                    SSG2_1( W[a] ), W[b] ), SSG2_0( W[c] ) ), W[d] );
+  mm128_add4_32( SSG2_1( W[a] ), W[b], SSG2_0( W[c] ), W[d] );
 
 #define CHs(X, Y, Z) \
    _mm_xor_si128( _mm_and_si128( _mm_xor_si128( Y, Z ), X ), Z ) 
@@ -115,9 +114,8 @@ static const sph_u32 K256[64] = {
 #define SHA2s_4WAY_STEP(A, B, C, D, E, F, G, H, i, j) \
 do { \
   register __m128i T1, T2; \
-  T1 = _mm_add_epi32( _mm_add_epi32( _mm_add_epi32( \
-       _mm_add_epi32( H, BSG2_1(E) ), CHs(E, F, G) ), \
-                          _mm_set1_epi32( K256[( (j)+(i) )] ) ), W[i] ); \
+  T1 = _mm_add_epi32( H, mm128_add4_32( BSG2_1(E), CHs(E, F, G), \
+                             _mm_set1_epi32( K256[( (j)+(i) )] ), W[i] ) ); \
   T2 = _mm_add_epi32( BSG2_0(A), MAJs(A, B, C) ); \
   D  = _mm_add_epi32( D,  T1 ); \
   H  = _mm_add_epi32( T1, T2 ); \
@@ -129,22 +127,8 @@ sha256_4way_round( __m128i *in, __m128i r[8] )
    register  __m128i A, B, C, D, E, F, G, H;
    __m128i W[16];
 
-   W[ 0] = mm128_bswap_32( in[ 0] );
-   W[ 1] = mm128_bswap_32( in[ 1] );
-   W[ 2] = mm128_bswap_32( in[ 2] );
-   W[ 3] = mm128_bswap_32( in[ 3] );
-   W[ 4] = mm128_bswap_32( in[ 4] );
-   W[ 5] = mm128_bswap_32( in[ 5] );
-   W[ 6] = mm128_bswap_32( in[ 6] );
-   W[ 7] = mm128_bswap_32( in[ 7] );
-   W[ 8] = mm128_bswap_32( in[ 8] );
-   W[ 9] = mm128_bswap_32( in[ 9] );
-   W[10] = mm128_bswap_32( in[10] );
-   W[11] = mm128_bswap_32( in[11] );
-   W[12] = mm128_bswap_32( in[12] );
-   W[13] = mm128_bswap_32( in[13] );
-   W[14] = mm128_bswap_32( in[14] );
-   W[15] = mm128_bswap_32( in[15] );
+   mm128_block_bswap_32( W, in );
+   mm128_block_bswap_32( W+8, in+8 );
 
    A = r[0];
    B = r[1];
@@ -266,7 +250,7 @@ void sha256_4way( sha256_4way_context *sc, const void *data, size_t len )
 
 void sha256_4way_close( sha256_4way_context *sc, void *dst )
 {
-    unsigned ptr, u;
+    unsigned ptr;
     uint32_t low, high;
     const int buf_size = 64;
     const int pad = buf_size - 8;
@@ -294,8 +278,7 @@ void sha256_4way_close( sha256_4way_context *sc, void *dst )
                  mm128_bswap_32( _mm_set1_epi32( low ) );
     sha256_4way_round( sc->buf, sc->val );
 
-    for ( u = 0; u < 8; u ++ )
-       ((__m128i*)dst)[u] = mm128_bswap_32( sc->val[u] );
+    mm128_block_bswap_32( dst, sc->val );
 }
 
 #if defined(__AVX2__)
@@ -326,15 +309,13 @@ void sha256_4way_close( sha256_4way_context *sc, void *dst )
        mm256_ror_32(x, 17), mm256_ror_32(x, 19) ), _mm256_srli_epi32(x, 10) )
 
 #define SHA2x_MEXP( a, b, c, d ) \
-     _mm256_add_epi32( _mm256_add_epi32( _mm256_add_epi32( \
-                    SSG2_1x( W[a] ), W[b] ), SSG2_0x( W[c] ) ), W[d] );
+     mm256_add4_32( SSG2_1x( W[a] ), W[b], SSG2_0x( W[c] ), W[d] );
 
 #define SHA2s_8WAY_STEP(A, B, C, D, E, F, G, H, i, j) \
 do { \
   register __m256i T1, T2; \
-  T1 = _mm256_add_epi32( _mm256_add_epi32( _mm256_add_epi32( \
-       _mm256_add_epi32( H, BSG2_1x(E) ), CHx(E, F, G) ), \
-                          _mm256_set1_epi32( K256[( (j)+(i) )] ) ), W[i] ); \
+   T1 = _mm256_add_epi32( H, mm256_add4_32( BSG2_1x(E), CHx(E, F, G), \
+                          _mm256_set1_epi32( K256[( (j)+(i) )] ), W[i] ) ); \
   T2 = _mm256_add_epi32( BSG2_0x(A), MAJx(A, B, C) ); \
   D  = _mm256_add_epi32( D,  T1 ); \
   H  = _mm256_add_epi32( T1, T2 ); \
@@ -346,22 +327,8 @@ sha256_8way_round( __m256i *in, __m256i r[8] )
    register  __m256i A, B, C, D, E, F, G, H;
    __m256i W[16];
 
-   W[ 0] = mm256_bswap_32( in[ 0] );
-   W[ 1] = mm256_bswap_32( in[ 1] );
-   W[ 2] = mm256_bswap_32( in[ 2] );
-   W[ 3] = mm256_bswap_32( in[ 3] );
-   W[ 4] = mm256_bswap_32( in[ 4] );
-   W[ 5] = mm256_bswap_32( in[ 5] );
-   W[ 6] = mm256_bswap_32( in[ 6] );
-   W[ 7] = mm256_bswap_32( in[ 7] );
-   W[ 8] = mm256_bswap_32( in[ 8] );
-   W[ 9] = mm256_bswap_32( in[ 9] );
-   W[10] = mm256_bswap_32( in[10] );
-   W[11] = mm256_bswap_32( in[11] );
-   W[12] = mm256_bswap_32( in[12] );
-   W[13] = mm256_bswap_32( in[13] );
-   W[14] = mm256_bswap_32( in[14] );
-   W[15] = mm256_bswap_32( in[15] );
+   mm256_block_bswap_32( W  , in   );
+   mm256_block_bswap_32( W+8, in+8 );
 
    A = r[0];
    B = r[1];
@@ -484,7 +451,7 @@ void sha256_8way( sha256_8way_context *sc, const void *data, size_t len )
 
 void sha256_8way_close( sha256_8way_context *sc, void *dst )
 {
-    unsigned ptr, u;
+    unsigned ptr;
     uint32_t low, high;
     const int buf_size = 64;
     const int pad = buf_size - 8;
@@ -513,8 +480,7 @@ void sha256_8way_close( sha256_8way_context *sc, void *dst )
 
     sha256_8way_round( sc->buf, sc->val );
 
-    for ( u = 0; u < 8; u ++ )
-       ((__m256i*)dst)[u] = mm256_bswap_32( sc->val[u] );
+    mm256_block_bswap_32( dst, sc->val );
 }
 
 
@@ -596,9 +562,8 @@ static const sph_u64 K512[80] = {
 #define SHA3_4WAY_STEP(A, B, C, D, E, F, G, H, i) \
 do { \
   register __m256i T1, T2; \
-  T1 = _mm256_add_epi64( _mm256_add_epi64( _mm256_add_epi64( \
-       _mm256_add_epi64( H, BSG5_1(E) ), CH(E, F, G) ), \
-                         _mm256_set1_epi64x( K512[i] ) ), W[i] ); \
+  T1 = _mm256_add_epi64( H, mm256_add4_64( BSG5_1(E), CH(E, F, G), \
+                                _mm256_set1_epi64x( K512[i] ), W[i] ) ); \
   T2 = _mm256_add_epi64( BSG5_0(A), MAJ(A, B, C) ); \
   D  = _mm256_add_epi64( D, T1 ); \
   H  = _mm256_add_epi64( T1, T2 ); \
@@ -611,11 +576,12 @@ sha512_4way_round( __m256i *in, __m256i r[8] )
    register __m256i A, B, C, D, E, F, G, H;
    __m256i W[80];
 
-   for ( i = 0; i < 16; i++ )
-      W[i] = mm256_bswap_64( in[i] );
+   mm256_block_bswap_64( W  , in );
+   mm256_block_bswap_64( W+8, in+8 );
+
    for ( i = 16; i < 80; i++ )
-      W[i] = _mm256_add_epi64( _mm256_add_epi64( _mm256_add_epi64(
-           SSG5_1( W[ i-2 ] ), W[ i-7 ] ), SSG5_0( W[ i-15 ] ) ), W[ i-16 ] );
+      W[i] = mm256_add4_64( SSG5_1( W[ i- 2 ] ), W[ i- 7 ],
+                            SSG5_0( W[ i-15 ] ), W[ i-16 ] );
 
    A = r[0];
    B = r[1];
@@ -689,12 +655,12 @@ void sha512_4way( sha512_4way_context *sc, const void *data, size_t len )
 
 void sha512_4way_close( sha512_4way_context *sc, void *dst )
 {
-    unsigned ptr, u;
+    unsigned ptr;
     const int buf_size = 128;
     const int pad = buf_size - 16;
 
     ptr = (unsigned)sc->count & (buf_size - 1U);
-    sc->buf[ ptr>>3 ] = _mm256_set1_epi64x( 0x80 );
+    sc->buf[ ptr>>3 ] = m256_const1_64( 0x80 );
     ptr += 8;
     if ( ptr > pad )
     {
@@ -711,8 +677,7 @@ void sha512_4way_close( sha512_4way_context *sc, void *dst )
                  mm256_bswap_64( _mm256_set1_epi64x( sc->count << 3 ) );
     sha512_4way_round( sc->buf, sc->val );
 
-    for ( u = 0; u < 8; u ++ )
-       ((__m256i*)dst)[u] = mm256_bswap_64( sc->val[u] );
+    mm256_block_bswap_64( dst, sc->val );
 }
 
 #endif  // __AVX2__

algo/sha/sha256q-4way.c

@@ -36,7 +36,6 @@ int scanhash_sha256q_8way( struct work *work, uint32_t max_nonce,
 {
    uint32_t vdata[20*8] __attribute__ ((aligned (64)));
    uint32_t hash[8*8] __attribute__ ((aligned (32)));
-   uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t lane_hash[8] __attribute__ ((aligned (32)));
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
@@ -60,10 +59,7 @@ int scanhash_sha256q_8way( struct work *work, uint32_t max_nonce,
                                         0 };
 
    // Need big endian data
-   swab32_array( edata, pdata, 20 );
-   mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
-                             edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_8x32( vdata, pdata );
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
    sha256_8way_init( &sha256_ctx8 );
    sha256_8way( &sha256_ctx8, vdata, 64 );
 
@@ -84,7 +80,7 @@ int scanhash_sha256q_8way( struct work *work, uint32_t max_nonce,
          if ( !( hash7[ lane ] & mask ) )
          { 
             // deinterleave hash for lane
-	         mm256_extr_lane_8x32( lane_hash, hash, lane, 256 );
+	         extr_lane_8x32( lane_hash, hash, lane, 256 );
 
 	         if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
             {
@@ -133,7 +129,6 @@ int scanhash_sha256q_4way( struct work *work, uint32_t max_nonce,
 {
    uint32_t vdata[20*4] __attribute__ ((aligned (64)));
    uint32_t hash[8*4] __attribute__ ((aligned (32)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t *hash7 = &(hash[7<<2]);
    uint32_t lane_hash[8] __attribute__ ((aligned (32)));
    uint32_t *pdata = work->data;
@@ -157,9 +152,7 @@ int scanhash_sha256q_4way( struct work *work, uint32_t max_nonce,
                                0xFFFF0000,
                                         0 };
 
-   swab32_array( edata, pdata, 20 );
-   mm128_intrlv_4x32( vdata, edata, edata, edata, edata, 640 );
-//   mm128_bswap_intrlv80_4x32( vdata, pdata );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    sha256_4way_init( &sha256_ctx4 );
    sha256_4way( &sha256_ctx4, vdata, 64 );
 

algo/sha/sha256t-4way.c

@@ -72,7 +72,7 @@ int scanhash_sha256t_11way( struct work *work, uint32_t max_nonce,
    casti_m256i( dataz, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
    casti_m128i( dataz, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
 
-   mm256_intrlv_8x32( datax, dataz, dataz, dataz, dataz,
+   intrlv_8x32( datax, dataz, dataz, dataz, dataz,
                                  dataz, dataz, dataz, dataz, 640 );
    mm64_interleave_2x32( datay, dataz, dataz, 640 );
 
@@ -99,7 +99,7 @@ int scanhash_sha256t_11way( struct work *work, uint32_t max_nonce,
         for ( i = 0; i < 8; i++ ) if ( !( hash7[ i ] & mask ) )
         { 
             // deinterleave hash for lane
-            mm256_extr_lane_8x32( lane_hash, hashx, i, 256 );
+            extr_lane_8x32( lane_hash, hashx, i, 256 );
             if ( fulltest( lane_hash, ptarget ) )
             {
 	            pdata[19] = n + i;
@@ -163,7 +163,6 @@ int scanhash_sha256t_8way( struct work *work, uint32_t max_nonce,
 {
    uint32_t vdata[20*8]  __attribute__ ((aligned (64)));
    uint32_t hash[8*8]    __attribute__ ((aligned (32)));
-   uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t lane_hash[8] __attribute__ ((aligned (32)));
    uint32_t *hash7 = &(hash[7<<3]);
    uint32_t *pdata = work->data;
@@ -187,12 +186,9 @@ int scanhash_sha256t_8way( struct work *work, uint32_t max_nonce,
                                0xFFFF0000,
                                         0 };
 
-   swab32_array( edata, pdata, 20 );
-   mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
-                             edata, edata, edata, edata, 640 );
 
    // Need big endian data
-//   mm256_bswap_intrlv80_8x32( vdata, pdata );
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
    sha256_8way_init( &sha256_ctx8 );
    sha256_8way( &sha256_ctx8, vdata, 64 );
 
@@ -209,7 +205,7 @@ int scanhash_sha256t_8way( struct work *work, uint32_t max_nonce,
          if ( !( hash7[ lane ] & mask ) )
          {
             // deinterleave hash for lane
-            mm256_extr_lane_8x32( lane_hash, hash, lane, 256 );
+            extr_lane_8x32( lane_hash, hash, lane, 256 );
             if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
             {
               pdata[19] = n + lane;
@@ -253,7 +249,6 @@ int scanhash_sha256t_4way( struct work *work, uint32_t max_nonce,
 {
    uint32_t vdata[20*4] __attribute__ ((aligned (64)));
    uint32_t hash[8*4] __attribute__ ((aligned (32)));
-   uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t lane_hash[8] __attribute__ ((aligned (64)));
    uint32_t *hash7 = &(hash[7<<2]);
    uint32_t *pdata = work->data;
@@ -277,10 +272,7 @@ int scanhash_sha256t_4way( struct work *work, uint32_t max_nonce,
                               0xFFFF0000,
                                        0 };
 
-   swab32_array( edata, pdata, 20 );
-   mm128_intrlv_4x32( vdata, edata, edata, edata, edata, 640 );
-
-//   mm128_bswap_intrlv80_4x32( vdata, pdata );
+   mm128_bswap32_intrlv80_4x32( vdata, pdata );
    sha256_4way_init( &sha256_ctx4 );
    sha256_4way( &sha256_ctx4, vdata, 64 );
 

algo/sha/sha256t-gate.c

@@ -11,7 +11,7 @@ bool register_sha256t_algo( algo_gate_t* gate )
     gate->scanhash   = (void*)&scanhash_sha256t_4way;
     gate->hash       = (void*)&sha256t_4way_hash;
 #else
-gate->optimizations = SHA_OPT;
+    gate->optimizations = SHA_OPT;
     gate->scanhash   = (void*)&scanhash_sha256t;
     gate->hash       = (void*)&sha256t_hash;
 #endif
@@ -21,7 +21,11 @@ gate->optimizations = SHA_OPT;
 
 bool register_sha256q_algo( algo_gate_t* gate )
 {
-#if defined(SHA256T_4WAY)
+#if defined(SHA256T_8WAY)
+    gate->optimizations = SSE2_OPT | AVX2_OPT | SHA_OPT;
+    gate->scanhash   = (void*)&scanhash_sha256q_8way;
+    gate->hash       = (void*)&sha256q_8way_hash;
+#elif defined(SHA256T_4WAY)
     gate->optimizations = SSE2_OPT | AVX2_OPT | SHA_OPT;
     gate->scanhash   = (void*)&scanhash_sha256q_4way;
     gate->hash       = (void*)&sha256q_4way_hash;

algo/sha/sha256t-gate.h

@@ -8,7 +8,7 @@
 #if !defined(__SHA__)
  #if defined(__AVX2__)
   #define SHA256T_8WAY
-#elif defined(__SSE2__)
+ #elif defined(__SSE2__)
   #define SHA256T_4WAY
  #endif
 #endif

algo/shavite/shavite-hash-2way.c

@@ -20,6 +20,7 @@ static const uint32_t IV512[] =
 static void
 c512_2way( shavite512_2way_context *ctx, const void *msg )
 {
+   const __m128i zero = _mm_setzero_si128();
    __m256i p0, p1, p2, p3, x;
    __m256i k00, k01, k02, k03, k10, k11, k12, k13;
    __m256i *m = (__m256i*)msg;
@@ -33,24 +34,24 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
 
    // round
    k00 = m[0];
-   x = mm256_aesenc_2x128( _mm256_xor_si256( p1, k00 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( p1, k00 ), zero );
    k01 = m[1];
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
    k02 = m[2];
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
    k03 = m[3];
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
    p0 = _mm256_xor_si256( p0, x );
 
    k10 = m[4];
-   x = mm256_aesenc_2x128( _mm256_xor_si256( p3, k10 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( p3, k10 ), zero );
    k11 = m[5];
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
    k12 = m[6];
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
    k13 = m[7];
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );
 
    p2 = _mm256_xor_si256( p2, x );
 
@@ -59,129 +60,129 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
       // round 1, 5, 9
 
      k00 = _mm256_xor_si256( k13, mm256_ror1x32_128(
-                                  mm256_aesenc_2x128( k00 ) ) );
+                                  mm256_aesenc_2x128( k00, zero ) ) );
 
      if ( r == 0 )
         k00 = _mm256_xor_si256( k00, _mm256_set_epi32( 
 		      ~ctx->count3, ctx->count2, ctx->count1, ctx->count0,
                       ~ctx->count3, ctx->count2, ctx->count1, ctx->count0 ) );
 
-     x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k00 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k00 ), zero );
      k01 = _mm256_xor_si256( k00,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k01 ) ) );
+		     mm256_ror1x32_128( mm256_aesenc_2x128( k01, zero ) ) );
 
      if ( r == 1 )
         k01 = _mm256_xor_si256( k01, _mm256_set_epi32(
 	               ~ctx->count0, ctx->count1, ctx->count2, ctx->count3,
                        ~ctx->count0, ctx->count1, ctx->count2, ctx->count3 ) );
 
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
      k02 = _mm256_xor_si256( k01,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k02 ) ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ) );
+		     mm256_ror1x32_128( mm256_aesenc_2x128( k02, zero ) ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
      k03 = _mm256_xor_si256( k02,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k03 ) ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ) );
+		     mm256_ror1x32_128( mm256_aesenc_2x128( k03, zero ) ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
      p3 = _mm256_xor_si256( p3, x );
 
      k10 = _mm256_xor_si256( k03,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k10 ) ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k10 ) );
+		     mm256_ror1x32_128( mm256_aesenc_2x128( k10, zero ) ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k10 ), zero );
      k11 = _mm256_xor_si256( k10,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k11 ) ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ) );
+		     mm256_ror1x32_128( mm256_aesenc_2x128( k11, zero ) ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
      k12 = _mm256_xor_si256( k11,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k12 ) ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ) );
+		     mm256_ror1x32_128( mm256_aesenc_2x128( k12, zero ) ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
      k13 = _mm256_xor_si256( k12,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k13 ) ) );
+		     mm256_ror1x32_128( mm256_aesenc_2x128( k13, zero ) ) );
 
      if ( r == 2 )
         k13 = _mm256_xor_si256( k13, _mm256_set_epi32(
                   ~ctx->count1, ctx->count0, ctx->count3, ctx->count2,
                   ~ctx->count1, ctx->count0, ctx->count3, ctx->count2 ) );
  
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );
      p1 = _mm256_xor_si256( p1, x );
      
      // round 2, 6, 10
 
      k00 = _mm256_xor_si256( k00, mm256_ror2x256hi_1x32( k12, k13 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( p3, k00 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( p3, k00 ), zero );
      k01 = _mm256_xor_si256( k01, mm256_ror2x256hi_1x32( k13, k00 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
      k02 = _mm256_xor_si256( k02, mm256_ror2x256hi_1x32( k00, k01 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
      k03 = _mm256_xor_si256( k03, mm256_ror2x256hi_1x32( k01, k02 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
      p2 = _mm256_xor_si256( p2, x );
 
      k10 = _mm256_xor_si256( k10, mm256_ror2x256hi_1x32( k02, k03 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( p1, k10 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( p1, k10 ), zero );
      k11 = _mm256_xor_si256( k11, mm256_ror2x256hi_1x32( k03, k10 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
      k12 = _mm256_xor_si256( k12, mm256_ror2x256hi_1x32( k10, k11 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
      k13 = _mm256_xor_si256( k13, mm256_ror2x256hi_1x32( k11, k12 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );
 
      p0 = _mm256_xor_si256( p0, x );
 
      // round 3, 7, 11
 
      k00 = _mm256_xor_si256( mm256_ror1x32_128(
-                                     mm256_aesenc_2x128( k00 ) ), k13 );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k00 ) );
+                                     mm256_aesenc_2x128( k00, zero ) ), k13 );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k00 ), zero );
      k01 = _mm256_xor_si256( mm256_ror1x32_128(
-                                     mm256_aesenc_2x128( k01 ) ), k00 );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ) );
+                                     mm256_aesenc_2x128( k01, zero ) ), k00 );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
      k02 = _mm256_xor_si256( mm256_ror1x32_128(
-                                     mm256_aesenc_2x128( k02 ) ), k01 );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ) );
+                                     mm256_aesenc_2x128( k02, zero ) ), k01 );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
      k03 = _mm256_xor_si256( mm256_ror1x32_128(
-                                     mm256_aesenc_2x128( k03 ) ), k02 );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ) );
+                                     mm256_aesenc_2x128( k03, zero ) ), k02 );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
      p1 = _mm256_xor_si256( p1, x );
 
      k10 = _mm256_xor_si256( mm256_ror1x32_128(
-                                     mm256_aesenc_2x128( k10 ) ), k03 );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k10 ) );
+                                     mm256_aesenc_2x128( k10, zero ) ), k03 );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k10 ), zero );
      k11 = _mm256_xor_si256( mm256_ror1x32_128(
-                                     mm256_aesenc_2x128( k11 ) ), k10 );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ) );
+                                     mm256_aesenc_2x128( k11, zero ) ), k10 );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
      k12 = _mm256_xor_si256( mm256_ror1x32_128(
-                                     mm256_aesenc_2x128( k12 ) ), k11 );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ) );
+                                     mm256_aesenc_2x128( k12, zero ) ), k11 );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
      k13 = _mm256_xor_si256( mm256_ror1x32_128(
-                                     mm256_aesenc_2x128( k13 ) ), k12 );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ) );
+                                     mm256_aesenc_2x128( k13, zero ) ), k12 );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );
 
      p3 = _mm256_xor_si256( p3, x );
 
      // round 4, 8, 12
 
      k00 = _mm256_xor_si256( k00, mm256_ror2x256hi_1x32( k12, k13 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( p1, k00 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( p1, k00 ), zero );
      k01 = _mm256_xor_si256( k01, mm256_ror2x256hi_1x32( k13, k00 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
      k02 = _mm256_xor_si256( k02, mm256_ror2x256hi_1x32( k00, k01 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
      k03 = _mm256_xor_si256( k03, mm256_ror2x256hi_1x32( k01, k02 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
      p0 = _mm256_xor_si256( p0, x );
 
      k10 = _mm256_xor_si256( k10, mm256_ror2x256hi_1x32( k02, k03 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( p3, k10 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( p3, k10 ), zero );
      k11 = _mm256_xor_si256( k11, mm256_ror2x256hi_1x32( k03, k10 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
      k12 = _mm256_xor_si256( k12, mm256_ror2x256hi_1x32( k10, k11 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
      k13 = _mm256_xor_si256( k13, mm256_ror2x256hi_1x32( k11, k12 ) );
-     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ) );
+     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );
 
      p2 = _mm256_xor_si256( p2, x );
 
@@ -190,36 +191,36 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
    // round 13
 
    k00 = _mm256_xor_si256( mm256_ror1x32_128(
-			             mm256_aesenc_2x128( k00 ) ), k13  );
-   x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k00 ) );
+			             mm256_aesenc_2x128( k00, zero ) ), k13  );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k00 ), zero );
    k01 = _mm256_xor_si256( mm256_ror1x32_128(
-			             mm256_aesenc_2x128( k01 ) ), k00 );
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ) );
+			             mm256_aesenc_2x128( k01, zero ) ), k00 );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
    k02 = _mm256_xor_si256( mm256_ror1x32_128(
-			             mm256_aesenc_2x128( k02 ) ), k01 );
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ) );
+			             mm256_aesenc_2x128( k02, zero ) ), k01 );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
    k03 = _mm256_xor_si256( mm256_ror1x32_128(
-			             mm256_aesenc_2x128( k03 ) ), k02 );
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ) );
+			             mm256_aesenc_2x128( k03, zero ) ), k02 );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
    p3 = _mm256_xor_si256( p3, x );
 
    k10 = _mm256_xor_si256( mm256_ror1x32_128(
-			             mm256_aesenc_2x128( k10 ) ), k03 );
-   x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k10 ) );
+			             mm256_aesenc_2x128( k10, zero ) ), k03 );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k10 ), zero );
    k11 = _mm256_xor_si256( mm256_ror1x32_128(
-			             mm256_aesenc_2x128( k11 ) ), k10 );
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ) );
+			             mm256_aesenc_2x128( k11, zero ) ), k10 );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
 
-   k12 = mm256_ror1x32_128( mm256_aesenc_2x128( k12 ) );
+   k12 = mm256_ror1x32_128( mm256_aesenc_2x128( k12, zero ) );
    k12 = _mm256_xor_si256( k12, _mm256_xor_si256( k11, _mm256_set_epi32(
 	       ~ctx->count2, ctx->count3, ctx->count0, ctx->count1,
 	       ~ctx->count2, ctx->count3, ctx->count0, ctx->count1 ) ) );
 
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ) );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
    k13 = _mm256_xor_si256( mm256_ror1x32_128(
-			             mm256_aesenc_2x128( k13 ) ), k12 );
-   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ) );
+			             mm256_aesenc_2x128( k13, zero ) ), k12 );
+   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );
 
    p1 = _mm256_xor_si256( p1, x );
 

algo/shavite/sph-shavite-aesni.c

@@ -87,6 +87,7 @@ static const sph_u32 IV512[] = {
 static void
 c512( sph_shavite_big_context *sc, const void *msg )
 {
+   const __m128i zero = _mm_setzero_si128();
    __m128i p0, p1, p2, p3, x;
    __m128i k00, k01, k02, k03, k10, k11, k12, k13;
    __m128i *m = (__m128i*)msg;
@@ -101,38 +102,38 @@ c512( sph_shavite_big_context *sc, const void *msg )
    // round
    k00 = m[0];
    x = _mm_xor_si128( p1, k00 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
    k01 = m[1];
    x = _mm_xor_si128( x, k01 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
    k02 = m[2];
    x = _mm_xor_si128( x, k02 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
    k03 = m[3];
    x = _mm_xor_si128( x, k03 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
 
    p0 = _mm_xor_si128( p0, x );
 
    k10 = m[4];
    x = _mm_xor_si128( p3, k10 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
    k11 = m[5];
    x = _mm_xor_si128( x, k11 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
    k12 = m[6];
    x = _mm_xor_si128( x, k12 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
    k13 = m[7];
    x = _mm_xor_si128( x, k13 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
 
    p2 = _mm_xor_si128( p2, x );
 
    for ( r = 0; r < 3; r ++ )
    {
       // round 1, 5, 9
-      k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, m128_zero ) );
+      k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, zero ) );
       k00 = _mm_xor_si128( k00, k13 ); 
 
       if ( r == 0 )
@@ -140,8 +141,8 @@ c512( sph_shavite_big_context *sc, const void *msg )
                   ~sc->count3, sc->count2, sc->count1, sc->count0 ) ); 
 
       x = _mm_xor_si128( p0, k00 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, zero ) );
       k01 = _mm_xor_si128( k01, k00 );
 
       if ( r == 1 )
@@ -149,32 +150,32 @@ c512( sph_shavite_big_context *sc, const void *msg )
                   ~sc->count0, sc->count1, sc->count2, sc->count3 ) );
 
       x = _mm_xor_si128( x, k01 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, zero ) );
       k02 = _mm_xor_si128( k02, k01 );
       x = _mm_xor_si128( x, k02 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, zero ) );
       k03 = _mm_xor_si128( k03, k02 );
       x = _mm_xor_si128( x, k03 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
 
       p3 = _mm_xor_si128( p3, x );
 
-      k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, m128_zero ) );
+      k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, zero ) );
       k10 = _mm_xor_si128( k10, k03 );
 
       x = _mm_xor_si128( p2, k10 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, zero ) );
       k11 = _mm_xor_si128( k11, k10 );
       x = _mm_xor_si128( x, k11 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, zero ) );
       k12 = _mm_xor_si128( k12, k11 );
       x = _mm_xor_si128( x, k12 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, zero ) );
       k13 = _mm_xor_si128( k13, k12 );
 
       if ( r == 2 )
@@ -182,78 +183,78 @@ c512( sph_shavite_big_context *sc, const void *msg )
                   ~sc->count1, sc->count0, sc->count3, sc->count2 ) );
 
       x = _mm_xor_si128( x, k13 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       p1 = _mm_xor_si128( p1, x );
 
       // round 2, 6, 10
 
       k00 = _mm_xor_si128( k00, mm128_ror256hi_1x32( k12, k13 ) );
       x = _mm_xor_si128( p3, k00 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k01 = _mm_xor_si128( k01, mm128_ror256hi_1x32( k13, k00 ) );
       x = _mm_xor_si128( x, k01 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k02 = _mm_xor_si128( k02, mm128_ror256hi_1x32( k00, k01 ) );
       x = _mm_xor_si128( x, k02 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k03 = _mm_xor_si128( k03, mm128_ror256hi_1x32( k01, k02 ) );
       x = _mm_xor_si128( x, k03 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
 
       p2 = _mm_xor_si128( p2, x );
 
       k10 = _mm_xor_si128( k10, mm128_ror256hi_1x32( k02, k03 ) );
       x = _mm_xor_si128( p1, k10 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k11 = _mm_xor_si128( k11, mm128_ror256hi_1x32( k03, k10 ) );
       x = _mm_xor_si128( x, k11 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k12 = _mm_xor_si128( k12, mm128_ror256hi_1x32( k10, k11 ) );
       x = _mm_xor_si128( x, k12 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k13 = _mm_xor_si128( k13, mm128_ror256hi_1x32( k11, k12 ) );
       x = _mm_xor_si128( x, k13 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
 
       p0 = _mm_xor_si128( p0, x );
 
       // round 3, 7, 11
 
-      k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, m128_zero ) );
+      k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, zero ) );
       k00 = _mm_xor_si128( k00, k13 );
       x = _mm_xor_si128( p2, k00 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, zero ) );
       k01 = _mm_xor_si128( k01, k00 );
       x = _mm_xor_si128( x, k01 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, zero ) );
       k02 = _mm_xor_si128( k02, k01 );
       x = _mm_xor_si128( x, k02 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, zero ) );
       k03 = _mm_xor_si128( k03, k02 );
       x = _mm_xor_si128( x, k03 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
 
       p1 = _mm_xor_si128( p1, x );
 
-      k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, m128_zero ) );
+      k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, zero ) );
       k10 = _mm_xor_si128( k10, k03 );
       x = _mm_xor_si128( p0, k10 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, zero ) );
       k11 = _mm_xor_si128( k11, k10 );
       x = _mm_xor_si128( x, k11 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, zero ) );
       k12 = _mm_xor_si128( k12, k11 );
       x = _mm_xor_si128( x, k12 );
-      x = _mm_aesenc_si128( x, m128_zero );
-      k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, m128_zero ) );
+      x = _mm_aesenc_si128( x, zero );
+      k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, zero ) );
       k13 = _mm_xor_si128( k13, k12 );
       x = _mm_xor_si128( x, k13 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
 
       p3 = _mm_xor_si128( p3, x );
 
@@ -261,73 +262,73 @@ c512( sph_shavite_big_context *sc, const void *msg )
 
       k00 = _mm_xor_si128( k00, mm128_ror256hi_1x32( k12, k13 ) );
       x = _mm_xor_si128( p1, k00 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k01 = _mm_xor_si128( k01, mm128_ror256hi_1x32( k13, k00 ) );
       x = _mm_xor_si128( x, k01 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k02 = _mm_xor_si128( k02, mm128_ror256hi_1x32( k00, k01 ) );
       x = _mm_xor_si128( x, k02 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k03 = _mm_xor_si128( k03, mm128_ror256hi_1x32( k01, k02 ) );
       x = _mm_xor_si128( x, k03 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
 
       p0 = _mm_xor_si128( p0, x );
 
       k10 = _mm_xor_si128( k10, mm128_ror256hi_1x32( k02, k03 ) );
       x = _mm_xor_si128( p3, k10 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k11 = _mm_xor_si128( k11, mm128_ror256hi_1x32( k03, k10 ) );
       x = _mm_xor_si128( x, k11 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k12 = _mm_xor_si128( k12, mm128_ror256hi_1x32( k10, k11 ) );
       x = _mm_xor_si128( x, k12 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
       k13 = _mm_xor_si128( k13, mm128_ror256hi_1x32( k11, k12 ) );
       x = _mm_xor_si128( x, k13 );
-      x = _mm_aesenc_si128( x, m128_zero );
+      x = _mm_aesenc_si128( x, zero );
 
       p2 = _mm_xor_si128( p2, x );
    }
 
    // round 13
 
-   k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, m128_zero ) );
+   k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, zero ) );
    k00 = _mm_xor_si128( k00, k13 );
    x = _mm_xor_si128( p0, k00 );
-   x = _mm_aesenc_si128( x, m128_zero );
-   k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, m128_zero ) ); 
+   x = _mm_aesenc_si128( x, zero );
+   k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, zero ) ); 
    k01 = _mm_xor_si128( k01, k00 );
    x = _mm_xor_si128( x, k01 );
-   x = _mm_aesenc_si128( x, m128_zero );
-   k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, m128_zero ) );
+   x = _mm_aesenc_si128( x, zero );
+   k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, zero ) );
    k02 = _mm_xor_si128( k02, k01 );
    x = _mm_xor_si128( x, k02 );
-   x = _mm_aesenc_si128( x, m128_zero );
-   k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, m128_zero ) );
+   x = _mm_aesenc_si128( x, zero );
+   k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, zero ) );
    k03 = _mm_xor_si128( k03, k02 );
    x = _mm_xor_si128( x, k03 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
 
    p3 = _mm_xor_si128( p3, x );
 
-   k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, m128_zero ) );
+   k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, zero ) );
    k10 = _mm_xor_si128( k10, k03 );
    x = _mm_xor_si128( p2, k10 );
-   x = _mm_aesenc_si128( x, m128_zero );
-   k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, m128_zero ) );
+   x = _mm_aesenc_si128( x, zero );
+   k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, zero ) );
    k11 = _mm_xor_si128( k11, k10 );
    x = _mm_xor_si128( x, k11 );
-   x = _mm_aesenc_si128( x, m128_zero );
-   k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, m128_zero ) );
+   x = _mm_aesenc_si128( x, zero );
+   k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, zero ) );
    k12 = _mm_xor_si128( k12, _mm_xor_si128( k11, _mm_set_epi32(
                ~sc->count2, sc->count3, sc->count0, sc->count1 ) ) );
    x = _mm_xor_si128( x, k12 );
-   x = _mm_aesenc_si128( x, m128_zero );
-   k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, m128_zero ) );
+   x = _mm_aesenc_si128( x, zero );
+   k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, zero ) );
    k13 = _mm_xor_si128( k13, k12 );
    x = _mm_xor_si128( x, k13 );
-   x = _mm_aesenc_si128( x, m128_zero );
+   x = _mm_aesenc_si128( x, zero );
 
    p1 = _mm_xor_si128( p1, x );
 

algo/simd/simd-hash-2way.c

@@ -342,6 +342,7 @@ void fft128_2way( void *a )
 
 void fft128_2way_msg( uint16_t *a, const uint8_t *x, int final )
 {
+  const __m256i zero = _mm256_setzero_si256();
   static const m256_v16 Tweak      = {{ 0,0,0,0,0,0,0,1, 0,0,0,0,0,0,0,1, }};
   static const m256_v16 FinalTweak = {{ 0,0,0,0,0,1,0,1, 0,0,0,0,0,1,0,1, }};
 
@@ -352,10 +353,10 @@ void fft128_2way_msg( uint16_t *a, const uint8_t *x, int final )
 #define UNPACK( i ) \
 do { \
     __m256i t = X[i]; \
-    A[2*i]   = _mm256_unpacklo_epi8( t, m256_zero ); \
+    A[2*i]   = _mm256_unpacklo_epi8( t, zero ); \
     A[2*i+8] = _mm256_mullo_epi16( A[2*i], FFT128_Twiddle[2*i].v256 ); \
     A[2*i+8] = REDUCE(A[2*i+8]); \
-    A[2*i+1] = _mm256_unpackhi_epi8( t, m256_zero ); \
+    A[2*i+1] = _mm256_unpackhi_epi8( t, zero ); \
     A[2*i+9] = _mm256_mullo_epi16(A[2*i+1], FFT128_Twiddle[2*i+1].v256 ); \
     A[2*i+9] = REDUCE(A[2*i+9]); \
 } while(0)
@@ -365,10 +366,10 @@ do { \
 do { \
     __m256i t = X[i]; \
     __m256i tmp; \
-    A[2*i]   = _mm256_unpacklo_epi8( t, m256_zero ); \
+    A[2*i]   = _mm256_unpacklo_epi8( t, zero ); \
     A[2*i+8] = _mm256_mullo_epi16( A[ 2*i ], FFT128_Twiddle[ 2*i ].v256 ); \
     A[2*i+8] = REDUCE( A[ 2*i+8 ] ); \
-    tmp      = _mm256_unpackhi_epi8( t, m256_zero ); \
+    tmp      = _mm256_unpackhi_epi8( t, zero ); \
     A[2*i+1] = _mm256_add_epi16( tmp, tw ); \
     A[2*i+9] = _mm256_mullo_epi16( _mm256_sub_epi16( tmp, tw ), \
                                    FFT128_Twiddle[ 2*i+1 ].v256 );\
@@ -392,6 +393,7 @@ do { \
 
 void fft256_2way_msg( uint16_t *a, const uint8_t *x, int final )
 {
+  const __m256i zero = _mm256_setzero_si256();
   static const m256_v16 Tweak      = {{ 0,0,0,0,0,0,0,1, 0,0,0,0,0,0,0,1, }};
   static const m256_v16 FinalTweak = {{ 0,0,0,0,0,1,0,1, 0,0,0,0,0,1,0,1, }};
 
@@ -402,11 +404,11 @@ void fft256_2way_msg( uint16_t *a, const uint8_t *x, int final )
 #define UNPACK( i ) \
 do { \
     __m256i t = X[i]; \
-    A[ 2*i      ] = _mm256_unpacklo_epi8( t, m256_zero ); \
+    A[ 2*i      ] = _mm256_unpacklo_epi8( t, zero ); \
     A[ 2*i + 16 ] = _mm256_mullo_epi16( A[ 2*i ], \
                                         FFT256_Twiddle[ 2*i ].v256 ); \
     A[ 2*i + 16 ] = REDUCE( A[ 2*i + 16 ] ); \
-    A[ 2*i +  1 ] = _mm256_unpackhi_epi8( t, m256_zero ); \
+    A[ 2*i +  1 ] = _mm256_unpackhi_epi8( t, zero ); \
     A[ 2*i + 17 ] = _mm256_mullo_epi16( A[ 2*i + 1 ], \
                                         FFT256_Twiddle[ 2*i + 1 ].v256 ); \
     A[ 2*i + 17 ] = REDUCE( A[ 2*i + 17 ] ); \
@@ -417,11 +419,11 @@ do { \
 do { \
     __m256i t = X[i]; \
     __m256i tmp; \
-    A[ 2*i      ] = _mm256_unpacklo_epi8( t, m256_zero ); \
+    A[ 2*i      ] = _mm256_unpacklo_epi8( t, zero ); \
     A[ 2*i + 16 ] = _mm256_mullo_epi16( A[ 2*i ], \
                                         FFT256_Twiddle[ 2*i ].v256 ); \
     A[ 2*i + 16 ] = REDUCE( A[ 2*i + 16 ] ); \
-    tmp           = _mm256_unpackhi_epi8( t, m256_zero ); \
+    tmp           = _mm256_unpackhi_epi8( t, zero ); \
     A[ 2*i +  1 ] = _mm256_add_epi16( tmp, tw ); \
     A[ 2*i + 17 ] = _mm256_mullo_epi16( _mm256_sub_epi16( tmp, tw ), \
                                         FFT256_Twiddle[ 2*i + 1 ].v256 ); \
@@ -446,6 +448,8 @@ do { \
   fft128_2way( a+256 );
 }
 
+#define c1_16( x ) {{ x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x }}
+
 void rounds512_2way( uint32_t *state, const uint8_t *msg, uint16_t *fft )
 {
   register __m256i S0l, S1l, S2l, S3l;
@@ -453,7 +457,8 @@ void rounds512_2way( uint32_t *state, const uint8_t *msg, uint16_t *fft )
   __m256i *S = (__m256i*) state;
   __m256i *M = (__m256i*) msg;
   __m256i *W = (__m256i*) fft;
-  static const m256_v16 code[] = { mm256_const1_16(185), mm256_const1_16(233) };
+  static const m256_v16 code[] = { c1_16(185), c1_16(233) };
+  
 
   S0l = _mm256_xor_si256( S[0], M[0] );
   S0h = _mm256_xor_si256( S[1], M[1] );

algo/skein/skein-4way.c

@@ -12,7 +12,7 @@
 
 void skeinhash_4way( void *state, const void *input )
 {
-     uint64_t vhash64[8*4] __attribute__ ((aligned (64)));
+     uint64_t vhash64[16*4] __attribute__ ((aligned (64)));
      skein512_4way_context ctx_skein;
 #if defined(__SHA__)
      uint32_t hash0[16] __attribute__ ((aligned (64)));
@@ -30,7 +30,7 @@ void skeinhash_4way( void *state, const void *input )
      skein512_4way_close( &ctx_skein, vhash64 );
 
 #if defined(__SHA__)      
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 512 );
 
      SHA256_Init( &ctx_sha256 );
      SHA256_Update( &ctx_sha256, (unsigned char*)hash0, 64 );
@@ -50,7 +50,7 @@ void skeinhash_4way( void *state, const void *input )
 
      intrlv_4x32( state, hash0, hash1, hash2, hash3, 256 );
 #else
-     mm256_rintrlv_4x64_4x32( vhash32, vhash64, 512 );
+     rintrlv_4x64_4x32( vhash32, vhash64, 512 );
 
      sha256_4way_init( &ctx_sha256 );
      sha256_4way( &ctx_sha256, vhash32, 64 );
@@ -62,8 +62,7 @@ int scanhash_skein_4way( struct work *work, uint32_t max_nonce,
                          uint64_t *hashes_done, struct thr_info *mythr )
 {
     uint32_t vdata[20*4] __attribute__ ((aligned (64)));
-    uint32_t hash[8*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
+    uint32_t hash[16*4] __attribute__ ((aligned (64)));
     uint32_t lane_hash[8] __attribute__ ((aligned (32)));
     uint32_t *hash7 = &(hash[7<<2]);
     uint32_t *pdata = work->data;
@@ -74,9 +73,7 @@ int scanhash_skein_4way( struct work *work, uint32_t max_nonce,
     __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
     int thr_id = mythr->id;  // thr_id arg is deprecated
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_4x64( vdata, pdata );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
    do
    {
        *noncev = mm256_intrlv_blend_32( mm256_bswap_32(

algo/skein/skein-hash-4way.c

@@ -279,10 +279,7 @@ do { \
                                               _mm256_xor_si256( k2, k3 ) ), \
                             _mm256_xor_si256( _mm256_xor_si256( k4, k5 ), \
                                               _mm256_xor_si256( k6, k7 ) ) ), \
-                         _mm256_set_epi64x( SPH_C64(0x1BD11BDAA9FC1A22), \
-                                            SPH_C64(0x1BD11BDAA9FC1A22), \
-                                            SPH_C64(0x1BD11BDAA9FC1A22), \
-                                            SPH_C64(0x1BD11BDAA9FC1A22) ) ); \
+                         m256_const1_64( 0x1BD11BDAA9FC1A22) ); \
   t2 = t0 ^ t1; \
 } while (0)
 
@@ -294,13 +291,11 @@ do { \
   w3 = _mm256_add_epi64( w3, SKBI(k,s,3) ); \
   w4 = _mm256_add_epi64( w4, SKBI(k,s,4) ); \
   w5 = _mm256_add_epi64( w5, _mm256_add_epi64( SKBI(k,s,5), \
-                           _mm256_set_epi64x( SKBT(t,s,0), SKBT(t,s,0), \
-                                              SKBT(t,s,0), SKBT(t,s,0) ) ) ); \
+                                         m256_const1_64( SKBT(t,s,0) ) ) ); \
   w6 = _mm256_add_epi64( w6, _mm256_add_epi64( SKBI(k,s,6), \
-                           _mm256_set_epi64x( SKBT(t,s,1), SKBT(t,s,1), \
-                                              SKBT(t,s,1), SKBT(t,s,1) ) ) ); \
+                                         m256_const1_64( SKBT(t,s,1) ) ) ); \
   w7 = _mm256_add_epi64( w7, _mm256_add_epi64( SKBI(k,s,7), \
-                                      _mm256_set_epi64x( s, s, s, s ) ) ); \
+                                         m256_const1_64( s ) ) ); \
 } while (0)
 
 

algo/skein/skein2-4way.c

@@ -8,7 +8,7 @@
 void skein2hash_4way( void *output, const void *input )
 {
    skein512_4way_context ctx;
-   uint64_t hash[8*4] __attribute__ ((aligned (64)));
+   uint64_t hash[16*4] __attribute__ ((aligned (64)));
 
    skein512_4way_init( &ctx );
    skein512_4way( &ctx, input, 80 );
@@ -22,8 +22,7 @@ void skein2hash_4way( void *output, const void *input )
 int scanhash_skein2_4way( struct work *work, uint32_t max_nonce,
                           uint64_t *hashes_done, struct thr_info *mythr )
 {
-    uint32_t hash[8*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
+    uint32_t hash[16*4] __attribute__ ((aligned (64)));
     uint32_t vdata[20*4] __attribute__ ((aligned (64)));
     uint32_t lane_hash[8] __attribute__ ((aligned (64)));
     uint32_t *hash7 = &(hash[25]);
@@ -34,20 +33,10 @@ int scanhash_skein2_4way( struct work *work, uint32_t max_nonce,
     uint32_t n = first_nonce;
     __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
     int thr_id = mythr->id;  // thr_id arg is deprecated
-//    uint32_t *noncep = vdata + 73;   // 9*8 + 1
 
-
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-
-//    mm256_bswap_intrlv80_4x64( vdata, pdata );
+    mm256_bswap32_intrlv80_4x64( vdata, pdata );
     do 
     {
-//       be32enc( noncep,   n   );
-//       be32enc( noncep+2, n+1 );
-//       be32enc( noncep+4, n+2 );
-//       be32enc( noncep+6, n+3 );
-
        *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
                 _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
 
@@ -56,7 +45,7 @@ int scanhash_skein2_4way( struct work *work, uint32_t max_nonce,
        for ( int lane = 0; lane < 4; lane++ )
        if ( hash7[ lane<<1 ] <= Htarg )
        {
-          mm256_extr_lane_4x64( lane_hash, hash, lane, 256 );
+          extr_lane_4x64( lane_hash, hash, lane, 256 );
           if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
           {
              pdata[19] = n + lane;

algo/x11/c11-4way.c

@@ -69,7 +69,7 @@ void c11_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 3 Groestl
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -81,7 +81,7 @@ void c11_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // 4way
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      // 4 JH
      jh512_4way( &ctx.jh, vhash, 64 );
@@ -96,16 +96,16 @@ void c11_4way_hash( void *state, const void *input )
      skein512_4way_close( &ctx.skein, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 7 Luffa
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
-     mm256_intrlv_2x128( vhashB, hash2, hash3, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhashB, hash2, hash3, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashB, vhashB, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhashB, 512 );
 
      // 8 Cubehash
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
@@ -133,13 +133,13 @@ void c11_4way_hash( void *state, const void *input )
      sph_shavite512_close( &ctx.shavite, hash3 );
 
      // 10 Simd
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
-     mm256_intrlv_2x128( vhashB, hash2, hash3, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhashB, hash2, hash3, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhashB, 512 );
 
      // 11 Echo
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -165,7 +165,6 @@ int scanhash_c11_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-     uint32_t edata[20] __attribute__((aligned(64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      uint32_t n = pdata[19];
@@ -178,9 +177,7 @@ int scanhash_c11_4way( struct work *work, uint32_t max_nonce,
      uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      for (int m=0; m < 6; m++) 
        if (Htarg <= htmax[m])

algo/x11/timetravel-4way.c

@@ -87,19 +87,16 @@ void timetravel_4way_hash(void *output, const void *input)
            blake512_4way( &ctx.blake, vhashA, dataLen );
            blake512_4way_close( &ctx.blake, vhashB );
            if ( i == 7 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 1:
            bmw512_4way( &ctx.bmw, vhashA, dataLen );
            bmw512_4way_close( &ctx.bmw, vhashB );
            if ( i == 7 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 2:
-           mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                    vhashA, dataLen<<3 );
+           dintrlv_4x64( hash0, hash1, hash2, hash3, vhashA, dataLen<<3 );
            update_and_final_groestl( &ctx.groestl, (char*)hash0,
                                                    (char*)hash0, dataLen<<3 );
            reinit_groestl( &ctx.groestl );
@@ -112,47 +109,40 @@ void timetravel_4way_hash(void *output, const void *input)
            update_and_final_groestl( &ctx.groestl, (char*)hash3,
                                                    (char*)hash3, dataLen<<3 );
            if ( i != 7 )
-              mm256_intrlv_4x64( vhashB,
-                                     hash0, hash1, hash2, hash3, dataLen<<3 );
+              intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, dataLen<<3 );
         break;
         case 3:
            skein512_4way( &ctx.skein, vhashA, dataLen );
            skein512_4way_close( &ctx.skein, vhashB );
            if ( i == 7 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 4:
            jh512_4way( &ctx.jh, vhashA, dataLen );
            jh512_4way_close( &ctx.jh, vhashB );
            if ( i == 7 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 5:
            keccak512_4way( &ctx.keccak, vhashA, dataLen );
            keccak512_4way_close( &ctx.keccak, vhashB );
            if ( i == 7 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 6:
-           mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                    vhashA, dataLen<<3 );
-           mm256_intrlv_2x128( vhashA, hash0, hash1, dataLen<<3 );
+           dintrlv_4x64( hash0, hash1, hash2, hash3, vhashA, dataLen<<3 );
+           intrlv_2x128( vhashA, hash0, hash1, dataLen<<3 );
            luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, dataLen );
-           mm256_dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
-           mm256_intrlv_2x128( vhashA, hash2, hash3, dataLen<<3 );
+           dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
+           intrlv_2x128( vhashA, hash2, hash3, dataLen<<3 );
            luffa_2way_init( &ctx.luffa, 512 );
            luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, dataLen );
-           mm256_dintrlv_2x128( hash2, hash3, vhashA, dataLen<<3 );
+           dintrlv_2x128( hash2, hash3, vhashA, dataLen<<3 );
            if ( i != 7 )           
-              mm256_intrlv_4x64( vhashB,
-                                     hash0, hash1, hash2, hash3, dataLen<<3 );
+              intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, dataLen<<3 );
         break;
         case 7:
-           mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                    vhashA, dataLen<<3 );
+           dintrlv_4x64( hash0, hash1, hash2, hash3, vhashA, dataLen<<3 );
            cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
                                       (const byte*)hash0, dataLen );
            memcpy( &ctx.cube, &tt8_4way_ctx.cube, sizeof(cubehashParam) );
@@ -165,8 +155,7 @@ void timetravel_4way_hash(void *output, const void *input)
            cubehashUpdateDigest( &ctx.cube, (byte*)hash3,
                                       (const byte*)hash3, dataLen );
            if ( i != 7 )           
-              mm256_intrlv_4x64( vhashB,
-                                     hash0, hash1, hash2, hash3, dataLen<<3 );
+              intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, dataLen<<3 );
         break;
         default:
            applog(LOG_ERR,"SWERR: timetravel invalid permutation");
@@ -215,7 +204,7 @@ int scanhash_timetravel_4way( struct work *work, uint32_t max_nonce,
    }
 
    uint64_t *edata = (uint64_t*)endiandata;
-   mm256_intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+   intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
 
    do
    {

algo/x11/timetravel10-4way.c

@@ -93,19 +93,16 @@ void timetravel10_4way_hash(void *output, const void *input)
            blake512_4way( &ctx.blake, vhashA, dataLen );
            blake512_4way_close( &ctx.blake, vhashB );
            if ( i == 9 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 1:
            bmw512_4way( &ctx.bmw, vhashA, dataLen );
            bmw512_4way_close( &ctx.bmw, vhashB );
            if ( i == 9 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 2:
-           mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                    vhashA, dataLen<<3 );
+           dintrlv_4x64( hash0, hash1, hash2, hash3, vhashA, dataLen<<3 );
            update_and_final_groestl( &ctx.groestl, (char*)hash0,
                                                    (char*)hash0, dataLen<<3 );
            reinit_groestl( &ctx.groestl );
@@ -118,46 +115,40 @@ void timetravel10_4way_hash(void *output, const void *input)
            update_and_final_groestl( &ctx.groestl, (char*)hash3,
                                                    (char*)hash3, dataLen<<3 );
            if ( i != 9 )
-              mm256_intrlv_4x64( vhashB,
-                                     hash0, hash1, hash2, hash3, dataLen<<3 );
+              intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, dataLen<<3 );
         break;
         case 3:
            skein512_4way( &ctx.skein, vhashA, dataLen );
            skein512_4way_close( &ctx.skein, vhashB );
            if ( i == 9 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 4:
            jh512_4way( &ctx.jh, vhashA, dataLen );
            jh512_4way_close( &ctx.jh, vhashB );
            if ( i == 9 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 5:
            keccak512_4way( &ctx.keccak, vhashA, dataLen );
            keccak512_4way_close( &ctx.keccak, vhashB );
            if ( i == 9 )
-              mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                       vhashB, dataLen<<3 );
+              dintrlv_4x64( hash0, hash1, hash2, hash3, vhashB, dataLen<<3 );
         break;
         case 6:
-           mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                    vhashA, dataLen<<3 );
-           mm256_intrlv_2x128( vhashA, hash0, hash1, dataLen<<3 );
+           dintrlv_4x64( hash0, hash1, hash2, hash3, vhashA, dataLen<<3 );
+           intrlv_2x128( vhashA, hash0, hash1, dataLen<<3 );
            luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, dataLen );
-           mm256_dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
-           mm256_intrlv_2x128( vhashA, hash2, hash3, dataLen<<3 );
+           dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
+           intrlv_2x128( vhashA, hash2, hash3, dataLen<<3 );
            luffa_2way_init( &ctx.luffa, 512 );
            luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, dataLen );
-           mm256_dintrlv_2x128( hash2, hash3, vhashA, dataLen<<3 );
+           dintrlv_2x128( hash2, hash3, vhashA, dataLen<<3 );
            if ( i != 9 )           
-              mm256_intrlv_4x64( vhashB,
-                                     hash0, hash1, hash2, hash3, dataLen<<3 );
+              intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, dataLen<<3 );
         break;
         case 7:
-           mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
+           dintrlv_4x64( hash0, hash1, hash2, hash3,
                                     vhashA, dataLen<<3 );
            cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
                                       (const byte*)hash0, dataLen );
@@ -171,12 +162,10 @@ void timetravel10_4way_hash(void *output, const void *input)
            cubehashUpdateDigest( &ctx.cube, (byte*)hash3,
                                       (const byte*)hash3, dataLen );
            if ( i != 9 )           
-              mm256_intrlv_4x64( vhashB,
-                                     hash0, hash1, hash2, hash3, dataLen<<3 );
+              intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, dataLen<<3 );
         break;
         case 8:
-           mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                    vhashA, dataLen<<3 );
+           dintrlv_4x64( hash0, hash1, hash2, hash3, vhashA, dataLen<<3 );
            sph_shavite512( &ctx.shavite, hash0, dataLen );
            sph_shavite512_close( &ctx.shavite, hash0 );
            memcpy( &ctx.shavite, &tt10_4way_ctx.shavite, sizeof ctx.shavite );
@@ -189,22 +178,19 @@ void timetravel10_4way_hash(void *output, const void *input)
            sph_shavite512( &ctx.shavite, hash3, dataLen );
            sph_shavite512_close( &ctx.shavite, hash3 );
            if ( i != 9 )
-              mm256_intrlv_4x64( vhashB,
-                                     hash0, hash1, hash2, hash3, dataLen<<3 );
+              intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, dataLen<<3 );
         break;
         case 9:
-           mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                    vhashA, dataLen<<3 );
-           mm256_intrlv_2x128( vhashA, hash0, hash1, dataLen<<3 );
+           dintrlv_4x64( hash0, hash1, hash2, hash3, vhashA, dataLen<<3 );
+           intrlv_2x128( vhashA, hash0, hash1, dataLen<<3 );
            simd_2way_update_close( &ctx.simd, vhashA, vhashA, dataLen<<3 );
-           mm256_dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
-           mm256_intrlv_2x128( vhashA, hash2, hash3, dataLen<<3 );
+           dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
+           intrlv_2x128( vhashA, hash2, hash3, dataLen<<3 );
            simd_2way_init( &ctx.simd, 512 );
            simd_2way_update_close( &ctx.simd, vhashA, vhashA, dataLen<<3 );
-           mm256_dintrlv_2x128( hash2, hash3, vhashA, dataLen<<3 );
+           dintrlv_2x128( hash2, hash3, vhashA, dataLen<<3 );
            if ( i != 9 )
-              mm256_intrlv_4x64( vhashB,
-                                     hash0, hash1, hash2, hash3, dataLen<<3 );
+              intrlv_4x64( vhashB, hash0, hash1, hash2, hash3, dataLen<<3 );
         break;
         default:
            applog(LOG_ERR,"SWERR: timetravel invalid permutation");
@@ -253,7 +239,7 @@ int scanhash_timetravel10_4way( struct work *work,
    }
 
    uint64_t *edata = (uint64_t*)endiandata;
-   mm256_intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+   intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
 
    do
    {

algo/x11/tribus-4way.c

@@ -37,7 +37,7 @@ void tribus_hash_4way(void *state, const void *input)
      keccak512_4way( &ctx_keccak, vhash, 64 );
      keccak512_4way_close( &ctx_keccak, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // hash echo serially
      init_echo( &ctx_echo, 512 );
@@ -64,7 +64,6 @@ int scanhash_tribus_4way( struct work *work, uint32_t max_nonce,
 {
    uint32_t hash[4*8] __attribute__ ((aligned (64)));
    uint32_t vdata[20*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
    const uint32_t first_nonce = pdata[19];
@@ -87,9 +86,7 @@ int scanhash_tribus_4way( struct work *work, uint32_t max_nonce,
                         0xFFFF0000,
                                  0 };
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_4x64( vdata, pdata );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
    // precalc midstate
    // doing it one way then then interleaving would be faster but too

algo/x11/x11-4way.c

@@ -69,7 +69,7 @@ void x11_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 3 Groestl
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -81,7 +81,7 @@ void x11_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // 4way
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      // 4 Skein
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -95,16 +95,16 @@ void x11_4way_hash( void *state, const void *input )
      keccak512_4way( &ctx.keccak, vhash, 64 );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 7 Luffa parallel 2 way 128 bit
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
-     mm256_intrlv_2x128( vhashB, hash2, hash3, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhashB, hash2, hash3, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashB, vhashB, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhashB, 512 );
 
      // 8 Cubehash
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
@@ -132,13 +132,13 @@ void x11_4way_hash( void *state, const void *input )
      sph_shavite512_close( &ctx.shavite, hash3 );
 
      // 10 Simd
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
-     mm256_intrlv_2x128( vhashB, hash2, hash3, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhashB, hash2, hash3, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhashB, 512 );
 
      // 11 Echo
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -164,7 +164,6 @@ int scanhash_x11_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-     uint32_t edata[20] __attribute__((aligned(64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      uint32_t n = pdata[19];
@@ -177,9 +176,7 @@ int scanhash_x11_4way( struct work *work, uint32_t max_nonce,
      uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      for (int m=0; m < 6; m++) 
        if (Htarg <= htmax[m])

algo/x11/x11evo-4way.c

@@ -87,19 +87,16 @@ void x11evo_4way_hash( void *state, const void *input )
          case 0:
             blake512_4way( &ctx.blake, input, 80 );
             blake512_4way_close( &ctx.blake, vhash );
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                        vhash, 64<<3 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
          break;
          case 1:
             bmw512_4way( &ctx.bmw, vhash, 64 );
             bmw512_4way_close( &ctx.bmw, vhash );
             if ( i >= len-1 )
-               mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                        vhash, 64<<3 );
+               dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
          break;
          case 2:
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                     vhash, 64<<3 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
             update_and_final_groestl( &ctx.groestl, (char*)hash0,
                                                     (char*)hash0, 512 );
             reinit_groestl( &ctx.groestl );
@@ -112,47 +109,40 @@ void x11evo_4way_hash( void *state, const void *input )
             update_and_final_groestl( &ctx.groestl, (char*)hash3,
                                                       (char*)hash3, 512 );
             if ( i < len-1 )
-               mm256_intrlv_4x64( vhash,
-                                      hash0, hash1, hash2, hash3, 64<<3 );
+               intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 64<<3 );
          break;
          case 3:
             skein512_4way( &ctx.skein, vhash, 64 );
             skein512_4way_close( &ctx.skein, vhash );
             if ( i >= len-1 )
-               mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                        vhash, 64<<3 );
+               dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
          break;
          case 4:
             jh512_4way( &ctx.jh, vhash, 64 );
             jh512_4way_close( &ctx.jh, vhash );
             if ( i >= len-1 )
-               mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                        vhash, 64<<3 );
+               dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
          break;
          case 5:
             keccak512_4way( &ctx.keccak, vhash, 64 );
             keccak512_4way_close( &ctx.keccak, vhash );
             if ( i >= len-1 )
-               mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                        vhash, 64<<3 );
+               dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
          break;
          case 6:
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                     vhash, 64<<3 );
-            mm256_intrlv_2x128( vhash, hash0, hash1, 64<<3 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
+            intrlv_2x128( vhash, hash0, hash1, 64<<3 );
             luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-            mm256_dintrlv_2x128( hash0, hash1, vhash, 64<<3 );
-            mm256_intrlv_2x128( vhash, hash2, hash3, 64<<3 );
+            dintrlv_2x128( hash0, hash1, vhash, 64<<3 );
+            intrlv_2x128( vhash, hash2, hash3, 64<<3 );
             luffa_2way_init( &ctx.luffa, 512 );
             luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-            mm256_dintrlv_2x128( hash2, hash3, vhash, 64<<3 );
+            dintrlv_2x128( hash2, hash3, vhash, 64<<3 );
             if ( i < len-1 )
-               mm256_intrlv_4x64( vhash,
-                                      hash0, hash1, hash2, hash3, 64<<3 );
+               intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 64<<3 );
          break;
          case 7:
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                     vhash, 64<<3 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
             cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
                                       (const byte*) hash0, 64 );
             memcpy( &ctx.cube, &x11evo_4way_ctx.cube, sizeof(cubehashParam) );
@@ -165,12 +155,10 @@ void x11evo_4way_hash( void *state, const void *input )
             cubehashUpdateDigest( &ctx.cube, (byte*)hash3,
                                       (const byte*) hash3, 64 );
             if ( i < len-1 )
-               mm256_intrlv_4x64( vhash,
-                                      hash0, hash1, hash2, hash3, 64<<3 );
+               intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 64<<3 );
          break;
          case 8:
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                     vhash, 64<<3 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
             sph_shavite512( &ctx.shavite, hash0, 64 );
             sph_shavite512_close( &ctx.shavite, hash0 );
             memcpy( &ctx.shavite, &x11evo_4way_ctx.shavite,
@@ -186,26 +174,22 @@ void x11evo_4way_hash( void *state, const void *input )
             sph_shavite512( &ctx.shavite, hash3, 64 );
             sph_shavite512_close( &ctx.shavite, hash3 );
             if ( i < len-1 )
-               mm256_intrlv_4x64( vhash,
-                                      hash0, hash1, hash2, hash3, 64<<3 );
+               intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 64<<3 );
          break;
          case 9:
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                     vhash, 64<<3 );
-            mm256_intrlv_2x128( vhash, hash0, hash1, 64<<3 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
+            intrlv_2x128( vhash, hash0, hash1, 64<<3 );
             simd_2way_update_close( &ctx.simd, vhash, vhash, 64<<3 );
-            mm256_dintrlv_2x128( hash0, hash1, vhash, 64<<3 );
-            mm256_intrlv_2x128( vhash, hash2, hash3, 64<<3 );
+            dintrlv_2x128( hash0, hash1, vhash, 64<<3 );
+            intrlv_2x128( vhash, hash2, hash3, 64<<3 );
             simd_2way_init( &ctx.simd, 512 );
             simd_2way_update_close( &ctx.simd, vhash, vhash, 64<<3 );
-            mm256_dintrlv_2x128( hash2, hash3, vhash, 64<<3 );
+            dintrlv_2x128( hash2, hash3, vhash, 64<<3 );
             if ( i < len-1 )
-               mm256_intrlv_4x64( vhash,
-                                      hash0, hash1, hash2, hash3, 64<<3 );
+               intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 64<<3 );
          break;
          case 10:
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
-                                     vhash, 64<<3 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 64<<3 );
             update_final_echo( &ctx.echo, (BitSequence *)hash0,
                                    (const BitSequence *) hash0, 512 );
             memcpy( &ctx.echo, &x11evo_4way_ctx.echo, sizeof(hashState_echo) );
@@ -218,8 +202,7 @@ void x11evo_4way_hash( void *state, const void *input )
             update_final_echo( &ctx.echo, (BitSequence *)hash3,
                                    (const BitSequence *) hash3, 512 );
             if ( i < len-1 )
-               mm256_intrlv_4x64( vhash,
-                                      hash0, hash1, hash2, hash3, 64<<3 );
+               intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 64<<3 );
          break;
       }
    }
@@ -269,7 +252,7 @@ int scanhash_x11evo_4way( struct work* work, uint32_t max_nonce,
       }
 
      uint64_t *edata = (uint64_t*)endiandata;
-     mm256_intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+     intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
 
      do
      {

algo/x11/x11gost-4way.c

@@ -70,7 +70,7 @@ void x11gost_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
      memcpy( &ctx.groestl, &x11gost_4way_ctx.groestl,
@@ -84,7 +84,7 @@ void x11gost_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // 4way
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      skein512_4way( &ctx.skein, vhash, 64 );
      skein512_4way_close( &ctx.skein, vhash );
@@ -96,7 +96,7 @@ void x11gost_4way_hash( void *state, const void *input )
      keccak512_4way_close( &ctx.keccak, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_gost512( &ctx.gost, hash0, 64 );
      sph_gost512_close( &ctx.gost, hash0 );
@@ -110,13 +110,13 @@ void x11gost_4way_hash( void *state, const void *input )
      sph_gost512( &ctx.gost, hash3, 64 );
      sph_gost512_close( &ctx.gost, hash3 );
 
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
      memcpy( &ctx.cube, &x11gost_4way_ctx.cube, sizeof(cubehashParam) );
@@ -141,12 +141,12 @@ void x11gost_4way_hash( void *state, const void *input )
      sph_shavite512( &ctx.shavite, hash3, 64 );
      sph_shavite512_close( &ctx.shavite, hash3 );
 
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
                        (const BitSequence *) hash0, 512 );
@@ -171,7 +171,6 @@ int scanhash_x11gost_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      uint32_t n = pdata[19];
@@ -184,9 +183,7 @@ int scanhash_x11gost_4way( struct work *work, uint32_t max_nonce,
      uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      for (int m=0; m < 6; m++) 
        if (Htarg <= htmax[m])

algo/x12/x12-4way.c

@@ -33,7 +33,6 @@ typedef struct {
     simd_2way_context       simd;
     hashState_echo          echo;
     hamsi512_4way_context   hamsi;
-//    sph_fugue512_context    fugue;
 } x12_4way_ctx_holder;
 
 x12_4way_ctx_holder x12_4way_ctx __attribute__ ((aligned (64)));
@@ -52,7 +51,6 @@ void init_x12_4way_ctx()
      simd_2way_init( &x12_4way_ctx.simd, 512 );
      init_echo( &x12_4way_ctx.echo, 512 );
      hamsi512_4way_init( &x12_4way_ctx.hamsi );
-//     sph_fugue512_init( &x12_4way_ctx.fugue );
 };
 
 void x12_4way_hash( void *state, const void *input )
@@ -74,7 +72,7 @@ void x12_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 3 Groestl
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -86,7 +84,7 @@ void x12_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // Parallel 4way 64 bit
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      // 4 Skein
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -101,16 +99,16 @@ void x12_4way_hash( void *state, const void *input )
      keccak512_4way_close( &ctx.keccak, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 7 Luffa
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     intrlv_2x128( hash2, hash3, vhash, 512 );
 
      // 8 Cubehash
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
@@ -138,13 +136,13 @@ void x12_4way_hash( void *state, const void *input )
      sph_shavite512_close( &ctx.shavite, hash3 );
 
      // 10 Simd
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // 11 Echo
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -160,11 +158,11 @@ void x12_4way_hash( void *state, const void *input )
                        (const BitSequence *) hash3, 512 );
 
      // 12 Hamsi parallel 4way 32 bit
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( state, state+32, state+64, state+96, vhash, 256 );
+     dintrlv_4x64( state, state+32, state+64, state+96, vhash, 256 );
 }
 
 int scanhash_x12_4way( struct work *work, uint32_t max_nonce,
@@ -189,7 +187,7 @@ int scanhash_x12_4way( struct work *work, uint32_t max_nonce,
      swab32_array( endiandata, pdata, 20 );
 
      uint64_t *edata = (uint64_t*)endiandata;
-     mm256_intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+     intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
 
      for ( int m=0; m < 6; m++ )
        if ( Htarg <= htmax[m] )

algo/x13/phi1612-4way.c

@@ -53,7 +53,7 @@ void phi1612_4way_hash( void *state, const void *input )
      jh512_4way_close( &ctx.jh, vhash );
 
      // Serial to the end
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // Cubehash
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
@@ -114,7 +114,6 @@ int scanhash_phi1612_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-     uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      const uint32_t first_nonce = pdata[19];
@@ -125,9 +124,7 @@ int scanhash_phi1612_4way( struct work *work, uint32_t max_nonce,
 
      if ( opt_benchmark )
           ( (uint32_t*)ptarget )[7] = 0x0cff;
-     swab32_array( edata, pdata, 20 );
-     mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      do {
            *noncev = mm256_intrlv_blend_32( mm256_bswap_32(

algo/x13/skunk-4way.c

@@ -33,7 +33,7 @@ void skunk_4way_hash( void *output, const void *input )
 
      skein512_4way( &ctx.skein, input, 80 );
      skein512_4way_close( &ctx.skein, vhash );
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      cubehashUpdateDigest( &ctx.cube, (byte*) hash0, (const byte*)hash0, 64 );
      memcpy( &ctx.cube, &skunk_4way_ctx.cube, sizeof(cubehashParam) );
@@ -78,7 +78,6 @@ int scanhash_skunk_4way( struct work *work, uint32_t max_nonce,
 {
    uint32_t hash[4*8] __attribute__ ((aligned (64)));
    uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-   uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
    const uint32_t first_nonce = pdata[19];
@@ -91,9 +90,7 @@ int scanhash_skunk_4way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ((uint32_t*)ptarget)[7] = 0x0cff;
 
-   swab32_array( edata, pdata, 20 );
-   mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_4x64( vdata, pdata );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
    do
    {
       *noncev = mm256_intrlv_blend_32( mm256_bswap_32(

algo/x13/x13-4way.c

@@ -74,7 +74,7 @@ void x13_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 3 Groestl
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -86,7 +86,7 @@ void x13_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // Parallel 4way 64 bit
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      // 4 Skein
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -101,16 +101,16 @@ void x13_4way_hash( void *state, const void *input )
      keccak512_4way_close( &ctx.keccak, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 7 Luffa
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // 8 Cubehash
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
@@ -138,13 +138,13 @@ void x13_4way_hash( void *state, const void *input )
      sph_shavite512_close( &ctx.shavite, hash3 );
 
      // 10 Simd
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // 11 Echo
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -160,10 +160,10 @@ void x13_4way_hash( void *state, const void *input )
                        (const BitSequence *) hash3, 512 );
 
      // 12 Hamsi parallel 4way 32 bit
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 13 Fugue serial
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -189,7 +189,6 @@ int scanhash_x13_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-     uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      uint32_t n = pdata[19];
@@ -202,9 +201,7 @@ int scanhash_x13_4way( struct work *work, uint32_t max_nonce,
      uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
-     swab32_array( edata, pdata, 20 );
-     mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      for ( int m=0; m < 6; m++ )
        if ( Htarg <= htmax[m] )

algo/x13/x13sm3-4way.c

@@ -81,7 +81,7 @@ void x13sm3_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // Groestl
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -93,7 +93,7 @@ void x13sm3_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // Parallel 4way
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      // Skein
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -108,16 +108,16 @@ void x13sm3_4way_hash( void *state, const void *input )
      keccak512_4way_close( &ctx.keccak, vhash );
 
      // Serial to the end
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // Luffa
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // Cubehash
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
@@ -145,13 +145,13 @@ void x13sm3_4way_hash( void *state, const void *input )
      sph_shavite512_close( &ctx.shavite, hash3 );
 
      // Simd
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // Echo
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -185,10 +185,10 @@ void x13sm3_4way_hash( void *state, const void *input )
      dintrlv_4x32( hash0, hash1, hash2, hash3, sm3_vhash, 512 );
 
      // Hamsi parallel 4x32x2
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // Fugue serial
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -214,7 +214,6 @@ int scanhash_x13sm3_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-     uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      uint32_t n = pdata[19];
@@ -227,9 +226,7 @@ int scanhash_x13sm3_4way( struct work *work, uint32_t max_nonce,
      uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
-     swab32_array( edata, pdata, 20 );
-     mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      blake512_4way_init( &x13sm3_ctx_mid );
      blake512_4way( &x13sm3_ctx_mid, vdata, 64 );

algo/x14/polytimos-4way.c

@@ -39,7 +39,7 @@ void polytimos_4way_hash( void *output, const void *input )
 
      // Need to convert from 64 bit interleaved to 32 bit interleaved.
      uint32_t vhash32[16*4];
-     mm256_rintrlv_4x64_4x32( vhash32, vhash, 512 );
+     rintrlv_4x64_4x32( vhash32, vhash, 512 );
      shabal512_4way_init( &ctx.shabal );
      shabal512_4way( &ctx.shabal, vhash32, 64 );
      shabal512_4way_close( &ctx.shabal, vhash32 );
@@ -58,15 +58,15 @@ void polytimos_4way_hash( void *output, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, 512 );
 
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      sph_fugue512_init( &ctx.fugue );
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -105,7 +105,6 @@ int scanhash_polytimos_4way( struct work *work, uint32_t max_nonce,
 {
    uint32_t hash[4*8] __attribute__ ((aligned (64)));
    uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
    const uint32_t first_nonce = pdata[19];
@@ -118,9 +117,7 @@ int scanhash_polytimos_4way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ptarget[7] = 0x0cff;
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_4x64( vdata, pdata );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
    do {
       *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
                  _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );

algo/x14/veltor-4way.c

@@ -40,7 +40,7 @@ void veltor_4way_hash( void *output, const void *input )
 
      skein512_4way( &ctx.skein, input, 80 );
      skein512_4way_close( &ctx.skein, vhash );
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_shavite512( &ctx.shavite, hash0, 64 );
      sph_shavite512_close( &ctx.shavite, hash0 );
@@ -82,7 +82,6 @@ int scanhash_veltor_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      const uint32_t Htarg = ptarget[7];
@@ -95,9 +94,7 @@ int scanhash_veltor_4way( struct work *work, uint32_t max_nonce,
      if ( opt_benchmark )
         ptarget[7] = 0x0cff;
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      do
      {

algo/x14/x14-4way.c

@@ -78,7 +78,7 @@ void x14_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 3 Groestl
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -90,7 +90,7 @@ void x14_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // Parallel 4way
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      // 4 Skein
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -105,16 +105,16 @@ void x14_4way_hash( void *state, const void *input )
      keccak512_4way_close( &ctx.keccak, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 7 Luffa
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // 8 Cubehash
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
@@ -142,13 +142,13 @@ void x14_4way_hash( void *state, const void *input )
      sph_shavite512_close( &ctx.shavite, hash3 );
 
      // 10 Simd
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // 11 Echo
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -164,10 +164,10 @@ void x14_4way_hash( void *state, const void *input )
                        (const BitSequence *) hash3, 512 );
 
      // 12 Hamsi parallel 4way 32 bit
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 13 Fugue serial
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -193,7 +193,6 @@ int scanhash_x14_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*16] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      uint32_t n = pdata[19];
@@ -206,9 +205,7 @@ int scanhash_x14_4way( struct work *work, uint32_t max_nonce,
      uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      for ( int m=0; m < 6; m++ )
        if ( Htarg <= htmax[m] )

algo/x15/x15-4way.c

@@ -81,7 +81,7 @@ void x15_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 3 Groestl
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -93,7 +93,7 @@ void x15_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // Parallel 4way
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      // 4 Skein
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -108,16 +108,16 @@ void x15_4way_hash( void *state, const void *input )
      keccak512_4way_close( &ctx.keccak, vhash );
 
      // Serial to the end
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 7 Luffa
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // 8 Cubehash
      cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
@@ -145,13 +145,13 @@ void x15_4way_hash( void *state, const void *input )
      sph_shavite512_close( &ctx.shavite, hash3 );
 
      // 10 Simd
-     mm256_intrlv_2x128( vhash, hash0, hash1, 512 );
+     intrlv_2x128( vhash, hash0, hash1, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-     mm256_intrlv_2x128( vhash, hash2, hash3, 512 );
+     dintrlv_2x128( hash0, hash1, vhash, 512 );
+     intrlv_2x128( vhash, hash2, hash3, 512 );
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhash, vhash, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );
 
      // 11 Echo
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -167,10 +167,10 @@ void x15_4way_hash( void *state, const void *input )
                        (const BitSequence *) hash3, 512 );
 
      // 12 Hamsi parallel 4way 32 bit
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 13 Fugue
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -217,7 +217,6 @@ int scanhash_x15_4way( struct work *work, uint32_t max_nonce,
                        uint64_t *hashes_done, struct thr_info *mythr )
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
-     uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
@@ -231,10 +230,8 @@ int scanhash_x15_4way( struct work *work, uint32_t max_nonce,
      uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
-     swab32_array( edata, pdata, 20 );
-     mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
 
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
      for ( int m=0; m < 6; m++ )
        if ( Htarg <= htmax[m] )

algo/x16/x16r-4way.c

@@ -67,7 +67,7 @@ void x16r_4way_hash( void* output, const void* input )
    void *in3 = (void*) hash3;
    int size = 80;
 
-   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, input, 640 );
+   dintrlv_4x64( hash0, hash1, hash2, hash3, input, 640 );
  
    if ( s_ntime == UINT32_MAX )
    {
@@ -96,11 +96,11 @@ void x16r_4way_hash( void* output, const void* input )
                blake512_4way( &ctx.blake, input, size );
             else
             {
-               mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
+               intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
                blake512_4way( &ctx.blake, vhash, size );
             }
             blake512_4way_close( &ctx.blake, vhash );
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
          break;
          case BMW:
             bmw512_4way_init( &ctx.bmw );
@@ -108,11 +108,11 @@ void x16r_4way_hash( void* output, const void* input )
                bmw512_4way( &ctx.bmw, input, size );
             else
             {
-               mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
+               intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
                bmw512_4way( &ctx.bmw, vhash, size );
             }
             bmw512_4way_close( &ctx.bmw, vhash );
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
          break;
          case GROESTL:
                init_groestl( &ctx.groestl, 64 );
@@ -134,11 +134,11 @@ void x16r_4way_hash( void* output, const void* input )
                skein512_4way( &ctx.skein, input, size );
             else
             {
-               mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
+               intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
                skein512_4way( &ctx.skein, vhash, size );
             }
             skein512_4way_close( &ctx.skein, vhash );
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
          break;
          case JH:
             jh512_4way_init( &ctx.jh );
@@ -146,11 +146,11 @@ void x16r_4way_hash( void* output, const void* input )
                jh512_4way( &ctx.jh, input, size );
             else
             {
-               mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
+               intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
                jh512_4way( &ctx.jh, vhash, size );
             }
             jh512_4way_close( &ctx.jh, vhash );
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
          break;
          case KECCAK:
             keccak512_4way_init( &ctx.keccak );
@@ -158,21 +158,21 @@ void x16r_4way_hash( void* output, const void* input )
                keccak512_4way( &ctx.keccak, input, size );
             else
             {
-               mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
+               intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
                keccak512_4way( &ctx.keccak, vhash, size );
             }
             keccak512_4way_close( &ctx.keccak, vhash );
-            mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+            dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
          break;
          case LUFFA:
-            mm256_intrlv_2x128( vhash, in0, in1, size<<3 );
+            intrlv_2x128( vhash, in0, in1, size<<3 );
             luffa_2way_init( &ctx.luffa, 512 );
             luffa_2way_update_close( &ctx.luffa, vhash, vhash, size );
-            mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-            mm256_intrlv_2x128( vhash, in2, in3, size<<3 );
+            dintrlv_2x128( hash0, hash1, vhash, 512 );
+            intrlv_2x128( vhash, in2, in3, size<<3 );
             luffa_2way_init( &ctx.luffa, 512 );
             luffa_2way_update_close( &ctx.luffa, vhash, vhash, size);
-            mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+            dintrlv_2x128( hash2, hash3, vhash, 512 );
          break;
          case CUBEHASH:
             cubehashInit( &ctx.cube, 512, 16, 32 );
@@ -203,14 +203,14 @@ void x16r_4way_hash( void* output, const void* input )
             sph_shavite512_close( &ctx.shavite, hash3 );
          break;
          case SIMD:
-            mm256_intrlv_2x128( vhash, in0, in1, size<<3 );
+            intrlv_2x128( vhash, in0, in1, size<<3 );
             simd_2way_init( &ctx.simd, 512 );
             simd_2way_update_close( &ctx.simd, vhash, vhash, size<<3 );
-            mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
-            mm256_intrlv_2x128( vhash, in2, in3, size<<3 );
+            dintrlv_2x128( hash0, hash1, vhash, 512 );
+            intrlv_2x128( vhash, in2, in3, size<<3 );
             simd_2way_init( &ctx.simd, 512 );
             simd_2way_update_close( &ctx.simd, vhash, vhash, size<<3 );
-            mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
+            dintrlv_2x128( hash2, hash3, vhash, 512 );
          break;
          case ECHO:
              init_echo( &ctx.echo, 512 );
@@ -227,11 +227,11 @@ void x16r_4way_hash( void* output, const void* input )
                                 (const BitSequence*)in3, size<<3 );
          break;
          case HAMSI:
-             mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
+             intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
              hamsi512_4way_init( &ctx.hamsi );
              hamsi512_4way( &ctx.hamsi, vhash, size );
              hamsi512_4way_close( &ctx.hamsi, vhash );
-             mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+             dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
          break;
          case FUGUE:
              sph_fugue512_init( &ctx.fugue );
@@ -269,11 +269,11 @@ void x16r_4way_hash( void* output, const void* input )
              sph_whirlpool_close( &ctx.whirlpool, hash3 );
          break;
          case SHA_512:
-             mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
+             intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
              sha512_4way_init( &ctx.sha512 );
              sha512_4way( &ctx.sha512, vhash, size );
              sha512_4way_close( &ctx.sha512, vhash );
-             mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+             dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
          break;
       }
       size = 64;
@@ -316,7 +316,7 @@ int scanhash_x16r_4way( struct work *work, uint32_t max_nonce,
       ptarget[7] = 0x0cff;
 
    uint64_t *edata = (uint64_t*)endiandata;
-   mm256_intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+   intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
 
    do
    {

algo/x17/sonoa-4way.c

@@ -69,7 +69,7 @@ void sonoa_4way_hash( void *state, const void *input )
      bmw512_4way( &ctx.bmw, vhash, 64 );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -80,7 +80,7 @@ void sonoa_4way_hash( void *state, const void *input )
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -94,7 +94,7 @@ void sonoa_4way_hash( void *state, const void *input )
      keccak512_4way( &ctx.keccak, vhash, 64 );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
@@ -116,8 +116,8 @@ void sonoa_4way_hash( void *state, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128_512( hash0, hash1, vhashA );
+     dintrlv_2x128_512( hash2, hash3, vhashB );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -134,13 +134,13 @@ void sonoa_4way_hash( void *state, const void *input )
 
 // 2
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      bmw512_4way_init( &ctx.bmw );
      bmw512_4way( &ctx.bmw, vhash, 64 );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -151,7 +151,7 @@ void sonoa_4way_hash( void *state, const void *input )
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -165,7 +165,7 @@ void sonoa_4way_hash( void *state, const void *input )
      keccak512_4way( &ctx.keccak, vhash, 64 );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
@@ -187,8 +187,8 @@ void sonoa_4way_hash( void *state, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128_512( hash0, hash1, vhashA );
+     dintrlv_2x128_512( hash2, hash3, vhashB );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -203,7 +203,7 @@ void sonoa_4way_hash( void *state, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
@@ -215,7 +215,7 @@ void sonoa_4way_hash( void *state, const void *input )
      bmw512_4way( &ctx.bmw, vhash, 64 );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -226,7 +226,7 @@ void sonoa_4way_hash( void *state, const void *input )
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -240,7 +240,7 @@ void sonoa_4way_hash( void *state, const void *input )
      keccak512_4way( &ctx.keccak, vhash, 64 );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
@@ -262,8 +262,8 @@ void sonoa_4way_hash( void *state, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128_512( hash0, hash1, vhashA );
+     dintrlv_2x128_512( hash2, hash3, vhashB );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -278,13 +278,13 @@ void sonoa_4way_hash( void *state, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_fugue512_init( &ctx.fugue );
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -300,13 +300,13 @@ void sonoa_4way_hash( void *state, const void *input )
      sph_fugue512_close( &ctx.fugue, hash3 );
 
 // 4
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      bmw512_4way_init( &ctx.bmw );
      bmw512_4way( &ctx.bmw, vhash, 64 );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -317,7 +317,7 @@ void sonoa_4way_hash( void *state, const void *input )
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -331,7 +331,7 @@ void sonoa_4way_hash( void *state, const void *input )
      keccak512_4way( &ctx.keccak, vhash, 64 );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
@@ -353,8 +353,8 @@ void sonoa_4way_hash( void *state, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128_512( hash0, hash1, vhashA );
+     dintrlv_2x128_512( hash2, hash3, vhashB );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -369,13 +369,13 @@ void sonoa_4way_hash( void *state, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_fugue512_init( &ctx.fugue );
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -396,13 +396,13 @@ void sonoa_4way_hash( void *state, const void *input )
      shabal512_4way( &ctx.shabal, vhash, 64 );
      shabal512_4way_close( &ctx.shabal, vhash );
 
-     mm256_rintrlv_4x32_4x64( vhashB, vhash, 512 ); 
+     rintrlv_4x32_4x64( vhashB, vhash, 512 ); 
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhashB, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -417,8 +417,8 @@ void sonoa_4way_hash( void *state, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, 512 );
 
-     mm256_intrlv_2x128( vhashA, hash0, hash1, 512 );
-     mm256_intrlv_2x128( vhashB, hash2, hash3, 512 );
+     intrlv_2x128_512( vhashA, hash0, hash1 );
+     intrlv_2x128_512( vhashB, hash2, hash3 );
 
      shavite512_2way_init( &ctx.shavite );
      shavite512_2way_update_close( &ctx.shavite, vhashA, vhashA, 64 );
@@ -426,13 +426,13 @@ void sonoa_4way_hash( void *state, const void *input )
      shavite512_2way_update_close( &ctx.shavite, vhashB, vhashB, 64 );
 
 // 5
-     mm256_rintrlv_2x128_4x64( vhash, vhashA, vhashB, 512 );
+     rintrlv_2x128_4x64( vhash, vhashA, vhashB, 512 );
 
      bmw512_4way_init( &ctx.bmw );
      bmw512_4way( &ctx.bmw, vhash, 64 );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_rintrlv_4x64_4x32( vhashB, vhash,  512 );
+     rintrlv_4x64_4x32( vhashB, vhash,  512 );
 
      shabal512_4way_init( &ctx.shabal );
      shabal512_4way( &ctx.shabal, vhashB, 64 );
@@ -449,7 +449,7 @@ void sonoa_4way_hash( void *state, const void *input )
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -463,7 +463,7 @@ void sonoa_4way_hash( void *state, const void *input )
      keccak512_4way( &ctx.keccak, vhash, 64 );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
@@ -485,8 +485,8 @@ void sonoa_4way_hash( void *state, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128_512( hash0, hash1, vhashA );
+     dintrlv_2x128_512( hash2, hash3, vhashB );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -501,13 +501,13 @@ void sonoa_4way_hash( void *state, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_fugue512_init( &ctx.fugue );
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -545,13 +545,13 @@ void sonoa_4way_hash( void *state, const void *input )
 
 // 6
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
      
      bmw512_4way_init( &ctx.bmw );
      bmw512_4way( &ctx.bmw, vhash, 64 );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -562,7 +562,7 @@ void sonoa_4way_hash( void *state, const void *input )
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -576,7 +576,7 @@ void sonoa_4way_hash( void *state, const void *input )
      keccak512_4way( &ctx.keccak, vhash, 64 );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
@@ -598,8 +598,8 @@ void sonoa_4way_hash( void *state, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128_512( hash0, hash1, vhashA );
+     dintrlv_2x128_512( hash2, hash3, vhashB );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -614,13 +614,13 @@ void sonoa_4way_hash( void *state, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_fugue512_init( &ctx.fugue );
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -656,13 +656,13 @@ void sonoa_4way_hash( void *state, const void *input )
      sph_whirlpool( &ctx.whirlpool, hash3, 64 );
      sph_whirlpool_close( &ctx.whirlpool, hash3 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      sha512_4way_init( &ctx.sha512 );
      sha512_4way( &ctx.sha512, vhash, 64 );
      sha512_4way_close( &ctx.sha512, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_whirlpool_init( &ctx.whirlpool );
      sph_whirlpool( &ctx.whirlpool, hash0, 64 );
@@ -679,13 +679,13 @@ void sonoa_4way_hash( void *state, const void *input )
 
 // 7
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      bmw512_4way_init( &ctx.bmw );
      bmw512_4way( &ctx.bmw, vhash, 64 );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
@@ -696,7 +696,7 @@ void sonoa_4way_hash( void *state, const void *input )
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, 64 );
@@ -710,7 +710,7 @@ void sonoa_4way_hash( void *state, const void *input )
      keccak512_4way( &ctx.keccak, vhash, 64 );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
@@ -732,8 +732,8 @@ void sonoa_4way_hash( void *state, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, 512 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, 512 );
+     dintrlv_2x128_512( hash0, hash1, vhashA );
+     dintrlv_2x128_512( hash2, hash3, vhashB );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -748,13 +748,13 @@ void sonoa_4way_hash( void *state, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, 512 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      sph_fugue512_init( &ctx.fugue );
      sph_fugue512( &ctx.fugue, hash0, 64 );
@@ -790,13 +790,13 @@ void sonoa_4way_hash( void *state, const void *input )
      sph_whirlpool( &ctx.whirlpool, hash3, 64 );
      sph_whirlpool_close( &ctx.whirlpool, hash3 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      sha512_4way_init( &ctx.sha512 );
      sha512_4way( &ctx.sha512, vhash, 64 );
      sha512_4way_close( &ctx.sha512, vhash );
 
-     mm256_rintrlv_4x64_4x32( vhashB, vhash,  512 );
+     rintrlv_4x64_4x32( vhashB, vhash,  512 );
 
      haval256_5_4way_init( &ctx.haval );
      haval256_5_4way( &ctx.haval, vhashB, 64 );
@@ -808,7 +808,6 @@ int scanhash_sonoa_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t lane_hash[8] __attribute__ ((aligned (32)));
      uint32_t *hash7 = &(hash[7<<2]);
      uint32_t *pdata = work->data;
@@ -824,9 +823,7 @@ int scanhash_sonoa_4way( struct work *work, uint32_t max_nonce,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
      // Need big endian data
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
      for ( int m=0; m < 6; m++ ) if ( Htarg <= htmax[m] )
      {
         uint32_t mask = masks[m];

algo/x17/x17-4way.c

@@ -68,7 +68,7 @@ void x17_4way_hash( void *state, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serialize
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 3 Groestl
      init_groestl( &ctx.groestl, 64 );
@@ -81,7 +81,7 @@ void x17_4way_hash( void *state, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
 
      // Parallellize
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      // 4 Skein parallel 4 way 64 bit 
      skein512_4way_init( &ctx.skein );
@@ -99,7 +99,7 @@ void x17_4way_hash( void *state, const void *input )
      keccak512_4way_close( &ctx.keccak, vhash );
 
      // 7 Luffa  parallel 2 way 128 bit
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
@@ -124,8 +124,8 @@ void x17_4way_hash( void *state, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, 512 );
 
-     mm256_dintrlv_2x128_512( hash0, hash1, vhashA );
-     mm256_dintrlv_2x128_512( hash2, hash3, vhashB );
+     dintrlv_2x128_512( hash0, hash1, vhashA );
+     dintrlv_2x128_512( hash2, hash3, vhashB );
 
      // 11 Echo serial
      init_echo( &ctx.echo, 512 );
@@ -142,13 +142,13 @@ void x17_4way_hash( void *state, const void *input )
                        (const BitSequence *) hash3, 512 );
 
      // 12 Hamsi parallel 4 way 64 bit
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, 64 );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
 
      // 13 Fugue serial
      sph_fugue512_init( &ctx.fugue );
@@ -188,14 +188,14 @@ void x17_4way_hash( void *state, const void *input )
      sph_whirlpool_close( &ctx.whirlpool, hash3 );
 
      // 16 SHA512 parallel 64 bit 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
 
      sha512_4way_init( &ctx.sha512 );
      sha512_4way( &ctx.sha512, vhash, 64 );
      sha512_4way_close( &ctx.sha512, vhash );     
 
      // 17 Haval parallel 32 bit
-     mm256_rintrlv_4x64_4x32( vhashB, vhash,  512 );
+     rintrlv_4x64_4x32( vhashB, vhash,  512 );
 
      haval256_5_4way_init( &ctx.haval );
      haval256_5_4way( &ctx.haval, vhashB, 64 );
@@ -207,7 +207,6 @@ int scanhash_x17_4way( struct work *work, uint32_t max_nonce,
 {
      uint32_t hash[4*8] __attribute__ ((aligned (64)));
      uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
      uint32_t lane_hash[8] __attribute__ ((aligned (32)));
      uint32_t *hash7 = &(hash[7<<2]);
      uint32_t *pdata = work->data;
@@ -223,9 +222,7 @@ int scanhash_x17_4way( struct work *work, uint32_t max_nonce,
                           0xFFFFF000, 0xFFFF0000,          0  };
 
      // Need big endian data
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//     mm256_bswap_intrlv80_4x64( vdata, pdata );
+     mm256_bswap32_intrlv80_4x64( vdata, pdata );
      for ( int m = 0; m < 6; m++ ) if ( Htarg <= htmax[m] )
      {
         uint32_t mask = masks[ m ];
@@ -235,7 +232,7 @@ int scanhash_x17_4way( struct work *work, uint32_t max_nonce,
 	              _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
            x17_4way_hash( hash, vdata );
 
-	     for ( int lane = 0; lane < 4; lane++ )
+           for ( int lane = 0; lane < 4; lane++ )
            if ( ( hash7[ lane ] & mask ) == 0 )
            {
               extr_lane_4x32( lane_hash, hash, lane, 256 );

algo/x17/xevan-4way.c

@@ -71,7 +71,7 @@ void xevan_4way_hash( void *output, const void *input )
      bmw512_4way_close( &ctx.bmw, vhash );
 
      // Serial
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, dataLen<<3 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, dataLen<<3 );
 
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0,
@@ -87,7 +87,7 @@ void xevan_4way_hash( void *output, const void *input )
                                dataLen<<3 );
 
      // Parallel 4way
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, dataLen );
@@ -101,7 +101,7 @@ void xevan_4way_hash( void *output, const void *input )
      keccak512_4way( &ctx.keccak, vhash, dataLen );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, dataLen<<3 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, dataLen<<3 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, dataLen );
@@ -123,8 +123,8 @@ void xevan_4way_hash( void *output, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, dataLen<<3 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, dataLen<<3 );
+     dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
+     dintrlv_2x128( hash2, hash3, vhashB, dataLen<<3 );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -139,13 +139,13 @@ void xevan_4way_hash( void *output, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, dataLen<<3 );
      // Parallel
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, dataLen );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, dataLen<<3 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, dataLen<<3 );
 
      sph_fugue512_init( &ctx.fugue );
      sph_fugue512( &ctx.fugue, hash0, dataLen );
@@ -183,19 +183,19 @@ void xevan_4way_hash( void *output, const void *input )
      sph_whirlpool( &ctx.whirlpool, hash3, dataLen );
      sph_whirlpool_close( &ctx.whirlpool, hash3 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
 
      sha512_4way_init( &ctx.sha512 );
      sha512_4way( &ctx.sha512, vhash, dataLen );
      sha512_4way_close( &ctx.sha512, vhash );
 
-     mm256_rintrlv_4x64_4x32( vhashA, vhash, dataLen<<3 );
+     rintrlv_4x64_4x32( vhashA, vhash, dataLen<<3 );
 
      haval256_5_4way_init( &ctx.haval );
      haval256_5_4way( &ctx.haval, vhashA, dataLen );
      haval256_5_4way_close( &ctx.haval, vhashA );
 
-     mm256_rintrlv_4x32_4x64( vhash, vhashA, dataLen<<3 );
+     rintrlv_4x32_4x64( vhash, vhashA, dataLen<<3 );
 
      memset( &vhash[ 4<<2 ], 0, (dataLen-32) << 2 );
 
@@ -207,7 +207,7 @@ void xevan_4way_hash( void *output, const void *input )
      bmw512_4way( &ctx.bmw, vhash, dataLen );
      bmw512_4way_close( &ctx.bmw, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, dataLen<<3 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, dataLen<<3 );
 
      init_groestl( &ctx.groestl, 64 );
      update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0,
@@ -222,7 +222,7 @@ void xevan_4way_hash( void *output, const void *input )
      update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3,
                                dataLen<<3 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
 
      skein512_4way_init( &ctx.skein );
      skein512_4way( &ctx.skein, vhash, dataLen );
@@ -236,7 +236,7 @@ void xevan_4way_hash( void *output, const void *input )
      keccak512_4way( &ctx.keccak, vhash, dataLen );
      keccak512_4way_close( &ctx.keccak, vhash );
 
-     mm256_rintrlv_4x64_2x128( vhashA, vhashB, vhash, dataLen<<3 );
+     rintrlv_4x64_2x128( vhashA, vhashB, vhash, dataLen<<3 );
 
      luffa_2way_init( &ctx.luffa, 512 );
      luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, dataLen );
@@ -258,8 +258,8 @@ void xevan_4way_hash( void *output, const void *input )
      simd_2way_init( &ctx.simd, 512 );
      simd_2way_update_close( &ctx.simd, vhashB, vhashB, dataLen<<3 );
 
-     mm256_dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
-     mm256_dintrlv_2x128( hash2, hash3, vhashB, dataLen<<3 );
+     dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
+     dintrlv_2x128( hash2, hash3, vhashB, dataLen<<3 );
 
      init_echo( &ctx.echo, 512 );
      update_final_echo( &ctx.echo, (BitSequence *)hash0,
@@ -274,13 +274,13 @@ void xevan_4way_hash( void *output, const void *input )
      update_final_echo( &ctx.echo, (BitSequence *)hash3,
                        (const BitSequence *) hash3, dataLen<<3 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
 
      hamsi512_4way_init( &ctx.hamsi );
      hamsi512_4way( &ctx.hamsi, vhash, dataLen );
      hamsi512_4way_close( &ctx.hamsi, vhash );
 
-     mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, dataLen<<3 );
+     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, dataLen<<3 );
 
      sph_fugue512_init( &ctx.fugue );
      sph_fugue512( &ctx.fugue, hash0, dataLen );
@@ -316,13 +316,13 @@ void xevan_4way_hash( void *output, const void *input )
      sph_whirlpool( &ctx.whirlpool, hash3, dataLen );
      sph_whirlpool_close( &ctx.whirlpool, hash3 );
 
-     mm256_intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
+     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, dataLen<<3 );
 
      sha512_4way_init( &ctx.sha512 );
      sha512_4way( &ctx.sha512, vhash, dataLen );
      sha512_4way_close( &ctx.sha512, vhash );
 
-     mm256_rintrlv_4x64_4x32( vhashA, vhash, dataLen<<3 );
+     rintrlv_4x64_4x32( vhashA, vhash, dataLen<<3 );
 
      haval256_5_4way_init( &ctx.haval );
      haval256_5_4way( &ctx.haval, vhashA, dataLen );
@@ -334,7 +334,6 @@ int scanhash_xevan_4way( struct work *work, uint32_t max_nonce,
 {
    uint32_t hash[4*8] __attribute__ ((aligned (64)));
    uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-    uint32_t edata[20] __attribute__ ((aligned (64)));
    uint32_t lane_hash[8] __attribute__ ((aligned (32)));
    uint32_t *hash7 = &(hash[7<<2]);
    uint32_t *pdata = work->data;
@@ -349,9 +348,7 @@ int scanhash_xevan_4way( struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ptarget[7] = 0x0cff;
 
-    swab32_array( edata, pdata, 20 );
-    mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
-//   mm256_bswap_intrlv80_4x64( vdata, pdata );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
    do {
       *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
                _mm256_set_epi32( n+3, 0,n+2, 0,n+1, 0, n, 0 ) ), *noncev );

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 3.9.5.2.
+# Generated by GNU Autoconf 2.69 for cpuminer-opt 3.9.5.4.
 #
 #
 # 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='3.9.5.2'
-PACKAGE_STRING='cpuminer-opt 3.9.5.2'
+PACKAGE_VERSION='3.9.5.4'
+PACKAGE_STRING='cpuminer-opt 3.9.5.4'
 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 3.9.5.2 to adapt to many kinds of systems.
+\`configure' configures cpuminer-opt 3.9.5.4 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 3.9.5.2:";;
+     short | recursive ) echo "Configuration of cpuminer-opt 3.9.5.4:";;
    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 3.9.5.2
+cpuminer-opt configure 3.9.5.4
 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 3.9.5.2, which was
+It was created by cpuminer-opt $as_me 3.9.5.4, 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='3.9.5.2'
+ VERSION='3.9.5.4'
 
 
 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 3.9.5.2, which was
+This file was extended by cpuminer-opt $as_me 3.9.5.4, 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 3.9.5.2
+cpuminer-opt config.status 3.9.5.4
 configured by $0, generated by GNU Autoconf 2.69,
   with options \\"\$ac_cs_config\\"
 

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([cpuminer-opt], [3.9.5.2])
+AC_INIT([cpuminer-opt], [3.9.5.4])
 
 AC_PREREQ([2.59c])
 AC_CANONICAL_SYSTEM

cpu-miner.c

@@ -848,7 +848,8 @@ static double   shash_sum   = 0.;
 static double   bhash_sum   = 0.;
 static double   time_sum    = 0.;
 static double   latency_sum = 0.;
-static uint64_t submits_sum = 0;
+static uint64_t submit_sum  = 0;
+static uint64_t reject_sum  = 0;
 
 struct share_stats_t
 {
@@ -943,7 +944,8 @@ static int share_result( int result, struct work *null_work,
    shash_sum   += share_hash;
    bhash_sum   += block_hash;
    time_sum    += share_time;
-   submits_sum ++;
+   submit_sum ++;
+   reject_sum += (uint64_t)!result;
    latency_sum += latency;
 
    pthread_mutex_unlock( &stats_lock );
@@ -2118,30 +2120,49 @@ static void *miner_thread( void *userdata )
           double   hash     = shash_sum;   shash_sum   = 0.;
           double   bhash    = bhash_sum;   bhash_sum   = 0.;
           double   time     = time_sum;    time_sum    = 0.;
-          uint64_t submits  = submits_sum; submits_sum = 0;
+          uint64_t submits  = submit_sum;  submit_sum  = 0;
+          uint64_t rejects  = reject_sum;  reject_sum  = 0;
           uint64_t latency  = latency_sum; latency_sum = 0;
           memcpy( &five_min_start, &time_now, sizeof time_now );
 
           pthread_mutex_unlock( &stats_lock );
 
-          double ghrate = global_hashrate;
-          double shrate = time == 0. ? 0. : hash / time;
-          double scaled_shrate = shrate;
-          double avg_share = bhash == 0. ? 0. : hash / bhash * 100.;
+          double   ghrate = global_hashrate;
+          double   scaled_ghrate = ghrate;
+          double   shrate = time == 0. ? 0. : hash / time;
+          double   scaled_shrate = shrate;
+          double   avg_share = bhash == 0. ? 0. : hash / bhash * 100.;
+          uint64_t avg_latency = 0;
+          double   latency_pc = 0.;
+          double   rejects_pc = 0.;
+          double   submit_rate = 0.;
           char shr[32];
           char shr_units[4] = {0};
+          char ghr[32];
+          char ghr_units[4] = {0};
           int temp = cpu_temp(0);
-          char timestr[32];
+          char tempstr[32];
 
-          latency = submits ? latency / submits : 0;
+          if ( submits )
+             avg_latency = latency / submits;
+
+          if ( time != 0. )
+          {
+            submit_rate = (double)submits*60. / time;
+            rejects_pc = (double)rejects / (time*10.);   
+            latency_pc =  (double)latency / ( time*10.);
+          }
+          
           scale_hash_for_display( &scaled_shrate, shr_units );
+          scale_hash_for_display( &scaled_ghrate, ghr_units );
+          sprintf( ghr, "%.2f %sH/s", scaled_ghrate, ghr_units );
 
           if ( use_colors )
           {
-             if ( shrate > (32.*ghrate) )
+             if ( shrate > (128.*ghrate) )
                 sprintf( shr, "%s%.2f %sH/s%s", CL_MAG, scaled_shrate,
                          shr_units, CL_WHT );
-             else if ( shrate > (8.*ghrate) )
+             else if ( shrate > (16.*ghrate) )
                 sprintf( shr, "%s%.2f %sH/s%s", CL_GRN, scaled_shrate,
                          shr_units, CL_WHT );
              else if ( shrate > 2.0*ghrate )
@@ -2153,53 +2174,99 @@ static void *miner_thread( void *userdata )
                 sprintf( shr, "%s%.2f %sH/s%s", CL_YLW, scaled_shrate,
                          shr_units, CL_WHT );
 
-             if ( temp >= 80 ) sprintf( timestr, "%s%d C%s",
+             if ( temp >= 80 ) sprintf( tempstr, "%s%d C%s",
                                                   CL_RED, temp, CL_WHT );
-             else if (temp >=70 ) sprintf( timestr, "%s%d C%s",
+             else if (temp >=70 ) sprintf( tempstr, "%s%d C%s",
                                                   CL_YLW, temp, CL_WHT );
-             else sprintf( timestr, "%d C", temp );
+             else sprintf( tempstr, "%d C", temp );
           }
           else
           {
              sprintf( shr, "%.2f %sH/s", scaled_shrate, shr_units );
-             sprintf( timestr, "%d C", temp );
+             sprintf( tempstr, "%d C", temp );
           }
 
+
+          applog(LOG_NOTICE,"Submitted %d shares in %dm%02ds.",
+                (uint64_t)submits, et.tv_sec / 60, et.tv_sec % 60 );
+          applog(LOG_NOTICE,"%d rejects (%.2f%%), %.5f%% block share.",     
+                rejects, rejects_pc, avg_share );
+          applog(LOG_NOTICE,"Avg hashrate: Miner %s, Share %s.", ghr, shr );
+          
+#if ((defined(_WIN64) || defined(__WINDOWS__)))
+          applog(LOG_NOTICE,"Shares/min: %.2f, latency %d ms (%.2f%%).",
+                submit_rate, avg_latency, latency_pc );
+
+#else
+          applog(LOG_NOTICE,"Shares/min: %.2f, latency %d ms (%.2f%%), temp: %s.",
+                submit_rate, avg_latency, latency_pc, tempstr );
+#endif
+
+/*
           applog(LOG_NOTICE,"Submitted %d shares in %dm%02ds, %.5f%% block share.",
                (uint64_t)submits, et.tv_sec / 60, et.tv_sec % 60, avg_share );    
 
 #if ((defined(_WIN64) || defined(__WINDOWS__)))
-          applog(LOG_NOTICE,"Share hashrate %s, latency %d ms.",
-                            shr, latency );
+          applog(LOG_NOTICE,"Share hashrate %s, latency %d ms (%.2f%%).",
+                            shr, avg_latency, latency_pc );
 #else
-          applog(LOG_NOTICE,"Share hashrate %s, latency %d ms, temp %s.",
-                            shr, latency, timestr );
+          applog(LOG_NOTICE,"Share hashrate %s, latency %d ms (%.2f%%), temp %s.",
+                            shr, avg_latency, latency_pc, tempstr );
 #endif
+*/
           applog(LOG_INFO,"- - - - - - - - - - - - - - - - - - - - - - - - - - -");
        }
 
        // display hashrate
-       if ( opt_hash_meter )
+       if ( !opt_quiet )
        {
           char hc[16];
           char hr[16];
           char hc_units[2] = {0,0};
           char hr_units[2] = {0,0};
-          double hashcount = thr_hashcount[thr_id];
-          double hashrate  = thr_hashrates[thr_id];
-          if ( hashcount )
+          double hashcount;
+          double hashrate;
+          if ( opt_hash_meter )
           {
-             scale_hash_for_display( &hashcount, hc_units );
-             scale_hash_for_display( &hashrate,  hr_units );
-             if ( hc_units[0] )
-                sprintf( hc, "%.2f", hashcount );
-             else // no fractions of a hash
-                sprintf( hc, "%.0f", hashcount );
-             sprintf( hr, "%.2f", hashrate );
-             applog( LOG_INFO, "CPU #%d: %s %sH, %s %sH/s",
-                               thr_id, hc, hc_units, hr, hr_units );
+             hashcount = thr_hashcount[thr_id];
+             hashrate  = thr_hashrates[thr_id];
+             if ( hashcount != 0. )
+             {
+                scale_hash_for_display( &hashcount, hc_units );
+                scale_hash_for_display( &hashrate,  hr_units );
+                if ( hc_units[0] )
+                   sprintf( hc, "%.2f", hashcount );
+                else // no fractions of a hash
+                   sprintf( hc, "%.0f", hashcount );
+                sprintf( hr, "%.2f", hashrate );
+                applog( LOG_INFO, "CPU #%d: %s %sH, %s %sH/s",
+                                  thr_id, hc, hc_units, hr, hr_units );
+             }
+          }
+          if ( thr_id == 0 )
+          {
+             hashcount = 0.;
+             hashrate = 0.;
+             for ( i = 0; i < opt_n_threads; i++ )
+             {
+                 hashrate  += thr_hashrates[i];
+                 hashcount += thr_hashcount[i];
+             }
+             if ( hashcount != 0. )
+             {
+                scale_hash_for_display( &hashcount, hc_units );
+                scale_hash_for_display( &hashrate,  hr_units );
+                if ( hc_units[0] )
+                   sprintf( hc, "%.2f", hashcount );
+                else  // no fractions of a hash
+                   sprintf( hc, "%.0f", hashcount );
+                sprintf( hr, "%.2f", hashrate );
+                applog( LOG_NOTICE, "Miner perf: %s %sH, %s %sH/s.",
+                                    hc, hc_units, hr, hr_units );
+             }
           }
        }
+
        // Display benchmark total
        // Update hashrate for API if no shares accepted yet.
        if ( ( opt_benchmark || !accepted_share_count ) 
@@ -2212,7 +2279,7 @@ static void *miner_thread( void *userdata )
               hashrate  += thr_hashrates[i];
               hashcount += thr_hashcount[i];
           }
-          if ( hashcount )
+          if ( hashcount != 0. )
           {
              global_hashcount = hashcount;
              global_hashrate  = hashrate;

simd-utils.h

@@ -174,33 +174,26 @@
 #if defined(__MMX__)
 
 // 64 bit vectors
-#include "simd-utils/simd-mmx.h"
-#include "simd-utils/intrlv-mmx.h"
+#include "simd-utils/simd-64.h"
+//#include "simd-utils/intrlv-mmx.h"
 
 #if defined(__SSE2__)
 
 // 128 bit vectors
-#include "simd-utils/simd-sse2.h"
-#include "simd-utils/intrlv-sse2.h"
+#include "simd-utils/simd-128.h"
 
 #if defined(__AVX__)
 
 // 256 bit vector basics
-#include "simd-utils/simd-avx.h"
-#include "simd-utils/intrlv-avx.h"
+#include "simd-utils/simd-256.h"
 
 #if defined(__AVX2__)
 
-// 256 bit everything else
-#include "simd-utils/simd-avx2.h"
-#include "simd-utils/intrlv-avx2.h"
-
 // Skylake-X has all these
 #if defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
 
 // 512 bit vectors
-#include "simd-utils/simd-avx512.h"
-#include "simd-utils/intrlv-avx512.h"
+#include "simd-utils/simd-512.h"
 
 #endif  // MMX
 #endif  // SSE2
@@ -208,7 +201,6 @@
 #endif  // AVX2
 #endif  // AVX512
 
-// Picks implementation based on available CPU features.
-#include "simd-utils/intrlv-selector.h"
+#include "simd-utils/intrlv.h"
 
 #endif  // SIMD_UTILS_H__

simd-utils/intrlv-avx.h

@@ -1,867 +0,0 @@
-#if !defined(INTRLV_AVX_H__)
-#define INTRLV_AVX_H__ 1
-
-#if  defined(__AVX__)
-
-// Convenient short cuts for local use only
-
-// Extract 64 bits from the low 128 bits of 256 bit vector.
-#define extr64_cast128_256( a, n ) \
-   _mm_extract_epi64( _mm256_castsi256_si128( a ), n )
-
-// Extract 32 bits from the low 128 bits of 256 bit vector.
-#define extr32_cast128_256( a, n ) \
-   _mm_extract_epi32( _mm256_castsi256_si128( a ), n )
-
-///////////////////////////////////////////////////////////
-//
-//          AVX 256 Bit Vectors
-//
-//  256 bit interleaving can be done with AVX.
-
-#define mm256_put_64( s0, s1, s2, s3) \
-  _mm256_set_epi64x( *((const uint64_t*)(s3)), *((const uint64_t*)(s2)), \
-                     *((const uint64_t*)(s1)), *((const uint64_t*)(s0)) )
-
-#define mm256_put_32( s00, s01, s02, s03, s04, s05, s06, s07 ) \
-  _mm256_set_epi32( *((const uint32_t*)(s07)), *((const uint32_t*)(s06)), \
-                    *((const uint32_t*)(s05)), *((const uint32_t*)(s04)), \
-                    *((const uint32_t*)(s03)), *((const uint32_t*)(s02)), \
-                    *((const uint32_t*)(s01)), *((const uint32_t*)(s00)) )
-
-#define mm256_get_64( s, i0, i1, i2, i3 ) \
-  _mm256_set_epi64x( ((const uint64_t*)(s))[i3], ((const uint64_t*)(s))[i2], \
-                     ((const uint64_t*)(s))[i1], ((const uint64_t*)(s))[i0] )
-
-#define mm256_get_32( s, i0, i1, i2, i3, i4, i5, i6, i7 ) \
-  _mm256_set_epi32( ((const uint32_t*)(s))[i7], ((const uint32_t*)(s))[i6], \
-                    ((const uint32_t*)(s))[i5], ((const uint32_t*)(s))[i4], \
-                    ((const uint32_t*)(s))[i3], ((const uint32_t*)(s))[i2], \
-                    ((const uint32_t*)(s))[i1], ((const uint32_t*)(s))[i0] )
-
-/*
-// Blend 2 vectors alternating hi & lo: { hi[n], lo[n-1], ... hi[1], lo[0] }
-#define mm256_intrlv_blend_128( hi, lo ) \
-                _mm256_blend_epi32( hi, lo, 0x0f )
-
-#define mm256_intrlv_blend_64( hi, lo ) \
-                _mm256_blend_epi32( hi, lo, 0x33 )
-
-#define mm256_intrlv_blend_32( hi, lo ) \
-           _mm256_blend_epi32( hi, lo, 0x55 )
-*/
-
-// Interleave 8x32_256
-#define mm256_intrlv_8x32_256( d, s0, s1, s2, s3, s4, s5, s6, s7 ) \
-{ \
-   __m128i s0hi = mm128_extr_hi128_256( s0 ); \
-   __m128i s1hi = mm128_extr_hi128_256( s1 ); \
-   __m128i s2hi = mm128_extr_hi128_256( s2 ); \
-   __m128i s3hi = mm128_extr_hi128_256( s3 ); \
-   __m128i s4hi = mm128_extr_hi128_256( s4 ); \
-   __m128i s5hi = mm128_extr_hi128_256( s5 ); \
-   __m128i s6hi = mm128_extr_hi128_256( s6 ); \
-   __m128i s7hi = mm128_extr_hi128_256( s7 ); \
-   casti_m256i( d,0 ) = _mm256_set_epi32( \
-                        extr32_cast128_256(s7,0), extr32_cast128_256(s6,0), \
-                        extr32_cast128_256(s5,0), extr32_cast128_256(s4,0), \
-                        extr32_cast128_256(s3,0), extr32_cast128_256(s2,0), \
-                        extr32_cast128_256(s1,0), extr32_cast128_256(s0,0) ); \
-   casti_m256i( d,1 ) = _mm256_set_epi32( \
-                        extr32_cast128_256(s7,1), extr32_cast128_256(s6,1), \
-                        extr32_cast128_256(s5,1), extr32_cast128_256(s4,1), \
-                        extr32_cast128_256(s3,1), extr32_cast128_256(s2,1), \
-                        extr32_cast128_256(s1,1), extr32_cast128_256(s0,1) ); \
-   casti_m256i( d,2 ) = _mm256_set_epi32( \
-                        extr32_cast128_256(s7,2), extr32_cast128_256(s6,2), \
-                        extr32_cast128_256(s5,2), extr32_cast128_256(s4,2), \
-                        extr32_cast128_256(s3,2), extr32_cast128_256(s2,2), \
-                        extr32_cast128_256(s1,2), extr32_cast128_256(s0,2) ); \
-   casti_m256i( d,3 ) = _mm256_set_epi32( \
-                        extr32_cast128_256(s7,3), extr32_cast128_256(s6,3), \
-                        extr32_cast128_256(s5,3), extr32_cast128_256(s4,3), \
-                        extr32_cast128_256(s3,3), extr32_cast128_256(s2,3), \
-                        extr32_cast128_256(s1,3), extr32_cast128_256(s0,3) ); \
-   casti_m256i( d,4 ) = _mm256_set_epi32( \
-                           mm128_extr_32(s7hi,0),    mm128_extr_32(s6hi,0), \
-                           mm128_extr_32(s5hi,0),    mm128_extr_32(s4hi,0), \
-                           mm128_extr_32(s3hi,0),    mm128_extr_32(s2hi,0), \
-                           mm128_extr_32(s1hi,0),    mm128_extr_32(s0hi,0) ); \
-   casti_m256i( d,5 ) = _mm256_set_epi32( \
-                           mm128_extr_32(s7hi,1),    mm128_extr_32(s6hi,1), \
-                           mm128_extr_32(s5hi,1),    mm128_extr_32(s4hi,1), \
-                           mm128_extr_32(s3hi,1),    mm128_extr_32(s2hi,1), \
-                           mm128_extr_32(s1hi,1),    mm128_extr_32(s0hi,1) ); \
-   casti_m256i( d,6 ) = _mm256_set_epi32( \
-                           mm128_extr_32(s7hi,2),    mm128_extr_32(s6hi,2), \
-                           mm128_extr_32(s5hi,2),    mm128_extr_32(s4hi,2), \
-                           mm128_extr_32(s3hi,2),    mm128_extr_32(s2hi,2), \
-                           mm128_extr_32(s1hi,2),    mm128_extr_32(s0hi,2) ); \
-   casti_m256i( d,7 ) = _mm256_set_epi32( \
-                           mm128_extr_32(s7hi,3),    mm128_extr_32(s6hi,3), \
-                           mm128_extr_32(s5hi,3),    mm128_extr_32(s4hi,3), \
-                           mm128_extr_32(s3hi,3),    mm128_extr_32(s2hi,3), \
-                           mm128_extr_32(s1hi,3),    mm128_extr_32(s0hi,3) ); \
-} while(0)
-
-#define mm256_intrlv_8x32_128( d, s0, s1, s2, s3, s4, s5, s6, s7 ) \
-{ \
-   casti_m256i( d,0 ) = _mm256_set_epi32( \
-                           mm128_extr_32(s7,0), mm128_extr_32(s6,0), \
-                           mm128_extr_32(s5,0), mm128_extr_32(s4,0), \
-                           mm128_extr_32(s3,0), mm128_extr_32(s2,0), \
-                           mm128_extr_32(s1,0), mm128_extr_32(s0,0) ); \
-   casti_m256i( d,1 ) = _mm256_set_epi32( \
-                           mm128_extr_32(s7,1), mm128_extr_32(s6,1), \
-                           mm128_extr_32(s5,1), mm128_extr_32(s4,1), \
-                           mm128_extr_32(s3,1), mm128_extr_32(s2,1), \
-                           mm128_extr_32(s1,1), mm128_extr_32(s0,1) ); \
-   casti_m256i( d,2 ) = _mm256_set_epi32( \
-                           mm128_extr_32(s7,2), mm128_extr_32(s6,2), \
-                           mm128_extr_32(s5,2), mm128_extr_32(s4,2), \
-                           mm128_extr_32(s3,2), mm128_extr_32(s2,2), \
-                           mm128_extr_32(s1,2), mm128_extr_32(s0,2) ); \
-   casti_m256i( d,3 ) = _mm256_set_epi32( \
-                           mm128_extr_32(s7,3), mm128_extr_32(s6,3), \
-                           mm128_extr_32(s5,3), mm128_extr_32(s4,3), \
-                           mm128_extr_32(s3,3), mm128_extr_32(s2,3), \
-                           mm128_extr_32(s1,3), mm128_extr_32(s0,3) ); \
-} while(0)
-
-/*
-#define mm256_bswap_intrlv_8x32_256( d, src ) \
-do { \
-  __m256i s0 = mm256_bswap_32( src ); \
-  __m128i s1 = _mm256_extracti128_si256( s0, 1 ); \
-  casti_m256i( d, 0 ) = _mm256_set1_epi32( _mm_extract_epi32( \
-                                     _mm256_castsi256_si128( s0 ), 0 ) ); \
-  casti_m256i( d, 1 ) = _mm256_set1_epi32( _mm_extract_epi32( \
-                                     _mm256_castsi256_si128( s0 ), 1 ) ); \
-  casti_m256i( d, 2 ) = _mm256_set1_epi32( _mm_extract_epi32( \
-                                     _mm256_castsi256_si128( s0 ), 2 ) ); \
-  casti_m256i( d, 3 ) = _mm256_set1_epi32( _mm_extract_epi32( \
-                                     _mm256_castsi256_si128( s0 ), 3 ) ); \
-  casti_m256i( d, 4 ) = _mm256_set1_epi32( _mm_extract_epi32( s1,   0 ) ); \
-  casti_m256i( d, 5 ) = _mm256_set1_epi32( _mm_extract_epi32( s1,   1 ) ); \
-  casti_m256i( d, 6 ) = _mm256_set1_epi32( _mm_extract_epi32( s1,   2 ) ); \
-  casti_m256i( d, 7 ) = _mm256_set1_epi32( _mm_extract_epi32( s1,   3 ) ); \
-} while(0)
-
-
-#define mm256_bswap_intrlv_8x32_128( d, src ) \
-do { \
-  __m128i ss = mm128_bswap_32( src ); \
-  casti_m256i( d, 0 ) = _mm256_set1_epi32( _mm_extract_epi32( ss, 0 ) ); \
-  casti_m256i( d, 1 ) = _mm256_set1_epi32( _mm_extract_epi32( ss, 1 ) ); \
-  casti_m256i( d, 2 ) = _mm256_set1_epi32( _mm_extract_epi32( ss, 2 ) ); \
-  casti_m256i( d, 3 ) = _mm256_set1_epi32( _mm_extract_epi32( ss, 3 ) ); \
-} while(0)
-*/
-
-#define mm256_dintrlv_8x32_256( d0, d1, d2, d3, d4, d5, d6, d7, s ) \
-do { \
-  __m256i s0 = casti_m256i(s,0); \
-  __m256i s1 = casti_m256i(s,1); \
-  __m256i s2 = casti_m256i(s,2); \
-  __m256i s3 = casti_m256i(s,3); \
-  __m256i s4 = casti_m256i(s,4); \
-  __m256i s5 = casti_m256i(s,5); \
-  __m256i s6 = casti_m256i(s,6); \
-  __m256i s7 = casti_m256i(s,7); \
-  __m128i s0hi = _mm256_extracti128_si256( s0, 1 ); \
-  __m128i s1hi = _mm256_extracti128_si256( s1, 1 ); \
-  __m128i s2hi = _mm256_extracti128_si256( s2, 1 ); \
-  __m128i s3hi = _mm256_extracti128_si256( s3, 1 ); \
-  __m128i s4hi = _mm256_extracti128_si256( s4, 1 ); \
-  __m128i s5hi = _mm256_extracti128_si256( s5, 1 ); \
-  __m128i s6hi = _mm256_extracti128_si256( s6, 1 ); \
-  __m128i s7hi = _mm256_extracti128_si256( s7, 1 ); \
-   d0 = _mm256_set_epi32( \
-              extr32_cast128_256( s7, 0 ), extr32_cast128_256( s6, 0 ), \
-              extr32_cast128_256( s5, 0 ), extr32_cast128_256( s4, 0 ), \
-              extr32_cast128_256( s3, 0 ), extr32_cast128_256( s2, 0 ), \
-              extr32_cast128_256( s1, 0 ), extr32_cast128_256( s0, 0 ) );\
-   d1 = _mm256_set_epi32( \
-              extr32_cast128_256( s7, 1 ), extr32_cast128_256( s6, 1 ), \
-              extr32_cast128_256( s5, 1 ), extr32_cast128_256( s4, 1 ), \
-              extr32_cast128_256( s3, 1 ), extr32_cast128_256( s2, 1 ), \
-              extr32_cast128_256( s1, 1 ), extr32_cast128_256( s0, 1 ) );\
-   d2 = _mm256_set_epi32( \
-              extr32_cast128_256( s7, 2 ), extr32_cast128_256( s6, 2 ), \
-              extr32_cast128_256( s5, 2 ), extr32_cast128_256( s4, 2 ), \
-              extr32_cast128_256( s3, 2 ), extr32_cast128_256( s2, 2 ), \
-              extr32_cast128_256( s1, 2 ), extr32_cast128_256( s0, 2 ) );\
-   d3 = _mm256_set_epi32( \
-              extr32_cast128_256( s7, 3 ), extr32_cast128_256( s6, 3 ), \
-              extr32_cast128_256( s5, 3 ), extr32_cast128_256( s4, 3 ), \
-              extr32_cast128_256( s3, 3 ), extr32_cast128_256( s2, 3 ), \
-              extr32_cast128_256( s1, 3 ), extr32_cast128_256( s0, 3 ) );\
-   d4 = _mm256_set_epi32( \
-              _mm_extract_epi32( s7hi, 0 ), _mm_extract_epi32( s6hi, 0 ), \
-              _mm_extract_epi32( s5hi, 0 ), _mm_extract_epi32( s4hi, 0 ), \
-              _mm_extract_epi32( s3hi, 0 ), _mm_extract_epi32( s2hi, 0 ), \
-              _mm_extract_epi32( s1hi, 0 ), _mm_extract_epi32( s0hi, 0 ) ); \
-   d5 = _mm256_set_epi32( \
-              _mm_extract_epi32( s7hi, 1 ), _mm_extract_epi32( s6hi, 1 ), \
-              _mm_extract_epi32( s5hi, 1 ), _mm_extract_epi32( s4hi, 1 ), \
-              _mm_extract_epi32( s3hi, 1 ), _mm_extract_epi32( s2hi, 1 ), \
-              _mm_extract_epi32( s1hi, 1 ), _mm_extract_epi32( s0hi, 1 ) ); \
-   d6 = _mm256_set_epi32( \
-              _mm_extract_epi32( s7hi, 2 ), _mm_extract_epi32( s6hi, 2 ), \
-              _mm_extract_epi32( s5hi, 2 ), _mm_extract_epi32( s4hi, 2 ), \
-              _mm_extract_epi32( s3hi, 2 ), _mm_extract_epi32( s2hi, 2 ), \
-              _mm_extract_epi32( s1hi, 2 ), _mm_extract_epi32( s0hi, 2 ) ); \
-   d7 = _mm256_set_epi32( \
-              _mm_extract_epi32( s7hi, 3 ), _mm_extract_epi32( s6hi, 3 ), \
-              _mm_extract_epi32( s5hi, 3 ), _mm_extract_epi32( s4hi, 3 ), \
-              _mm_extract_epi32( s3hi, 3 ), _mm_extract_epi32( s2hi, 3 ), \
-              _mm_extract_epi32( s1hi, 3 ), _mm_extract_epi32( s0hi, 3 ) ); \
-} while(0)
-
-#define mm128_dintrlv_8x32_128( d0, d1, d2, d3, d4, d5, d6, d7, s ) \
-do { \
-   __m128i s0 = casti_m128i(s,0); \
-   __m128i s1 = casti_m128i(s,1); \
-   __m128i s2 = casti_m128i(s,2); \
-   __m128i s3 = casti_m128i(s,3); \
-   d0 = _mm_set_epi32( \
-              _mm_extract_epi32( s3, 0 ), _mm_extract_epi32( s2, 0 ), \
-              _mm_extract_epi32( s1, 0 ), _mm_extract_epi32( s0, 0 ) ); \
-   d1 = _mm_set_epi32( \
-              _mm_extract_epi32( s3, 1 ), _mm_extract_epi32( s2, 0 ), \
-              _mm_extract_epi32( s1, 1 ), _mm_extract_epi32( s0, 0 ) ); \
-   d2 = _mm_set_epi32( \
-              _mm_extract_epi32( s3, 0 ), _mm_extract_epi32( s2, 0 ), \
-              _mm_extract_epi32( s1, 0 ), _mm_extract_epi32( s0, 0 ) ); \
-   d3 = _mm_set_epi32( \
-              _mm_extract_epi32( s3, 0 ), _mm_extract_epi32( s2, 0 ), \
-              _mm_extract_epi32( s1, 0 ), _mm_extract_epi32( s0, 0 ) ); \
-   d4 = _mm_set_epi32( \
-              _mm_extract_epi32( s3, 0 ), _mm_extract_epi32( s2, 0 ), \
-              _mm_extract_epi32( s1, 0 ), _mm_extract_epi32( s0, 0 ) ); \
-   d5 = _mm_set_epi32( \
-              _mm_extract_epi32( s3, 0 ), _mm_extract_epi32( s2, 0 ), \
-              _mm_extract_epi32( s1, 0 ), _mm_extract_epi32( s0, 0 ) ); \
-   d6 = _mm_set_epi32( \
-              _mm_extract_epi32( s3, 0 ), _mm_extract_epi32( s2, 0 ), \
-              _mm_extract_epi32( s1, 0 ), _mm_extract_epi32( s0, 0 ) ); \
-   d7 = _mm_set_epi32( \
-              _mm_extract_epi32( s3, 0 ), _mm_extract_epi32( s2, 0 ), \
-              _mm_extract_epi32( s1, 0 ), _mm_extract_epi32( s0, 0 ) ); \
-} while(0)
-
-#define mm256_intrlv_4x64_256( d, s0, s1, s2, s3 ) \
-do { \
-  __m128i s0hi = _mm256_extracti128_si256( s0, 1 ); \
-  __m128i s1hi = _mm256_extracti128_si256( s1, 1 ); \
-  __m128i s2hi = _mm256_extracti128_si256( s2, 1 ); \
-  __m128i s3hi = _mm256_extracti128_si256( s3, 1 ); \
-  casti_m256i( d,0 ) = _mm256_set_epi64x( \
-                extr64_cast128_256( s3, 0 ), extr64_cast128_256( s2, 0 ), \
-                extr64_cast128_256( s1, 0 ), extr64_cast128_256( s0, 0 ) ); \
-  casti_m256i( d,1 ) = _mm256_set_epi64x( \
-                extr64_cast128_256( s3, 1 ), extr64_cast128_256( s2, 1 ), \
-                extr64_cast128_256( s1, 1 ), extr64_cast128_256( s0, 1 ) ); \
-  casti_m256i( d,2 ) = _mm256_set_epi64x( \
-                  _mm_extract_epi64( s3hi,0 ), _mm_extract_epi64( s2hi,0 ), \
-                  _mm_extract_epi64( s1hi,0 ), _mm_extract_epi64( s0hi,0 ) ); \
-  casti_m256i( d,3 ) = _mm256_set_epi64x( \
-                  _mm_extract_epi64( s3hi,1 ), _mm_extract_epi64( s2hi,1 ), \
-                  _mm_extract_epi64( s1hi,1 ), _mm_extract_epi64( s0hi,1 ) ); \
-} while(0)
-
-#define mm256_intrlv_4x64_128( d, s0, s1, s2, s3 ) \
-do { \
-  casti_m256i( d,0 ) = _mm256_set_epi64x( \
-                  _mm_extract_epi64( s3, 0 ), _mm_extract_epi64( s2, 0 ), \
-                  _mm_extract_epi64( s1, 0 ), _mm_extract_epi64( s0, 0 ) ); \
-  casti_m256i( d,1 ) = _mm256_set_epi64x( \
-                  _mm_extract_epi64( s3, 1 ), _mm_extract_epi64( s2, 1 ), \
-                  _mm_extract_epi64( s1, 1 ), _mm_extract_epi64( s0, 1 ) ); \
-} while(0)
-
-/*
-#define mm256_bswap_intrlv_4x64_256( d, src ) \
-do { \
-  __m256i s0 = mm256_bswap_32( src ); \
-  __m128i s1 = _mm256_extracti128_si256( s0, 1 ); \
-  casti_m256i( d,0 ) = _mm256_set1_epi64x( _mm_extract_epi64( \
-                                        _mm256_castsi256_si128( s0 ), 0 ) ); \
-  casti_m256i( d,1 ) = _mm256_set1_epi64x( _mm_extract_epi64( \
-                                        _mm256_castsi256_si128( s0 ), 1 ) ); \
-  casti_m256i( d,2 ) = _mm256_set1_epi64x(   _mm_extract_epi64( s1,   0 ) ); \
-  casti_m256i( d,3 ) = _mm256_set1_epi64x(   _mm_extract_epi64( s1,   1 ) ); \
-} while(0)
-
-#define mm256_bswap_intrlv_4x64_128( d, src ) \
-do { \
-  __m128i ss = mm128_bswap_32( src ); \
-  casti_m256i( d,0 ) = _mm256_set1_epi64x( _mm_extract_epi64( ss, 0 ) ); \
-  casti_m256i( d,1 ) = _mm256_set1_epi64x( _mm_extract_epi64( ss, 1 ) ); \
-} while(0)
-*/
-
-// 4 lanes of 256 bits using 64 bit interleaving (standard final hash size)
-static inline void mm256_dintrlv_4x64_256( void *d0, void *d1, void *d2,
-                            void *d3, const int n, const void *src )
-{
-   __m256i s0   = *( (__m256i*) src     );            // s[0][1:0]
-   __m256i s1   = *( (__m256i*)(src+32) );            // s[1][1:0]
-   __m256i s2   = *( (__m256i*)(src+64) );            // s[2][1:0]
-   __m256i s3   = *( (__m256i*)(src+96) );            // s[3][2:0]
-   __m128i s0hi = _mm256_extracti128_si256( s0, 1 );  // s[0][3:2]
-   __m128i s1hi = _mm256_extracti128_si256( s1, 1 );  // s[1][3:2]
-   __m128i s2hi = _mm256_extracti128_si256( s2, 1 );  // s[2][3:2]
-   __m128i s3hi = _mm256_extracti128_si256( s3, 1 );  // s[3][3:2]
-
-   casti_m256i( d0,n ) = _mm256_set_epi64x(
-              extr64_cast128_256( s3, 0 ), extr64_cast128_256( s2, 0 ),
-              extr64_cast128_256( s1, 0 ), extr64_cast128_256( s0, 0 ) );
-   casti_m256i( d1,n ) = _mm256_set_epi64x(
-              extr64_cast128_256( s3, 1 ), extr64_cast128_256( s2, 1 ),
-              extr64_cast128_256( s1, 1 ), extr64_cast128_256( s0, 1 ) );
-   casti_m256i( d2,n ) = _mm256_set_epi64x(
-              _mm_extract_epi64( s3hi, 0 ), _mm_extract_epi64( s2hi, 0 ),
-              _mm_extract_epi64( s1hi, 0 ), _mm_extract_epi64( s0hi, 0 ) );
-   casti_m256i( d3,n ) = _mm256_set_epi64x(
-              _mm_extract_epi64( s3hi, 1 ), _mm_extract_epi64( s2hi, 1 ),
-              _mm_extract_epi64( s1hi, 1 ), _mm_extract_epi64( s0hi, 1 ) );
-}
-
-
-// quarter avx2 block, 16 bytes * 4 lanes
-// 4 lanes of 128 bits using 64 bit interleaving
-// Used for last 16 bytes of 80 byte input, only used for testing.
-static inline void mm128_dintrlv_4x64_128( void *d0, void *d1, void *d2,
-                                  void *d3, const int n, const void *src )
-{
-  __m256i s0 = *( (__m256i*) src     );
-  __m256i s1 = *( (__m256i*)(src+32) );
-  __m128i s0hi = _mm256_extracti128_si256( s0, 1 );
-  __m128i s1hi = _mm256_extracti128_si256( s1, 1 );
-
-  casti_m128i( d0,n ) = _mm_set_epi64x( extr64_cast128_256( s1  , 0 ),
-                                        extr64_cast128_256( s0  , 0 ) );
-  casti_m128i( d1,n ) = _mm_set_epi64x( extr64_cast128_256( s1  , 1 ),
-                                        extr64_cast128_256( s0  , 1 ) );
-  casti_m128i( d2,n ) = _mm_set_epi64x( _mm_extract_epi64(    s1hi, 0 ),
-                                        _mm_extract_epi64(    s0hi, 0 ) );
-  casti_m128i( d3,n ) = _mm_set_epi64x( _mm_extract_epi64(    s1hi, 1 ),
-                                        _mm_extract_epi64(    s0hi, 1 ) );
-}
-
-/*
-static inline void mm256_dintrlv_2x128x256( void *d0, void *d1,
-                                                 const int n, const void *s )
-{
-   casti_m256i( d0,n ) = mm256_get_64( s, 0, 1, 4, 5 );
-   casti_m256i( d1,n ) = mm256_get_64( s, 2, 3, 6, 7 );
-}
-*/
-//
-
-#define mm256_intrlv_4x32_256( d, s0, s1, s2, s3 ) \
-do { \
-   casti_m256i( d,0 ) = _mm256_set_epi32( \
-                            mm128_extr_32( s3, 1 ), mm128_extr_32( s2, 1 ), \
-                            mm128_extr_32( s1, 1 ), mm128_extr_32( s0, 1 ), \
-                            mm128_extr_32( s3, 0 ), mm128_extr_32( s2, 0 ), \
-                            mm128_extr_32( s1, 0 ), mm128_extr_32( s0, 0 ) ); \
-   casti_m256i( d,1 ) = _mm256_set_epi32( \
-                            mm128_extr_32( s3, 3 ), mm128_extr_32( s2, 3 ), \
-                            mm128_extr_32( s1, 3 ), mm128_extr_32( s0, 3 ), \
-                            mm128_extr_32( s3, 2 ), mm128_extr_32( s2, 2 ), \
-                            mm128_extr_32( s1, 2 ), mm128_extr_32( s0, 2 ) ); \
-   casti_m256i( d,2 ) = _mm256_set_epi32( \
-                            mm128_extr_32( s3, 5 ), mm128_extr_32( s2, 5 ), \
-                            mm128_extr_32( s1, 5 ), mm128_extr_32( s0, 5 ), \
-                            mm128_extr_32( s3, 4 ), mm128_extr_32( s2, 4 ), \
-                            mm128_extr_32( s1, 4 ), mm128_extr_32( s0, 4 ) ); \
-   casti_m256i( d,3 ) = _mm256_set_epi32( \
-                            mm128_extr_32( s3, 7 ), mm128_extr_32( s2, 7 ), \
-                            mm128_extr_32( s1, 7 ), mm128_extr_32( s0, 7 ), \
-                            mm128_extr_32( s3, 6 ), mm128_extr_32( s2, 6 ), \
-                            mm128_extr_32( s1, 6 ), mm128_extr_32( s0, 6 ) ); \
-} while(0)
-
-// 256 bit versions of commmon 128 bit functions.
-static inline void mm256_intrlv_4x32( void *d, const void *s0,
-               const void *s1, const void *s2, const void *s3, int bit_len )
-{
-   mm256_intrlv_4x32_256( d     ,casti_m256i(s0,0), casti_m256i(s1,0),
-                                 casti_m256i(s2,0), casti_m256i(s3,0) );
-   if ( bit_len <= 256 ) return;
-   mm256_intrlv_4x32_256( d+128 ,casti_m256i(s0,1), casti_m256i(s1,1),
-                                 casti_m256i(s2,1), casti_m256i(s3,1) );
-   if ( bit_len <= 512 ) return;
-   if ( bit_len <= 640 )
-   {
-      mm128_intrlv_4x32_128( d+256, casti_m128i(s0,4), casti_m128i(s1,4),
-                                    casti_m128i(s2,4), casti_m128i(s3,4) );
-      return;
-   }
-   mm256_intrlv_4x32_256( d+256 ,casti_m256i(s0,2), casti_m256i(s1,2),
-                                 casti_m256i(s2,2), casti_m256i(s3,2) );
-   mm256_intrlv_4x32_256( d+384 ,casti_m256i(s0,3), casti_m256i(s1,3),
-                                 casti_m256i(s2,3), casti_m256i(s3,3) );
-}
-
-static inline void mm256_dintrlv_4x32_256( void *d0, void *d1, void *d2,
-                                           void *d3, const void *src )
-{
-   __m256i s0 = *(__m256i*) src;
-   __m256i s1 = *(__m256i*)(src+32);
-   __m256i s2 = *(__m256i*)(src+64);
-   __m256i s3 = *(__m256i*)(src+96);
-   *(__m256i*)d0 = _mm256_set_epi32(
-                  _mm256_extract_epi32( s3,4 ), _mm256_extract_epi32( s3,0 ),
-                  _mm256_extract_epi32( s2,4 ), _mm256_extract_epi32( s2,0 ),
-                  _mm256_extract_epi32( s1,4 ), _mm256_extract_epi32( s1,0 ),
-                  _mm256_extract_epi32( s0,4 ), _mm256_extract_epi32( s0,0 ) );
-   *(__m256i*)d1 = _mm256_set_epi32(
-                  _mm256_extract_epi32( s3,5 ), _mm256_extract_epi32( s3,1 ),
-                  _mm256_extract_epi32( s2,5 ), _mm256_extract_epi32( s2,1 ),
-                  _mm256_extract_epi32( s1,5 ), _mm256_extract_epi32( s1,1 ),
-                  _mm256_extract_epi32( s0,5 ), _mm256_extract_epi32( s0,1 ) );
-   *(__m256i*)d2 = _mm256_set_epi32(
-                  _mm256_extract_epi32( s3,6 ), _mm256_extract_epi32( s3,2 ),
-                  _mm256_extract_epi32( s2,6 ), _mm256_extract_epi32( s2,2 ),
-                  _mm256_extract_epi32( s1,6 ), _mm256_extract_epi32( s1,2 ),
-                  _mm256_extract_epi32( s0,6 ), _mm256_extract_epi32( s0,2 ) );
-   *(__m256i*)d3 = _mm256_set_epi32(
-                  _mm256_extract_epi32( s3,7 ), _mm256_extract_epi32( s3,3 ),
-                  _mm256_extract_epi32( s2,7 ), _mm256_extract_epi32( s2,3 ),
-                  _mm256_extract_epi32( s1,7 ), _mm256_extract_epi32( s1,3 ),
-                  _mm256_extract_epi32( s0,7 ), _mm256_extract_epi32( s0,3 ) );
-}
-
-static inline void mm256_dintrlv_4x32( void *d0, void *d1, void *d2,
-                                       void *d3, const void *s, int bit_len )
-{
-   mm256_dintrlv_4x32_256( d0    , d1    , d2    , d3    , s     );
-   if ( bit_len <= 256 ) return;
-   mm256_dintrlv_4x32_256( d0+ 32, d1+ 32, d2+ 32, d3+ 32, s+128 );
-   if ( bit_len <= 512 ) return;
-   if ( bit_len <= 640 )
-   {
-      mm128_dintrlv_4x32_128( d0+ 64, d1+ 64, d2+ 64, d3+ 64, s+256 );
-      return;
-   }
-   mm256_dintrlv_4x32_256( d0+ 64, d1+ 64, d2+ 64, d3+ 64, s+256 );
-   mm256_dintrlv_4x32_256( d0+ 96, d1+ 96, d2+ 96, d3+ 96, s+384 );
-}
-
-static inline void mm256_extr_lane_4x32( void *d, const void *s,
-                                         const int lane, const int bit_len )
-{
-  casti_m256i( d, 0 ) = mm256_get_32( s, lane   , lane+ 4, lane+ 8, lane+12,
-                                         lane+16, lane+20, lane+24, lane+28 );
-  if ( bit_len <= 256 ) return;
-  casti_m256i( d, 1 ) = mm256_get_32( s, lane+32, lane+36, lane+40, lane+44,
-                                         lane+48, lane+52, lane+56, lane+60 );
-}
-
-// Interleave 8 source buffers containing 32 bit data into the destination
-// vector
-static inline void mm256_intrlv_8x32( void *d, const void *s0,
-        const void *s1, const void *s2, const void *s3, const void *s4,
-        const void *s5, const void *s6, const void *s7, int bit_len )
-{
-   mm256_intrlv_8x32_256( d    , casti_m256i( s0,0 ), casti_m256i( s1,0 ),
-            casti_m256i( s2,0 ), casti_m256i( s3,0 ), casti_m256i( s4,0 ),
-            casti_m256i( s5,0 ), casti_m256i( s6,0 ), casti_m256i( s7,0 ) );
-   if ( bit_len <= 256 ) return;
-   mm256_intrlv_8x32_256( d+256, casti_m256i( s0,1 ), casti_m256i( s1,1 ), 
-            casti_m256i( s2,1 ), casti_m256i( s3,1 ), casti_m256i( s4,1 ),
-            casti_m256i( s5,1 ), casti_m256i( s6,1 ), casti_m256i( s7,1 ) );
-   if ( bit_len <= 512 ) return;
-   if ( bit_len <= 640 )
-   {
-      mm256_intrlv_8x32_128( d+512, casti_m128i( s0,4 ), casti_m128i( s1,4 ),
-               casti_m128i( s2,4 ), casti_m128i( s3,4 ), casti_m128i( s4,4 ),
-               casti_m128i( s5,4 ), casti_m128i( s6,4 ), casti_m128i( s7,4 ) );
-      return;
-   }
-   mm256_intrlv_8x32_256( d+512, casti_m256i( s0,2 ), casti_m256i( s1,2 ), 
-            casti_m256i( s2,2 ), casti_m256i( s3,2 ), casti_m256i( s4,2 ),
-            casti_m256i( s5,2 ), casti_m256i( s6,2 ), casti_m256i( s7,2 ) );
-   mm256_intrlv_8x32_256( d+768, casti_m256i( s0,3 ), casti_m256i( s1,3 ), 
-            casti_m256i( s2,3 ), casti_m256i( s3,3 ), casti_m256i( s4,3 ),
-            casti_m256i( s5,3 ), casti_m256i( s6,3 ), casti_m256i( s7,3 ) );
-   // bit_len == 1024
-}
-
-// A couple of mining specifi functions.
-/*
-// Interleave 80 bytes of 32 bit data for 8 lanes.
-static inline void mm256_bswap_intrlv80_8x32( void *d, const void *s )
-{
-   mm256_bswap_intrlv_8x32_256( d    , casti_m256i( s, 0 ) );
-   mm256_bswap_intrlv_8x32_256( d+256, casti_m256i( s, 1 ) );
-   mm256_bswap_intrlv_8x32_128( d+512, casti_m128i( s, 4 ) );
-}
-*/
-
-// Deinterleave 8 buffers of 32 bit data from the source buffer.
-// Sub-function can be called directly for 32 byte final hash.
-static inline void mm256_dintrlv_8x32( void *d0, void *d1, void *d2,
-                        void *d3, void *d4, void *d5, void *d6, void *d7,
-                        const void *s, int bit_len )
-{
-   mm256_dintrlv_8x32_256( casti_m256i(d0,0), casti_m256i(d1,0),
-        casti_m256i(d2,0), casti_m256i(d3,0), casti_m256i(d4,0),
-        casti_m256i(d5,0), casti_m256i(d6,0), casti_m256i(d7,0), s );
-   if ( bit_len <= 256 ) return;
-   mm256_dintrlv_8x32_256( casti_m256i(d0,1), casti_m256i(d1,1), 
-        casti_m256i(d2,1), casti_m256i(d3,1), casti_m256i(d4,1),
-        casti_m256i(d5,1), casti_m256i(d6,1), casti_m256i(d7,1), s+256 );
-   if ( bit_len <= 512 ) return;
-   // short block, final 16 bytes of input data
-   if ( bit_len <= 640 )
-   {
-      mm128_dintrlv_8x32_128( casti_m128i(d0,2), casti_m128i(d1,2), 
-           casti_m128i(d2,2), casti_m128i(d3,2), casti_m128i(d4,2),
-           casti_m128i(d5,2), casti_m128i(d6,2), casti_m128i(d7,2), s+512 );
-      return;
-   }
-   // bitlen == 1024
-   mm256_dintrlv_8x32_256( casti_m256i(d0,2), casti_m256i(d1,2), 
-        casti_m256i(d2,2), casti_m256i(d3,2), casti_m256i(d4,2),
-        casti_m256i(d5,2), casti_m256i(d6,2), casti_m256i(d7,2), s+512 );
-   mm256_dintrlv_8x32_256( casti_m256i(d0,3), casti_m256i(d1,3), 
-        casti_m256i(d2,3), casti_m256i(d3,3), casti_m256i(d4,3),
-        casti_m256i(d5,3), casti_m256i(d6,3), casti_m256i(d7,3), s+768 );
-}
-
-static inline void mm256_extr_lane_8x32( void *d, const void *s,
-                                            const int lane, const int bit_len )
-{
-  casti_m256i( d,0 ) = mm256_get_32(s, lane   , lane+  8, lane+ 16, lane+ 24,
-                                       lane+32, lane+ 40, lane+ 48, lane+ 56 );
-  if ( bit_len <= 256 ) return;
-  casti_m256i( d,1 ) = mm256_get_32(s, lane+64, lane+ 72, lane+ 80, lane+ 88,
-                                       lane+96, lane+104, lane+112, lane+120 );
-  // bit_len == 512
-}
-
-// Interleave 4 source buffers containing 64 bit data into the destination
-// buffer. Only bit_len 256, 512, 640 & 1024 are supported.
-static inline void mm256_intrlv_4x64( void *d, const void *s0,
-            const void *s1, const void *s2, const void *s3, int bit_len )
-{
-  mm256_intrlv_4x64_256( d    , casti_m256i(s0,0), casti_m256i(s1,0),
-                                casti_m256i(s2,0), casti_m256i(s3,0) );
-  if ( bit_len <= 256 ) return;
-  mm256_intrlv_4x64_256( d+128, casti_m256i(s0,1), casti_m256i(s1,1),
-                                casti_m256i(s2,1), casti_m256i(s3,1) );
-  if ( bit_len <= 512 ) return;
-  if ( bit_len <= 640 )
-  {
-    mm256_intrlv_4x64_128( d+256, casti_m128i(s0,4), casti_m128i(s1,4),
-                                  casti_m128i(s2,4), casti_m128i(s3,4) );
-    return;
-  }
-  // bit_len == 1024
-  mm256_intrlv_4x64_256( d+256, casti_m256i(s0,2), casti_m256i(s1,2),
-                                casti_m256i(s2,2), casti_m256i(s3,2) );
-  mm256_intrlv_4x64_256( d+384, casti_m256i(s0,3), casti_m256i(s1,3),
-                                casti_m256i(s2,3), casti_m256i(s3,3) );
-}
-/*
-// Interleave 80 bytes of 32 bit data for 8 lanes.
-static inline void mm256_bswap_intrlv80_4x64( void *d, const void *s )
-{
-   mm256_bswap_intrlv_4x64_256( d    , casti_m256i( s, 0 ) );
-   mm256_bswap_intrlv_4x64_256( d+128, casti_m256i( s, 1 ) );
-   mm256_bswap_intrlv_4x64_128( d+256, casti_m128i( s, 4 ) );
-}
-
-// Blend 32 byte lanes of hash from 2 sources according to control mask.
-// macro due to 256 bit value arg.
-#define mm256_blend_hash_4x64( dst, a, b, mask ) \
-do { \
-    dst[0] = _mm256_blendv_epi8( a[0], b[0], mask ); \
-    dst[1] = _mm256_blendv_epi8( a[1], b[1], mask ); \
-    dst[2] = _mm256_blendv_epi8( a[2], b[2], mask ); \
-    dst[3] = _mm256_blendv_epi8( a[3], b[3], mask ); \
-    dst[4] = _mm256_blendv_epi8( a[4], b[4], mask ); \
-    dst[5] = _mm256_blendv_epi8( a[5], b[5], mask ); \
-    dst[6] = _mm256_blendv_epi8( a[6], b[6], mask ); \
-    dst[7] = _mm256_blendv_epi8( a[7], b[7], mask ); \
-} while(0)
-*/
-
-// Deinterleave 4 buffers of 64 bit data from the source buffer.
-// bit_len must be 256, 512, 640 or 1024 bits.
-// Requires overrun padding for 640 bit len.
-static inline void mm256_dintrlv_4x64( void *d0, void *d1, void *d2,
-                                    void *d3, const void *s, int bit_len )
-{
-   mm256_dintrlv_4x64_256( d0, d1, d2, d3, 0, s );
-   if ( bit_len <= 256 ) return;
-   mm256_dintrlv_4x64_256( d0, d1, d2, d3, 1, s+128 );
-   if ( bit_len <= 512 ) return;
-   // short block, final 16 bytes of input data
-   if ( bit_len <= 640 )
-   {
-      mm128_dintrlv_4x64_128( d0, d1, d2, d3, 4, s+256 );
-      return;
-   }
-   // bit_len == 1024
-   mm256_dintrlv_4x64_256( d0, d1, d2, d3, 2, s+256 );
-   mm256_dintrlv_4x64_256( d0, d1, d2, d3, 3, s+384 );
-}
-
-// extract and deinterleave specified lane.
-#define mm256_extr_lane_4x64_256 \
-      casti_m256i( d, 0 ) = mm256_get_64( s, lane, lane+4, lane+8, lane+12 )
-static inline void mm256_extr_lane_4x64( void *d, const void *s,
-                                            const int lane, const int bit_len )
-{
-  casti_m256i( d, 0 ) = mm256_get_64( s, lane, lane+4, lane+8, lane+12 );
-  if ( bit_len <= 256 ) return;
-  casti_m256i( d, 1 ) = mm256_get_64( s, lane+16, lane+20, lane+24, lane+28 );
-  return;
-}
-
-
-// Convert from 4x32 SSE2 interleaving to 4x64 AVX2.
-// Can't do it in place
-static inline void mm256_rintrlv_4x32_4x64( void *dst, void *src,
-                                            int  bit_len )
-{
-   __m256i* d = (__m256i*)dst;
-   uint32_t *s = (uint32_t*)src;
-
-   d[0] = _mm256_set_epi32( s[ 7],s[ 3],s[ 6],s[ 2],s[ 5],s[ 1],s[ 4],s[ 0] );
-   d[1] = _mm256_set_epi32( s[15],s[11],s[14],s[10],s[13],s[ 9],s[12],s[ 8] );
-   d[2] = _mm256_set_epi32( s[23],s[19],s[22],s[18],s[21],s[17],s[20],s[16] );
-   d[3] = _mm256_set_epi32( s[31],s[27],s[30],s[26],s[29],s[25],s[28],s[24] );
-
-   if ( bit_len <= 256 ) return;
-
-   d[4] = _mm256_set_epi32( s[39],s[35],s[38],s[34],s[37],s[33],s[36],s[32] );
-   d[5] = _mm256_set_epi32( s[47],s[43],s[46],s[42],s[45],s[41],s[44],s[40] );
-   d[6] = _mm256_set_epi32( s[55],s[51],s[54],s[50],s[53],s[49],s[52],s[48] );
-   d[7] = _mm256_set_epi32( s[63],s[59],s[62],s[58],s[61],s[57],s[60],s[56] );
-
-   if ( bit_len <= 512 ) return;
-
-   d[8] = _mm256_set_epi32( s[71],s[67],s[70],s[66],s[69],s[65],s[68],s[64] );
-   d[9] = _mm256_set_epi32( s[79],s[75],s[78],s[74],s[77],s[73],s[76],s[72] );
-
-   if ( bit_len <= 640 ) return;
-
-  d[10] = _mm256_set_epi32(s[87],s[83],s[86],s[82],s[85],s[81],s[84],s[80]);
-  d[11] = _mm256_set_epi32(s[95],s[91],s[94],s[90],s[93],s[89],s[92],s[88]);
-
-  d[12] = _mm256_set_epi32(s[103],s[99],s[102],s[98],s[101],s[97],s[100],s[96]);
-  d[13] = _mm256_set_epi32(s[111],s[107],s[110],s[106],s[109],s[105],s[108],s[104]);
-  d[14] = _mm256_set_epi32(s[119],s[115],s[118],s[114],s[117],s[113],s[116],s[112]);
-  d[15] = _mm256_set_epi32(s[127],s[123],s[126],s[122],s[125],s[121],s[124],s[120]);
-   // bit_len == 1024
-}
-
-// Convert 4x64 byte (256 bit) vectors to 4x32 (128 bit) vectors for AVX
-// bit_len must be multiple of 64
-static inline void mm256_rintrlv_4x64_4x32( void *dst, void *src,
-                                            int  bit_len )
-{
-   __m256i  *d = (__m256i*)dst;
-   uint32_t *s = (uint32_t*)src;
-
-   d[0] = _mm256_set_epi32( s[ 7],s[ 5],s[ 3],s[ 1],s[ 6],s[ 4],s[ 2],s[ 0] );
-   d[1] = _mm256_set_epi32( s[15],s[13],s[11],s[ 9],s[14],s[12],s[10],s[ 8] );
-   d[2] = _mm256_set_epi32( s[23],s[21],s[19],s[17],s[22],s[20],s[18],s[16] );
-   d[3] = _mm256_set_epi32( s[31],s[29],s[27],s[25],s[30],s[28],s[26],s[24] );
-
-   if ( bit_len <= 256 ) return;
-
-   d[4] = _mm256_set_epi32( s[39],s[37],s[35],s[33],s[38],s[36],s[34],s[32] );
-   d[5] = _mm256_set_epi32( s[47],s[45],s[43],s[41],s[46],s[44],s[42],s[40] );
-   d[6] = _mm256_set_epi32( s[55],s[53],s[51],s[49],s[54],s[52],s[50],s[48] );
-   d[7] = _mm256_set_epi32( s[63],s[61],s[59],s[57],s[62],s[60],s[58],s[56] );
-
-   if ( bit_len <= 512 ) return;
-
-   d[8] = _mm256_set_epi32( s[71],s[69],s[67],s[65],s[70],s[68],s[66],s[64] );
-   d[9] = _mm256_set_epi32( s[79],s[77],s[75],s[73],s[78],s[76],s[74],s[72] );
-
-   if ( bit_len <= 640 ) return;
-
-   d[10] = _mm256_set_epi32( s[87],s[85],s[83],s[81],s[86],s[84],s[82],s[80] );
-   d[11] = _mm256_set_epi32( s[95],s[93],s[91],s[89],s[94],s[92],s[90],s[88] );
-
-   d[12] = _mm256_set_epi32( s[103],s[101],s[99],s[97],s[102],s[100],s[98],s[96] );
-   d[13] = _mm256_set_epi32( s[111],s[109],s[107],s[105],s[110],s[108],s[106],s[104] );
-   d[14] = _mm256_set_epi32( s[119],s[117],s[115],s[113],s[118],s[116],s[114],s[112] );
-   d[15] = _mm256_set_epi32( s[127],s[125],s[123],s[121],s[126],s[124],s[122],s[120] );
-   // bit_len == 1024
-}
-
-static inline void mm256_rintrlv_4x64_2x128( void *dst0, void *dst1,
-                                              const void *src, int  bit_len )
-{
-   __m256i* d0 = (__m256i*)dst0;
-   __m256i* d1 = (__m256i*)dst1;
-   uint64_t *s = (uint64_t*)src;
-
-   d0[0] = _mm256_set_epi64x( s[ 5], s[ 1], s[ 4], s[ 0] );
-   d1[0] = _mm256_set_epi64x( s[ 7], s[ 3], s[ 6], s[ 2] );
-
-   d0[1] = _mm256_set_epi64x( s[13], s[ 9], s[12], s[ 8] );
-   d1[1] = _mm256_set_epi64x( s[15], s[11], s[14], s[10] );
-
-   if ( bit_len <= 256 ) return;
-
-   d0[2] = _mm256_set_epi64x( s[21], s[17], s[20], s[16] );
-   d1[2] = _mm256_set_epi64x( s[23], s[19], s[22], s[18] );
-
-   d0[3] = _mm256_set_epi64x( s[29], s[25], s[28], s[24] );
-   d1[3] = _mm256_set_epi64x( s[31], s[27], s[30], s[26] );
-
-   if ( bit_len <= 512 ) return;
-
-   d0[4] = _mm256_set_epi64x( s[37], s[33], s[36], s[32] );
-   d1[4] = _mm256_set_epi64x( s[39], s[35], s[38], s[34] );
-
-   d0[5] = _mm256_set_epi64x( s[45], s[41], s[44], s[40] );
-   d1[5] = _mm256_set_epi64x( s[47], s[43], s[46], s[42] );
-
-   d0[6] = _mm256_set_epi64x( s[53], s[49], s[52], s[48] );
-   d1[6] = _mm256_set_epi64x( s[55], s[51], s[54], s[50] );
-
-   d0[7] = _mm256_set_epi64x( s[61], s[57], s[60], s[56] );
-   d1[7] = _mm256_set_epi64x( s[63], s[59], s[62], s[58] );
-}
-
-static inline void mm256_rintrlv_2x128_4x64( void *dst, const void *src0,
-                                         const void *src1, int  bit_len )
-{
-   __m256i* d = (__m256i*)dst;
-   uint64_t *s0 = (uint64_t*)src0;
-   uint64_t *s1 = (uint64_t*)src1;
-
-   d[ 0] = _mm256_set_epi64x( s1[2], s1[0], s0[2], s0[0] );
-   d[ 1] = _mm256_set_epi64x( s1[3], s1[1], s0[3], s0[1] );
-   d[ 2] = _mm256_set_epi64x( s1[6], s1[4], s0[6], s0[4] );
-   d[ 3] = _mm256_set_epi64x( s1[7], s1[5], s0[7], s0[5] );
-
-   if ( bit_len <= 256 ) return;
-
-   d[ 4] = _mm256_set_epi64x( s1[10], s1[ 8], s0[10], s0[ 8] );
-   d[ 5] = _mm256_set_epi64x( s1[11], s1[ 9], s0[11], s0[ 9] );
-   d[ 6] = _mm256_set_epi64x( s1[14], s1[12], s0[14], s0[12] );
-   d[ 7] = _mm256_set_epi64x( s1[15], s1[13], s0[15], s0[13] );
-
-   if ( bit_len <= 512 ) return;
-
-   d[ 8] = _mm256_set_epi64x( s1[18], s1[16], s0[18], s0[16] );
-   d[ 9] = _mm256_set_epi64x( s1[19], s1[17], s0[19], s0[17] );
-   d[10] = _mm256_set_epi64x( s1[22], s1[20], s0[22], s0[20] );
-   d[11] = _mm256_set_epi64x( s1[23], s1[21], s0[23], s0[21] );
-
-   d[12] = _mm256_set_epi64x( s1[26], s1[24], s0[26], s0[24] );
-   d[13] = _mm256_set_epi64x( s1[27], s1[25], s0[27], s0[25] );
-   d[14] = _mm256_set_epi64x( s1[30], s1[28], s0[30], s0[28] );
-   d[15] = _mm256_set_epi64x( s1[31], s1[29], s0[31], s0[29] );
-}
-
-
-static inline void mm256_intrlv_2x128( const void *d, const void *s0,
-                                      void *s1, const int bit_len )
-{
-  __m128i s1hi = _mm256_extracti128_si256( casti_m256i( s1,0 ), 1 );
-  __m128i s0hi = _mm256_extracti128_si256( casti_m256i( s0,0 ), 1 );
-  casti_m256i( d,0 ) = mm256_concat_128(
-                           _mm256_castsi256_si128( casti_m256i( s1,0 ) ),
-                           _mm256_castsi256_si128( casti_m256i( s0,0 ) ) );
-  casti_m256i( d,1 ) = mm256_concat_128( s1hi, s0hi );                  
-
-  if ( bit_len <= 256 ) return;
-  s0hi = _mm256_extracti128_si256( casti_m256i( s0,1 ), 1 );
-  s1hi = _mm256_extracti128_si256( casti_m256i( s1,1 ), 1 );
-  casti_m256i( d,2 ) = mm256_concat_128(
-                           _mm256_castsi256_si128( casti_m256i( s1,1 ) ),
-                           _mm256_castsi256_si128( casti_m256i( s0,1 ) ) );
-  casti_m256i( d,3 ) = mm256_concat_128( s1hi, s0hi );        
-
-  if ( bit_len <= 512 ) return;
-  if ( bit_len <= 640 )
-  {
-     casti_m256i( d,4 ) = mm256_concat_128(
-                           _mm256_castsi256_si128( casti_m256i( s1,2 ) ),
-                           _mm256_castsi256_si128( casti_m256i( s0,2 ) ) );
-     return;
-  }
-
-  s0hi = _mm256_extracti128_si256( casti_m256i( s0,2 ), 1 );
-  s1hi = _mm256_extracti128_si256( casti_m256i( s1,2 ), 1 );
-  casti_m256i( d,4 ) = mm256_concat_128(
-                           _mm256_castsi256_si128( casti_m256i( s1,2 ) ),
-                           _mm256_castsi256_si128( casti_m256i( s0,2 ) ) );
-  casti_m256i( d,5 ) = mm256_concat_128( s1hi, s0hi );        
-
-  s0hi = _mm256_extracti128_si256( casti_m256i( s0,3 ), 1 );
-  s1hi = _mm256_extracti128_si256( casti_m256i( s1,3 ), 1 );
-  casti_m256i( d,6 ) = mm256_concat_128(
-                           _mm256_castsi256_si128( casti_m256i( s1,3 ) ),
-                           _mm256_castsi256_si128( casti_m256i( s0,3 ) ) );
-  casti_m256i( d,7 ) = mm256_concat_128( s1hi, s0hi );        
-}
-
-// 512 is the bit len used by most, eliminate the conditionals
-static inline void mm256_dintrlv_2x128_512( void *dst0, void *dst1,
-                                            const void *s )
-{
-   __m256i *d0 = (__m256i*)dst0;
-   __m256i *d1 = (__m256i*)dst1;
-
-   __m256i s0 = casti_m256i( s, 0 );
-   __m256i s1 = casti_m256i( s, 1 );
-   d0[0] = _mm256_permute2x128_si256( s0, s1, 0x20 );
-   d1[0] = _mm256_permute2x128_si256( s0, s1, 0x31 );
-
-   s0 = casti_m256i( s, 2 );
-   s1 = casti_m256i( s, 3 );
-   d0[1] = _mm256_permute2x128_si256( s0, s1, 0x20 );
-   d1[1] = _mm256_permute2x128_si256( s0, s1, 0x31 );
-}   
-
-// Phase out usage for all 512 bit data lengths
-static inline void mm256_dintrlv_2x128( void *dst0, void *dst1, const void *s,
-                                             int bit_len )
-{
-   __m256i *d0 = (__m256i*)dst0;
-   __m256i *d1 = (__m256i*)dst1;
-
-   __m256i s0 = casti_m256i( s, 0 );
-   __m256i s1 = casti_m256i( s, 1 );
-   d0[0] = _mm256_permute2x128_si256( s0, s1, 0x20 );
-   d1[0] = _mm256_permute2x128_si256( s0, s1, 0x31 );
-
-   if ( bit_len <= 256 ) return;
-
-   s0 = casti_m256i( s, 2 );
-   s1 = casti_m256i( s, 3 );
-   d0[1] = _mm256_permute2x128_si256( s0, s1, 0x20 );
-   d1[1] = _mm256_permute2x128_si256( s0, s1, 0x31 );
-
-   if ( bit_len <= 512 ) return;
-
-   s0 = casti_m256i( s, 4 );
-   s1 = casti_m256i( s, 5 );
-   d0[2] = _mm256_permute2x128_si256( s0, s1, 0x20 );
-   d1[2] = _mm256_permute2x128_si256( s0, s1, 0x31 );
-
-   s0 = casti_m256i( s, 6 );
-   s1 = casti_m256i( s, 7 );
-   d0[3] = _mm256_permute2x128_si256( s0, s1, 0x20 );
-   d1[3] = _mm256_permute2x128_si256( s0, s1, 0x31 );
-}
-
-#undef extr64_cast128_256
-#undef extr32_cast128_256
-
-#endif // AVX
-#endif // INTRLV_AVX_H__

simd-utils/intrlv-avx2.h

@@ -1,104 +0,0 @@
-#if !defined(INTRLV_AVX2_H__)
-#define INTRLV_AVX2_H__ 1
-
-#if  defined(__AVX2__)
-
-///////////////////////////////////////////////////////////
-//
-//          AVX2 256 Bit Vectors
-//
-//  A few functions that need AVX2 for 256 bit.
-
-
-// Blend 2 vectors alternating hi & lo: { hi[n], lo[n-1], ... hi[1], lo[0] }
-#define mm256_intrlv_blend_128( hi, lo ) \
-                _mm256_blend_epi32( hi, lo, 0x0f )
-
-#define mm256_intrlv_blend_64( hi, lo ) \
-                _mm256_blend_epi32( hi, lo, 0x33 )
-
-#define mm256_intrlv_blend_32( hi, lo ) \
-           _mm256_blend_epi32( hi, lo, 0x55 )
-
-
-#define mm256_bswap_intrlv_8x32_256( d, src ) \
-do { \
-  __m256i s0 = mm256_bswap_32( src ); \
-  __m128i s1 = _mm256_extracti128_si256( s0, 1 ); \
-  casti_m256i( d, 0 ) = _mm256_set1_epi32( _mm_extract_epi32( \
-                                     _mm256_castsi256_si128( s0 ), 0 ) ); \
-  casti_m256i( d, 1 ) = _mm256_set1_epi32( _mm_extract_epi32( \
-                                     _mm256_castsi256_si128( s0 ), 1 ) ); \
-  casti_m256i( d, 2 ) = _mm256_set1_epi32( _mm_extract_epi32( \
-                                     _mm256_castsi256_si128( s0 ), 2 ) ); \
-  casti_m256i( d, 3 ) = _mm256_set1_epi32( _mm_extract_epi32( \
-                                     _mm256_castsi256_si128( s0 ), 3 ) ); \
-  casti_m256i( d, 4 ) = _mm256_set1_epi32( _mm_extract_epi32( s1,   0 ) ); \
-  casti_m256i( d, 5 ) = _mm256_set1_epi32( _mm_extract_epi32( s1,   1 ) ); \
-  casti_m256i( d, 6 ) = _mm256_set1_epi32( _mm_extract_epi32( s1,   2 ) ); \
-  casti_m256i( d, 7 ) = _mm256_set1_epi32( _mm_extract_epi32( s1,   3 ) ); \
-} while(0)
-
-#define mm256_bswap_intrlv_8x32_128( d, src ) \
-do { \
-  __m128i ss = mm128_bswap_32( src ); \
-  casti_m256i( d, 0 ) = _mm256_set1_epi32( _mm_extract_epi32( ss, 0 ) ); \
-  casti_m256i( d, 1 ) = _mm256_set1_epi32( _mm_extract_epi32( ss, 1 ) ); \
-  casti_m256i( d, 2 ) = _mm256_set1_epi32( _mm_extract_epi32( ss, 2 ) ); \
-  casti_m256i( d, 3 ) = _mm256_set1_epi32( _mm_extract_epi32( ss, 3 ) ); \
-} while(0)
-
-#define mm256_bswap_intrlv_4x64_256( d, src ) \
-do { \
-  __m256i s0 = mm256_bswap_32( src ); \
-  __m128i s1 = _mm256_extracti128_si256( s0, 1 ); \
-  casti_m256i( d,0 ) = _mm256_set1_epi64x( _mm_extract_epi64( \
-                                        _mm256_castsi256_si128( s0 ), 0 ) ); \
-  casti_m256i( d,1 ) = _mm256_set1_epi64x( _mm_extract_epi64( \
-                                        _mm256_castsi256_si128( s0 ), 1 ) ); \
-  casti_m256i( d,2 ) = _mm256_set1_epi64x(   _mm_extract_epi64( s1,   0 ) ); \
-  casti_m256i( d,3 ) = _mm256_set1_epi64x(   _mm_extract_epi64( s1,   1 ) ); \
-} while(0)
-
-#define mm256_bswap_intrlv_4x64_128( d, src ) \
-do { \
-  __m128i ss = mm128_bswap_32( src ); \
-  casti_m256i( d,0 ) = _mm256_set1_epi64x( _mm_extract_epi64( ss, 0 ) ); \
-  casti_m256i( d,1 ) = _mm256_set1_epi64x( _mm_extract_epi64( ss, 1 ) ); \
-} while(0)
-
-
-// A couple of mining specifi functions.
-
-// Interleave 80 bytes of 32 bit data for 8 lanes.
-static inline void mm256_bswap_intrlv80_8x32( void *d, const void *s )
-{
-   mm256_bswap_intrlv_8x32_256( d    , casti_m256i( s, 0 ) );
-   mm256_bswap_intrlv_8x32_256( d+256, casti_m256i( s, 1 ) );
-   mm256_bswap_intrlv_8x32_128( d+512, casti_m128i( s, 4 ) );
-}
-
-// Interleave 80 bytes of 32 bit data for 8 lanes.
-static inline void mm256_bswap_intrlv80_4x64( void *d, const void *s )
-{
-   mm256_bswap_intrlv_4x64_256( d    , casti_m256i( s, 0 ) );
-   mm256_bswap_intrlv_4x64_256( d+128, casti_m256i( s, 1 ) );
-   mm256_bswap_intrlv_4x64_128( d+256, casti_m128i( s, 4 ) );
-}
-
-// Blend 32 byte lanes of hash from 2 sources according to control mask.
-// macro due to 256 bit value arg.
-#define mm256_blend_hash_4x64( dst, a, b, mask ) \
-do { \
-    dst[0] = _mm256_blendv_epi8( a[0], b[0], mask ); \
-    dst[1] = _mm256_blendv_epi8( a[1], b[1], mask ); \
-    dst[2] = _mm256_blendv_epi8( a[2], b[2], mask ); \
-    dst[3] = _mm256_blendv_epi8( a[3], b[3], mask ); \
-    dst[4] = _mm256_blendv_epi8( a[4], b[4], mask ); \
-    dst[5] = _mm256_blendv_epi8( a[5], b[5], mask ); \
-    dst[6] = _mm256_blendv_epi8( a[6], b[6], mask ); \
-    dst[7] = _mm256_blendv_epi8( a[7], b[7], mask ); \
-} while(0)
-
-#endif // AVX2
-#endif // INTRLV_AVX2_H__

simd-utils/intrlv-avx512.h

@@ -1,679 +0,0 @@
-#if !defined(INTRLV_AVX512_H__)
-#define INTRLV_AVX512_H__ 1
-
-#if defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
-
-// SSE2 functions used in AVX512 interleaving
-
-// AVX512 block is 64 * 64 bytes
-
-// quarter avx512 block, 16 bytes * 16 lanes 
-static inline void mm128_dintrlv_16x32x128( void *d00, void *d01,
-        void *d02, void *d03, void *d04, void *d05, void *d06, void *d07,
-        void *d08, void *d09, void *d10, void *d11, void *d12, void *d13,
-        void *d14, void *d15, const int n, const void *s )
-{
-   cast_m128i( d00 ) = mm128_get_32( s, 0,  16, 32, 48 );
-   cast_m128i( d01 ) = mm128_get_32( s, 1,  17, 33, 49 );
-   cast_m128i( d02 ) = mm128_get_32( s, 2,  18, 34, 50 );
-   cast_m128i( d03 ) = mm128_get_32( s, 3,  19, 35, 51 );
-   cast_m128i( d04 ) = mm128_get_32( s, 4,  20, 36, 52 );
-   cast_m128i( d05 ) = mm128_get_32( s, 5,  21, 37, 53 );
-   cast_m128i( d06 ) = mm128_get_32( s, 6,  22, 38, 54 );
-   cast_m128i( d07 ) = mm128_get_32( s, 7,  23, 39, 55 );
-   cast_m128i( d08 ) = mm128_get_32( s, 8,  24, 40, 56 );
-   cast_m128i( d09 ) = mm128_get_32( s, 9,  25, 41, 57 );
-   cast_m128i( d10 ) = mm128_get_32( s, 10, 26, 42, 58 );
-   cast_m128i( d11 ) = mm128_get_32( s, 11, 27, 43, 59 );
-   cast_m128i( d12 ) = mm128_get_32( s, 12, 28, 44, 60 );
-   cast_m128i( d13 ) = mm128_get_32( s, 13, 29, 45, 61 );
-   cast_m128i( d14 ) = mm128_get_32( s, 14, 30, 46, 62 );
-   cast_m128i( d15 ) = mm128_get_32( s, 15, 31, 47, 63 );
-}
-
-// quarter avx512 block, 32 bytes * 8 lanes
-// 8 lanes of 128 bits using 64 bit interleaving
-// Used for last 16 bytes of 80 byte input, only used for testing.
-static inline void mm128_dintrlv_8x64x128( void *d0, void *d1, void *d2,
-                         void *d3, void *d4, void *d5, void *d6, void *d7,
-                         const int n, const void *s )
-{
-   casti_m128i( d0,n ) = mm128_get_64( s, 0,  8 );
-   casti_m128i( d1,n ) = mm128_get_64( s, 1,  9 );
-   casti_m128i( d2,n ) = mm128_get_64( s, 2, 10 );
-   casti_m128i( d3,n ) = mm128_get_64( s, 3, 11 );
-   casti_m128i( d4,n ) = mm128_get_64( s, 4, 12 );
-   casti_m128i( d5,n ) = mm128_get_64( s, 5, 13 );
-   casti_m128i( d6,n ) = mm128_get_64( s, 6, 14 );
-   casti_m128i( d7,n ) = mm128_get_64( s, 7, 15 );
-}
-
-static inline void mm128_dintrlv_4x128x128( void *d0, void *d1, void *d2,
-                                void *d3, const int n, const void *s )
-{
-   casti_m128i( d0,n ) = mm128_get_64( s, 0, 1 );
-   casti_m128i( d1,n ) = mm128_get_64( s, 2, 3 );
-   casti_m128i( d2,n ) = mm128_get_64( s, 4, 5 );
-   casti_m128i( d3,n ) = mm128_get_64( s, 5, 7 );
-}
-
-// AVX2 functions Used in AVX512 interleaving
-
-static inline void mm256_dintrlv_16x32x256( void *d00, void *d01,
-                 void *d02, void *d03, void *d04, void *d05,
-            void *d06, void *d07, void *d08, void *d09,
-            void *d10, void *d11, void *d12, void *d13,
-            void *d14, void *d15, const int n, const void *s )
-{
-   casti_m256i( d00,n ) = mm256_get_32( s,  0, 16, 32, 48, 64, 80, 96,112 );
-   casti_m256i( d01,n ) = mm256_get_32( s,  1, 17, 33, 49, 65, 81, 97,113 );
-   casti_m256i( d02,n ) = mm256_get_32( s,  2, 18, 34, 50, 66, 82, 98,114 );
-   casti_m256i( d03,n ) = mm256_get_32( s,  3, 19, 35, 51, 67, 83, 99,115 );
-   casti_m256i( d04,n ) = mm256_get_32( s,  4, 20, 36, 52, 68, 84,100,116 );
-   casti_m256i( d05,n ) = mm256_get_32( s,  5, 21, 37, 53, 69, 85,101,117 );
-   casti_m256i( d06,n ) = mm256_get_32( s,  6, 22, 38, 54, 70, 86,102,118 );
-   casti_m256i( d07,n ) = mm256_get_32( s,  7, 23, 39, 55, 71, 87,103,119 );
-   casti_m256i( d08,n ) = mm256_get_32( s,  8, 24, 40, 56, 72, 88,104,120 );
-   casti_m256i( d09,n ) = mm256_get_32( s,  9, 25, 41, 57, 73, 89,105,121 );
-   casti_m256i( d10,n ) = mm256_get_32( s, 10, 26, 42, 58, 74, 90,106,122 );
-   casti_m256i( d11,n ) = mm256_get_32( s, 11, 27, 43, 59, 75, 91,107,123 );
-   casti_m256i( d12,n ) = mm256_get_32( s, 12, 28, 44, 60, 76, 92,108,124 );
-   casti_m256i( d13,n ) = mm256_get_32( s, 13, 29, 45, 61, 77, 93,109,125 );
-   casti_m256i( d14,n ) = mm256_get_32( s, 14, 30, 46, 62, 78, 94,110,126 );
-   casti_m256i( d15,n ) = mm256_get_32( s, 15, 31, 47, 63, 79, 95,111,127 );
-}
-
-// 8 lanes of 256 bits using 64 bit interleaving (standard final hash size)
-static inline void mm256_dintrlv_8x64x256( void *d0, void *d1, void *d2,
-                            void *d3, void *d4, void *d5, void *d6, void *d7,
-                            const int n, const void *s )
-{
-   casti_m256i( d0,n ) = mm256_get_64( s,  0,  8, 16, 24 );
-   casti_m256i( d1,n ) = mm256_get_64( s,  1,  9, 17, 25 );
-   casti_m256i( d2,n ) = mm256_get_64( s,  2, 10, 18, 26 );
-   casti_m256i( d3,n ) = mm256_get_64( s,  3, 11, 19, 27 );
-   casti_m256i( d4,n ) = mm256_get_64( s,  4, 12, 20, 28 );
-   casti_m256i( d5,n ) = mm256_get_64( s,  5, 13, 21, 29 );
-   casti_m256i( d6,n ) = mm256_get_64( s,  6, 14, 22, 30 );
-   casti_m256i( d7,n ) = mm256_get_64( s,  7, 15, 23, 31 );
-}
-
-static inline void mm256_dintrlv_4x128x256( void *d0, void *d1, void *d2,
-                                void *d3, const int n, const void *s )
-{
-   casti_m256i( d0,n ) = mm256_get_64( s,  0,  1,  8,  9 );
-   casti_m256i( d1,n ) = mm256_get_64( s,  2,  3, 10, 11 );
-   casti_m256i( d2,n ) = mm256_get_64( s,  4,  5, 12, 13 );
-   casti_m256i( d3,n ) = mm256_get_64( s,  6,  7, 14, 15 );
-}
-
-// AVX 512 helper functions.
-//
-// Macro functions returning vector.
-// Abstracted typecasting, avoid temp pointers.
-// Source arguments may be any 64 or 32 byte aligned pointer as appropriate.
-
-#define mm512_put_64( s0, s1, s2, s3, s4, s5, s6, s7 ) \
-  _mm512_set_epi64( *((const uint64_t*)(s7)), *((const uint64_t*)(s6)), \
-                    *((const uint64_t*)(s5)), *((const uint64_t*)(s4)), \
-                    *((const uint64_t*)(s3)), *((const uint64_t*)(s2)), \
-                    *((const uint64_t*)(s1)), *((const uint64_t*)(s0)) ) 
-
-#define mm512_put_32( s00, s01, s02, s03, s04, s05, s06, s07, \
-                      s08, s09, s10, s11, s12, s13, s14, s15 ) \
-  _mm512_set_epi32( *((const uint32_t*)(s15)), *((const uint32_t*)(s14)), \
-                    *((const uint32_t*)(s13)), *((const uint32_t*)(s12)), \
-                    *((const uint32_t*)(s11)), *((const uint32_t*)(s10)), \
-                    *((const uint32_t*)(s09)), *((const uint32_t*)(s08)), \
-                    *((const uint32_t*)(s07)), *((const uint32_t*)(s06)), \
-                    *((const uint32_t*)(s05)), *((const uint32_t*)(s04)), \
-                    *((const uint32_t*)(s03)), *((const uint32_t*)(s02)), \
-                    *((const uint32_t*)(s01)), *((const uint32_t*)(s00)) ) 
-
-#define mm512_get_64( s, i0, i1, i2, i3, i4, i5, i6, i7 ) \
-  _mm512_set_epi64( ((const uint64_t*)(s))[i7], ((const uint64_t*)(s))[i6], \
-                    ((const uint64_t*)(s))[i5], ((const uint64_t*)(s))[i4], \
-                    ((const uint64_t*)(s))[i3], ((const uint64_t*)(s))[i2], \
-                    ((const uint64_t*)(s))[i1], ((const uint64_t*)(s))[i0] )
-
-#define mm512_get_32( s, i00, i01, i02, i03, i04, i05, i06, i07, \
-                         i08, i09, i10, i11, i12, i13, i14, i15 ) \
-  _mm512_set_epi32( ((const uint32_t*)(s))[i15], ((const uint32_t*)(s))[i14], \
-                    ((const uint32_t*)(s))[i13], ((const uint32_t*)(s))[i12], \
-                    ((const uint32_t*)(s))[i11], ((const uint32_t*)(s))[i10], \
-                    ((const uint32_t*)(s))[i09], ((const uint32_t*)(s))[i08], \
-                    ((const uint32_t*)(s))[i07], ((const uint32_t*)(s))[i06], \
-                    ((const uint32_t*)(s))[i05], ((const uint32_t*)(s))[i04], \
-                    ((const uint32_t*)(s))[i03], ((const uint32_t*)(s))[i02], \
-                    ((const uint32_t*)(s))[i01], ((const uint32_t*)(s))[i00] )
-
-// AVX512 has no blend, can be done with permute2xvar but at what cost?
-// Can also be done with shifting and mask-or'ing for 3 instructins with
-// 1 dependency. Finally it can be done with 1 _mm512_set but with 8 64 bit
-// array index calculations and 8 pointer reads.
-
-// Blend 2 vectors alternating hi & lo: { hi[n], lo[n-1], ... hi[1]. lo[0] }
-#define mm512_interleave_blend_128( hi, lo ) \
-  _mm256_permute2xvar_epi64( hi, lo, _mm512_set_epi64( \
-                            0x7, 0x6, 0x5, 0x4, 0xb, 0xa, 0x9, 0x8 )
-
-#define mm512_interleave_blend_64( hi, lo ) \
-  _mm256_permute2xvar_epi64( hi, lo, _mm512_set_epi64( \
-                            0x7, 0x6, 0xd, 0xc, 0x3, 0x2, 0x9, 0x8 )
-
-#define mm512_interleave_blend_32( hi, lo ) \
-  _mm256_permute2xvar_epi32( hi, lo, _mm512_set_epi32( \
-                  0x0f, 0x1e, 0x0d, 0x1c, 0x0b, 0x1a, 0x09, 0x18, \
-                            0x07, 0x16, 0x05, 0x14, 0x03, 0x12, 0x01, 0x10 )
-//
-
-static inline void mm512_intrlv_16x32x512( void *d, const void *s00,
-     const void *s01, const void *s02, const void *s03, const void *s04,
-     const void *s05, const void *s06, const void *s07, const void *s08,
-     const void *s09, const void *s10, const void *s11, const void *s12,
-     const void *s13, const void *s14, const void *s15 )
-{
-   casti_m512i( d, 0 ) = mm512_put_32(
-         s00,    s01,    s02,    s03,    s04,    s05,    s06,    s07,
-              s08,    s09,    s10,    s11,    s12,    s13,    s14,    s15 );
-   casti_m512i( d, 1 ) = mm512_put_32(
-         s00+ 4, s01+ 4, s02+ 4, s03+ 4, s04+ 4, s05+ 4, s06+ 4, s07+ 4,
-              s08+ 4, s09+ 4, s10+ 4, s11+ 4, s12+ 4, s13+ 4, s14+ 4, s15+ 4 );
-   casti_m512i( d, 2 ) = mm512_put_32(
-              s00+ 8, s01+ 8, s02+ 8, s03+ 8, s04+ 8, s05+ 8, s06+ 8, s07+ 8,
-              s08+ 8, s09+ 8, s10+ 8, s11+ 8, s12+ 8, s13+ 8, s14+ 8, s15+ 8 );
-   casti_m512i( d, 3 ) = mm512_put_32(
-         s00+12, s01+12, s02+12, s03+12, s04+12, s05+12, s06+12, s07+12,
-              s08+12, s09+12, s10+12, s11+12, s12+12, s13+12, s14+12, s15+12 );
-   casti_m512i( d, 4 ) = mm512_put_32(
-         s00+16, s01+16, s02+16, s03+16, s04+16, s05+16, s06+16, s07+16,
-              s08+16, s09+16, s10+16, s11+16, s12+16, s13+16, s14+16, s15+16 );
-   casti_m512i( d, 5 ) = mm512_put_32(
-         s00+20, s01+20, s02+20, s03+20, s04+20, s05+20, s06+20, s07+20,
-              s08+20, s09+20, s10+20, s11+20, s12+20, s13+20, s14+20, s15+20 );
-   casti_m512i( d, 6 ) = mm512_put_32(
-         s00+24, s01+24, s02+24, s03+24, s04+24, s05+24, s06+24, s07+24,
-              s08+24, s09+24, s10+24, s11+24, s12+24, s13+24, s14+24, s15+24 );
-   casti_m512i( d, 7 ) = mm512_put_32(
-         s00+28, s01+28, s02+28, s03+28, s04+28, s05+28, s06+28, s07+28,
-              s08+28, s09+28, s10+28, s11+28, s12+28, s13+28, s14+28, s15+28 );
-   casti_m512i( d, 8 ) = mm512_put_32(
-         s00+32, s01+28, s02+28, s03+28, s04+32, s05+28, s06+28, s07+28,
-              s08+32, s09+28, s10+28, s11+28, s12+32, s13+28, s14+28, s15+28 );
-   casti_m512i( d, 9 ) = mm512_put_32(
-         s00+36, s01+28, s02+28, s03+28, s04+36, s05+28, s06+28, s07+28,
-              s08+36, s09+28, s10+28, s11+28, s12+36, s13+28, s14+28, s15+28 );
-   casti_m512i( d,10 ) = mm512_put_32(
-         s00+40, s01+28, s02+28, s03+28, s04+40, s05+28, s06+28, s07+28,
-              s08+40, s09+28, s10+28, s11+28, s12+40, s13+28, s14+28, s15+28 );
-   casti_m512i( d,11 ) = mm512_put_32(
-         s00+44, s01+28, s02+28, s03+28, s04+44, s05+28, s06+28, s07+28,
-              s08+44, s09+28, s10+28, s11+28, s12+44, s13+28, s14+28, s15+28 );
-   casti_m512i( d,12 ) = mm512_put_32(
-         s00+48, s01+28, s02+28, s03+28, s04+48, s05+28, s06+28, s07+28,
-              s08+48, s09+28, s10+28, s11+28, s12+48, s13+28, s14+28, s15+28 );
-   casti_m512i( d,13 ) = mm512_put_32(
-         s00+52, s01+28, s02+28, s03+28, s04+52, s05+28, s06+28, s07+28,
-              s08+52, s09+28, s10+28, s11+28, s12+52, s13+28, s14+28, s15+28 );
-   casti_m512i( d,14 ) = mm512_put_32(
-         s00+56, s01+28, s02+28, s03+28, s04+56, s05+28, s06+28, s07+28,
-              s08+56, s09+28, s10+28, s11+28, s12+56, s13+28, s14+28, s15+28 );
-   casti_m512i( d,15 ) = mm512_put_32(
-         s00+60, s01+28, s02+28, s03+28, s04+60, s05+28, s06+28, s07+28,
-              s08+60, s09+28, s10+28, s11+28, s12+60, s13+28, s14+28, s15+28 );
-}
-
-static inline void mm512_intrlv_16x32x256( void *d, const void *s00,
-     const void *s01, const void *s02, const void *s03, const void *s04,
-     const void *s05, const void *s06, const void *s07, const void *s08,
-     const void *s09, const void *s10, const void *s11, const void *s12,
-     const void *s13, const void *s14, const void *s15 )
-{
-   casti_m512i( d, 0 ) = mm512_put_32(
-             s00,    s01,    s02,    s03,    s04,    s05,    s06,    s07,
-             s08,    s09,    s10,    s11,    s12,    s13,    s14,    s15 );
-   casti_m512i( d, 1 ) = mm512_put_32(
-        s00+ 4, s01+ 4, s02+ 4, s03+ 4, s04+ 4, s05+ 4, s06+ 4, s07+ 4,
-             s08+ 4, s09+ 4, s10+ 4, s11+ 4, s12+ 4, s13+ 4, s14+ 4, s15+ 4 );
-   casti_m512i( d, 2 ) = mm512_put_32(
-        s00+ 8, s01+ 8, s02+ 8, s03+ 8, s04+ 8, s05+ 8, s06+ 8, s07+ 8,
-             s08+ 8, s09+ 8, s10+ 8, s11+ 8, s12+ 8, s13+ 8, s14+ 8, s15+ 8 );
-   casti_m512i( d, 3 ) = mm512_put_32(
-        s00+12, s01+12, s02+12, s03+12, s04+12, s05+12, s06+12, s07+12,
-             s08+12, s09+12, s10+12, s11+12, s12+12, s13+12, s14+12, s15+12 );
-   casti_m512i( d, 4 ) = mm512_put_32(
-             s00+16, s01+16, s02+16, s03+16, s04+16, s05+16, s06+16, s07+16,
-             s08+16, s09+16, s10+16, s11+16, s12+16, s13+16, s14+16, s15+16 );
-   casti_m512i( d, 5 ) = mm512_put_32(
-        s00+20, s01+20, s02+20, s03+20, s04+20, s05+20, s06+20, s07+20,
-             s08+20, s09+20, s10+20, s11+20, s12+20, s13+20, s14+20, s15+20 );
-   casti_m512i( d, 6 ) = mm512_put_32(
-        s00+24, s01+24, s02+24, s03+24, s04+24, s05+24, s06+24, s07+24,
-             s08+24, s09+24, s10+24, s11+24, s12+24, s13+24, s14+24, s15+24 );
-   casti_m512i( d, 7 ) = mm512_put_32(
-        s00+28, s01+28, s02+28, s03+28, s04+28, s05+28, s06+28, s07+28,
-             s08+28, s09+28, s10+28, s11+28, s12+28, s13+28, s14+28, s15+28 );
-}
-
-// Last 16 bytes of input
-static inline void mm512_intrlv_16x32x128( void *d, const void *s00,
-     const void *s01, const void *s02, const void *s03, const void *s04,
-     const void *s05, const void *s06, const void *s07, const void *s08,
-     const void *s09, const void *s10, const void *s11, const void *s12,
-     const void *s13, const void *s14, const void *s15 )
-{
-   casti_m512i( d, 0 ) = mm512_put_32(
-        s00,    s01,    s02,    s03,    s04,    s05,    s06,    s07,
-        s08,    s09,    s10,    s11,    s12,    s13,    s14,    s15 );
-   casti_m512i( d, 1 ) = mm512_put_32(
-        s00+ 4, s01+ 4, s02+ 4, s03+ 4, s04+ 4, s05+ 4, s06+ 4, s07+ 4,
-             s08+ 4, s09+ 4, s10+ 4, s11+ 4, s12+ 4, s13+ 4, s14+ 4, s15+ 4 );
-   casti_m512i( d, 2 ) = mm512_put_32(
-        s00+ 8, s01+ 8, s02+ 8, s03+ 8, s04+ 8, s05+ 8, s06+ 8, s07+ 8,
-             s08+ 8, s09+ 8, s10+ 8, s11+ 8, s12+ 8, s13+ 8, s14+ 8, s15+ 8 );
-   casti_m512i( d, 3 ) = mm512_put_32(
-        s00+12, s01+12, s02+12, s03+12, s04+12, s05+12, s06+12, s07+12,
-             s08+12, s09+12, s10+12, s11+12, s12+12, s13+12, s14+12, s15+12 );
-}
-
-// can be called directly for 64 byte hash.
-static inline void mm512_dintrlv_16x32x512( void *d00, void *d01,
-                void *d02, void *d03, void *d04, void *d05, void *d06,
-                void *d07, void *d08, void *d09, void *d10, void *d11,
-                void *d12, void *d13, void *d14, void *d15, const int n,
-      const void *s )
-{
-   casti_m512i(d00,n) = mm512_get_32( s,  0, 16, 32, 48, 64, 80, 96,112,
-                              128,144,160,176,192,208,224,240 );
-   casti_m512i(d01,n) = mm512_get_32( s,  1, 17, 33, 49, 65, 81, 97,113,
-                              129,145,161,177,193,209,225,241 );
-   casti_m512i(d02,n) = mm512_get_32( s,  2, 18, 34, 50, 66, 82, 98,114,
-                    130,146,162,178,194,210,226,242 );
-   casti_m512i(d03,n) = mm512_get_32( s,  3, 19, 35, 51, 67, 83, 99,115,
-                                        131,147,163,179,195,211,227,243 );
-   casti_m512i(d04,n) = mm512_get_32( s,  4, 20, 36, 52, 68, 84,100,116,
-                              132,148,164,180,196,212,228,244 );
-   casti_m512i(d05,n) = mm512_get_32( s,  5, 21, 37, 53, 69, 85,101,117,
-                                        133,149,165,181,197,213,229,245 );
-   casti_m512i(d06,n) = mm512_get_32( s,  6, 22, 38, 54, 70, 86,102,118,
-                                        134,150,166,182,198,214,230,246 );
-   casti_m512i(d07,n) = mm512_get_32( s,  7, 23, 39, 55, 71, 87,103,119,
-                              135,151,167,183,199,215,231,247 );
-   casti_m512i(d08,n) = mm512_get_32( s,  8, 24, 40, 56, 72, 88,104,120,
-                              136,152,168,184,200,216,232,248 );
-   casti_m512i(d09,n) = mm512_get_32( s,  9, 25, 41, 57, 73, 89,105,121,
-                              137,153,169,185,201,217,233,249 );
-   casti_m512i(d10,n) = mm512_get_32( s, 10, 26, 42, 58, 74, 90,106,122,
-                              138,154,170,186,202,218,234,250 );
-   casti_m512i(d11,n) = mm512_get_32( s, 11, 27, 43, 59, 75, 91,107,123,
-                              139,155,171,187,203,219,235,251 );
-   casti_m512i(d12,n) = mm512_get_32( s, 12, 28, 44, 60, 76, 92,108,124,
-                              140,156,172,188,204,220,236,252 );
-   casti_m512i(d13,n) = mm512_get_32( s, 13, 29, 45, 61, 77, 93,109,125,
-                              141,157,173,189,205,221,237,253 );
-   casti_m512i(d14,n) = mm512_get_32( s, 14, 30, 46, 62, 78, 94,110,126,
-                                   142,158,174,190,206,222,238,254 );
-   casti_m512i(d15,n) = mm512_get_32( s, 15, 31, 47, 63, 79, 95,111,127,
-                                    143,159,175,191,207,223,239,255 );
-}
-
-static inline void mm512_intrlv_8x64x512( void *d, const void *s0,
-                   const void *s1, const void *s2, const void *s3,
-                   const void *s4, const void *s5, const void *s6,
-                   const void *s7 )
-{
-  casti_m512i( d,0 ) = mm512_put_64( s0,    s1,    s2,    s3,
-                                     s4,    s5,    s6,    s7 );
-  casti_m512i( d,1 ) = mm512_put_64( s0+ 8, s1+ 8, s2+ 8, s3+ 8,
-                                     s4+ 8, s5+ 8, s6+ 8, s7+ 8 );
-  casti_m512i( d,2 ) = mm512_put_64( s0+16, s1+16, s2+16, s3+16,
-                                     s4+16, s5+16, s6+16, s7+16 );
-  casti_m512i( d,3 ) = mm512_put_64( s0+24, s1+24, s2+24, s3+24,
-                                     s4+24, s5+24, s6+24, s7+24 );
-  casti_m512i( d,4 ) = mm512_put_64( s0+32, s1+32, s2+32, s3+32,
-                                     s4+32, s5+32, s6+32, s7+32 );
-  casti_m512i( d,5 ) = mm512_put_64( s0+40, s1+40, s2+40, s3+40,
-                                     s4+40, s5+40, s6+40, s7+40 );
-  casti_m512i( d,6 ) = mm512_put_64( s0+48, s1+48, s2+48, s3+48,
-                                     s4+48, s5+48, s6+48, s7+48 );
-  casti_m512i( d,7 ) = mm512_put_64( s0+56, s1+56, s2+56, s3+56,
-                                     s4+56, s5+56, s6+56, s7+56 );
-}
-
-static inline void mm512_intrlv_8x64x256( void *d, const void *s0,
-                   const void *s1, const void *s2, const void *s3,
-                   const void *s4, const void *s5, const void *s6,
-                   const void *s7 )
-{
-  casti_m512i( d,0 ) = mm512_put_64( s0,    s1,    s2,    s3,
-                                     s4,    s5,    s6,    s7 );
-  casti_m512i( d,1 ) = mm512_put_64( s0+ 8, s1+ 8, s2+ 8, s3+ 8,
-                                     s4+ 8, s5+ 8, s6+ 8, s7+ 8 );
-  casti_m512i( d,2 ) = mm512_put_64( s0+16, s1+16, s2+16, s3+16,
-                                     s4+16, s5+16, s6+16, s7+16 );
-  casti_m512i( d,3 ) = mm512_put_64( s0+24, s1+24, s2+24, s3+24,
-                                     s4+24, s5+24, s6+24, s7+24 );
-}
-
-
-// 8 lanes of 512 bits using 64 bit interleaving (typical intermediate hash) 
-static inline void mm512_dintrlv_8x64x512( void *d0, void *d1, void *d2,
-                            void *d3, void *d4, void *d5, void *d6, void *d7,
-             const int n, const void *s )
-{
-   casti_m512i( d0,n ) = mm512_get_64( s, 0,  8, 16, 24, 32, 40, 48, 56 );
-   casti_m512i( d1,n ) = mm512_get_64( s, 1,  9, 17, 25, 33, 41, 49, 57 );
-   casti_m512i( d2,n ) = mm512_get_64( s, 2, 10, 18, 26, 34, 42, 50, 58 );
-   casti_m512i( d3,n ) = mm512_get_64( s, 3, 11, 19, 27, 35, 43, 51, 59 );
-   casti_m512i( d4,n ) = mm512_get_64( s, 4, 12, 20, 28, 36, 44, 52, 60 );
-   casti_m512i( d5,n ) = mm512_get_64( s, 5, 13, 21, 29, 37, 45, 53, 61 );
-   casti_m512i( d6,n ) = mm512_get_64( s, 6, 14, 22, 30, 38, 46, 54, 62 );
-   casti_m512i( d7,n ) = mm512_get_64( s, 7, 15, 23, 31, 39, 47, 55, 63 );
-}
-
-static inline void mm512_dintrlv_4x128x512( void *d0, void *d1, void *d2,
-                                void *d3, const int n, const void *s )
-{
-   casti_m512i( d0,n ) = mm512_get_64( s, 0, 1,  8,  9, 16, 17, 24, 25 );
-   casti_m512i( d1,n ) = mm512_get_64( s, 2, 3, 10, 11, 18, 19, 16, 27 );
-   casti_m512i( d2,n ) = mm512_get_64( s, 4, 5, 12, 13, 20, 21, 28, 29 );
-   casti_m512i( d3,n ) = mm512_get_64( s, 6, 7, 14, 15, 22, 23, 30, 31 );
-}
-
-// AVX-512 user facing functions.
-
-static inline void mm512_intrlv_16x32( void *d, const void *s00,
-    const void *s01, const void *s02, const void *s03, const void *s04,
-    const void *s05, const void *s06, const void *s07, const void *s08,
-    const void *s09, const void *s10, const void *s11, const void *s12,
-    const void *s13, const void *s14, const void *s15, int bit_len )
-{
-   if ( bit_len <= 256 )
-   {
-      mm512_intrlv_16x32x256( d, s00, s01, s02, s03, s04, s05, s06, s07,
-                                     s08, s09, s10, s11, s12, s13, s14, s15 );
-      return;
-   }
-   mm512_intrlv_16x32x512( d, s00, s01, s02, s03, s04, s05, s06, s07,
-                        s08, s09, s10, s11, s12, s13, s14, s15 );
-   if ( bit_len <= 512 ) return;
-   if ( bit_len <= 640 )
-   {
-
-      mm512_intrlv_16x32x128( d+1024, s00+64, s01+64, s02+64, s03+64,
-                s04+64, s05+64, s06+64, s07+64, s08+64, s09+64,
-           s10+64, s11+64, s12+64, s13+64, s14+64, s15+64 );
-      return;
-   }
-   mm512_intrlv_16x32x512( d+1024, s00+64, s01+64, s02+64, s03+64,
-                       s04+64, s05+64, s06+64, s07+64, s08+64, s09+64,
-                       s10+64, s11+64, s12+64, s13+64, s14+64, s15+64 );
-   // bit_len == 1024
-}
-
-// sub-functions can be called directly for 32 & 64 byte hash.
-static inline void mm512_dintrlv_16x32( void *d00, void *d01, void *d02,
-            void *d03, void *d04, void *d05, void *d06, void *d07, void *d08,
-            void *d09, void *d10, void *d11, void *d12, void *d13, void *d14,
-            void *d15, const void *src, const int bit_len )
-{
-   if ( bit_len <= 256 )
-   {
-      mm256_dintrlv_16x32x256( d00, d01, d02, d03, d04, d05, d06, d07,
-                                    d08, d09, d10, d11, d12, d13, d14, d15,
-                                    0,src );
-      return;
-   }
-   mm512_dintrlv_16x32x512( d00, d01, d02, d03, d04, d05, d06, d07,
-                                 d08, d09, d10, d11, d12, d13, d14, d15,
-                                 0, src );
-   if ( bit_len <= 512 ) return;
-   if ( bit_len <= 640 )
-   {
-      // short block, final 16 bytes of input data.
-      mm128_dintrlv_16x32x128( d00, d01, d02, d03, d04, d05, d06, d07,
-                          d08, d09, d10, d11, d12, d13, d14, d15,
-                1, src+1024 );
-      return;
-   }
-   // bit_len == 1024
-   mm512_dintrlv_16x32x512( d00, d01, d02, d03, d04, d05, d06, d07,
-                       d08, d09, d10, d11, d12, d13, d14, d15,
-             1, src+1024 );
-}
-
-static inline void mm512_extr_lane_16x32( void *dst, const void *src,
-                                            const int lane, const int bit_len )
-{
-  if ( bit_len <= 256 )
-  {
-     cast_m256i( dst ) = mm256_get_32( src, lane, lane+16, lane+32, lane+48,
-                               lane+64, lane+80, lane+96, lane+112 );
-     return;
-  }
-  cast_m512i( dst ) = mm512_get_32( src, lane, lane+ 16, lane+ 32, lane+ 48,
-                 lane+ 64, lane+ 80, lane+ 96, lane+112, lane+128, lane+144,
-            lane+160, lane+176, lane+192, lane+208, lane+224, lane+248 );
-}
-
-//
-
-static inline void mm512_intrlv_8x64( void *d, const void *s0,
-                   const void *s1, const void *s2, const void *s3,
-                   const void *s4, const void *s5, const void *s6,
-                   const void *s7, int bit_len )
-{
-   if ( bit_len <= 256 )
-   {
-      mm512_intrlv_8x64x256( d, s0, s1, s2, s3, s4, s5, s6, s7 );
-      return;
-   }
-   mm512_intrlv_8x64x512( d, s0, s1, s2, s3, s4, s5, s6, s7 );
-   if ( bit_len <= 512 ) return;
-   if ( bit_len <= 640 )
-   {
-      casti_m512i( d, 8 ) = mm512_put_64( s7+64, s6+64, s5+64, s4+64,
-                                          s3+64, s2+64, s1+64, s0+64 );
-      casti_m512i( d, 9 ) = mm512_put_64( s7+72, s6+72, s5+72, s4+72,
-                                          s3+72, s2+72, s1+72, s0+72 );
-      return;
-   }
-   // bitlen == 1024
-   mm512_intrlv_8x64x512( d+512, s0+64, s1+64, s2+64, s3+64,
-                           s4+64, s5+64, s6+64, s7+64 );
-}
-
-
-static inline void mm512_dintrlv_8x64( void *d0, void *d1, void *d2,
-                        void *d3, void *d4, void *d5, void *d6, void *d7,
-                        const void *s, const int bit_len )
-{
-   if ( bit_len <= 256 )
-   {
-      mm256_dintrlv_8x64x256( d0, d1, d2, d3, d4, d5, d6, d7, 0, s );
-      return;
-   }
-   mm512_dintrlv_8x64x512( d0, d1, d2, d3, d4, d5, d6, d7, 0, s );
-   if ( bit_len <= 512 ) return;
-   if ( bit_len <= 640 )
-   {
-      // short block, final 16 bytes of input data.
-      mm128_dintrlv_8x64x128( d0, d1, d2, d3, d4, d5, d6, d7, 1, s+512 );
-      return;
-   }
-   // bit_len == 1024
-   mm512_dintrlv_8x64x512( d0, d1, d2, d3, d4, d5, d6, d7, 1, s+512 );
-}
-
-// Extract one lane from 64 bit interleaved data
-static inline void mm512_extr_lane_8x64( void *d, const void *s,
-                                            const int lane, const int bit_len )
-{
-  if ( bit_len <= 256 )
-  {
-     cast_m256i( d ) = mm256_get_64( s, lane, lane+8, lane+16, lane+24 );
-     return;
-  }
-  // else bit_len == 512
-  cast_m512i( d ) = mm512_get_64( s, lane   , lane+ 8, lane+16, lane+24,
-                               lane+32, lane+40, lane+48, lane+56 );
-}
-
-//
-
-static inline void mm512_intrlv_4x128( void *d, const void *s0,
-            const void *s1, const void *s2, const void *s3, const int bit_len )
-{
-  casti_m512i( d, 0 ) = mm512_put_64( s0,    s0+8,  s1,    s1+8,
-                                      s2,    s2+8,  s3,    s3+8 );
-  casti_m512i( d, 1 ) = mm512_put_64( s0+16, s0+24, s1+16, s1+24,
-                                      s2+16, s2+24, s3+16, s3+24 );
-  if ( bit_len <= 256 ) return;
-
-  casti_m512i( d, 2 ) = mm512_put_64( s0+32, s0+40, s1+32, s1+40,
-                                      s2+32, s2+40, s3+32, s3+40 );
-  casti_m512i( d, 3 ) = mm512_put_64( s0+48, s0+56, s1+48, s1+56,
-                                      s2+48, s2+56, s3+48, s3+56 );
-  if ( bit_len <= 512 ) return;
-
-  casti_m512i( d, 4 ) = mm512_put_64( s0+64, s0+72, s1+64, s1+72,
-                                      s2+64, s2+72, s3+64, s3+72 );
-  if ( bit_len <= 640 ) return;
-
-  casti_m512i( d, 5 ) = mm512_put_64( s0+ 80, s0+ 88, s1+ 80, s1+ 88,
-                                      s2+ 80, s2+ 88, s3+ 80, s3+ 88 );
-  casti_m512i( d, 6 ) = mm512_put_64( s0+ 96, s0+104, s1+ 96, s1+104,
-                                      s2+ 96, s2+104, s3+ 96, s3+104 );
-  casti_m512i( d, 7 ) = mm512_put_64( s0+112, s0+120, s1+112, s1+120,
-                                      s2+112, s2+120, s3+112, s3+120 );
-   // bit_len == 1024
-}
-
-static inline void mm512_dintrlv_4x128( void *d0, void *d1, void *d2,
-              void *d3, const void *s, const int bit_len )
-{
-   if ( bit_len <= 256 )
-   {
-      mm256_dintrlv_4x128x256( d0, d1, d2, d3, 0, s );
-      return;
-   }
-   mm512_dintrlv_4x128x512( d0, d1, d2, d3, 0, s );
-   if ( bit_len <= 512 ) return;
-   if ( bit_len <= 640 )
-   {
-      mm128_dintrlv_4x128x128( d0, d1, d2, d3, 1, s+256 );
-      return;
-   }
-   // bit_len == 1024
-   mm512_dintrlv_4x128x512( d0, d1, d2, d3, 1, s+256 );
-}
-
-// input one 8x64 buffer and return 2*4*128
-static inline void mm512_rintrlv_8x64_4x128( void *dst0, void *dst1,
-                                              const void *src, int  bit_len )
-{
-   __m512i* d0 = (__m512i*)dst0;
-   __m512i* d1 = (__m512i*)dst1;
-   uint64_t *s = (uint64_t*)src;
-
-   d0[0] = _mm512_set_epi64( s[ 11], s[  3], s[ 10], s[  2],
-                             s[  9], s[  1], s[  8], s[  0] );
-   d0[1] = _mm512_set_epi64( s[ 27], s[ 19], s[ 26], s[ 18],
-                   s[ 25], s[ 17], s[ 24], s[ 16] );
-   d0[2] = _mm512_set_epi64( s[ 15], s[  7], s[ 14], s[  6],
-                             s[ 13], s[  5], s[ 12], s[  4] );
-   d0[3] = _mm512_set_epi64( s[ 31], s[ 23], s[ 30], s[ 22],
-                             s[ 29], s[ 21], s[ 28], s[ 20] );
-   d1[0] = _mm512_set_epi64( s[ 43], s[ 35], s[ 42], s[ 34],
-                             s[ 41], s[ 33], s[ 40], s[ 32] );
-   d1[1] = _mm512_set_epi64( s[ 59], s[ 51], s[ 58], s[ 50],
-                             s[ 57], s[ 49], s[ 56], s[ 48] );
-   d1[2] = _mm512_set_epi64( s[ 47], s[ 39], s[ 46], s[ 38],
-                             s[ 45], s[ 37], s[ 44], s[ 36] );
-   d1[3] = _mm512_set_epi64( s[ 63], s[ 55], s[ 62], s[ 54],
-                              s[ 61], s[ 53], s[ 60], s[ 52] );
-
-   if ( bit_len <= 512 ) return;
-
-   d0[4] = _mm512_set_epi64( s[ 75], s[ 67], s[ 74], s[ 66],
-                             s[ 73], s[ 65], s[ 72], s[ 64] );
-   d0[5] = _mm512_set_epi64( s[ 91], s[ 83], s[ 90], s[ 82],
-                             s[ 89], s[ 81], s[ 88], s[ 80] );
-   d0[6] = _mm512_set_epi64( s[ 79], s[ 71], s[ 78], s[ 70],
-                             s[ 77], s[ 69], s[ 76], s[ 68] );
-   d0[7] = _mm512_set_epi64( s[ 95], s[ 87], s[ 94], s[ 86],
-                             s[ 93], s[ 85], s[ 92], s[ 84] );
-   d1[4] = _mm512_set_epi64( s[107], s[ 99], s[106], s[ 98],
-                             s[105], s[ 97], s[104], s[ 96] );
-   d1[5] = _mm512_set_epi64( s[123], s[115], s[122], s[114],
-                             s[121], s[113], s[120], s[112] );
-   d1[6] = _mm512_set_epi64( s[111], s[103], s[110], s[102],
-                             s[109], s[101], s[108], s[100] );
-   d1[7] = _mm512_set_epi64( s[127], s[119], s[126], s[118],
-                             s[125], s[117], s[124], s[116] );
-
-}
-
-// input 2 4x128  return 8x64
-static inline void mm512_rintrlv_4x128_8x64( void *dst, const void *src0,
-                                              const void *src1, int  bit_len )
-{
-   __m512i* d = (__m512i*)dst;
-   uint64_t *s0 = (uint64_t*)src0;
-   uint64_t *s1 = (uint64_t*)src1;
-
-   d[0] = _mm512_set_epi64( s1[ 6], s1[ 4], s1[ 2], s1[ 0],
-                            s0[ 6], s0[ 4], s0[ 2], s0[ 0] );
-   d[1] = _mm512_set_epi64( s1[ 7], s1[ 5], s1[ 3], s1[ 1],
-                            s0[ 7], s0[ 5], s0[ 3], s0[ 1] );
-   d[2] = _mm512_set_epi64( s1[14], s1[12], s1[10], s1[ 8],
-                            s0[14], s0[12], s0[10], s0[ 8] );
-   d[3] = _mm512_set_epi64( s1[15], s1[13], s1[11], s1[ 9],
-                            s0[15], s0[13], s0[11], s0[ 9] );
-   d[4] = _mm512_set_epi64( s1[22], s1[20], s1[18], s1[16],
-                            s0[22], s0[20], s0[18], s0[16] );
-   d[5] = _mm512_set_epi64( s1[23], s1[21], s1[19], s1[17],
-                            s0[24], s0[21], s0[19], s0[17] );
-   d[6] = _mm512_set_epi64( s1[22], s1[28], s1[26], s1[24],
-                            s0[22], s0[28], s0[26], s0[24] );
-   d[7] = _mm512_set_epi64( s1[31], s1[29], s1[27], s1[25],
-                            s0[31], s0[29], s0[27], s0[25] );
-
-   if ( bit_len <= 512 ) return;
-
-   d[0] = _mm512_set_epi64( s1[38], s1[36], s1[34], s1[32],
-                            s0[38], s0[36], s0[34], s0[32] );
-   d[1] = _mm512_set_epi64( s1[39], s1[37], s1[35], s1[33],
-                            s0[39], s0[37], s0[35], s0[33] );
-   d[2] = _mm512_set_epi64( s1[46], s1[44], s1[42], s1[40],
-                            s0[46], s0[44], s0[42], s0[40] );
-   d[3] = _mm512_set_epi64( s1[47], s1[45], s1[43], s1[41],
-                            s0[47], s0[45], s0[43], s0[41] );
-   d[4] = _mm512_set_epi64( s1[54], s1[52], s1[50], s1[48],
-                            s0[54], s0[52], s0[50], s0[48] );
-   d[5] = _mm512_set_epi64( s1[55], s1[53], s1[51], s1[49],
-                            s0[55], s0[53], s0[51], s0[49] );
-
-   d[6] = _mm512_set_epi64( s1[62], s1[60], s1[58], s1[56],
-                            s0[62], s0[60], s0[58], s0[56] );
-   d[7] = _mm512_set_epi64( s1[63], s1[61], s1[59], s1[57],
-                            s0[63], s0[61], s0[59], s0[57] );
-
-}
-
-static inline void mm512_extr_lane_4x128( void *d, const void *s,
-                                            const int lane, const int bit_len )
-{
-  int l = lane<<1;
-  if ( bit_len <= 256 )
-  {
-     cast_m256i( d ) = mm256_get_64( s, l, l+1, l+8, l+9 );
-     return;
-  }
-  // else bit_len == 512
-  cast_m512i( d ) = mm512_get_64( s, l   , l+ 1, l+ 8, l+ 9,
-                                l+16, l+17, l+24, l+25 );
-}
-
-#endif // AVX512
-#endif // INTRLV_AVX512_H__

simd-utils/intrlv-mmx.h

@@ -1,126 +0,0 @@
-#if !defined(INTRLV_MMX_H__)
-#define INTRLV_MMX_H__ 1
-
-#if defined(__MMX__)
-
-//////////////////////////////////////////////////////
-//
-//          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_intrlv_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_dintrlv_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_extr_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
-}
-
-
-
-#endif // MMX
-#endif // INTRLV_MMX_H__

simd-utils/intrlv-selector.h

@@ -1,77 +0,0 @@
-#if !defined(INTRLV_SELECTOR_H__)
-#define INTRLV_SELECTOR_H__
-
-//////////////////////////////////////////////////////////////
-//
-//  Generic interface for interleaving data for parallel processing.
-//
-//  Best tech is chosen atomatically.
-
-/*
-#if defined(__AVX512F__)
-
-#define intrlv_4x128      mm512_intrlv_4x128
-#define intrlv_4x128      mm512_intrlv_4x128
-
-#define intrlv_8x64       mm512_intrlv_8x64
-#define dintrlv_8x64      mm512_dintrlv_8x64
-#define extr_lane_8x64    mm512_extr_lane_8x64
-
-#define intrlv_16x32      mm512_intrlv_16x32
-#define dintrlv_16x32     mm512_dintrlv_16x32
-#define extr_lane_16x32    mm512_extr_lane_16x32
-
-#define intrlv_2x128      mm512_intrlv_2x128
-#define dintrlv_2x128     mm512_dintrlv_2x128
-
-#define intrlv_4x64       mm512_intrlv_4x64
-#define dintrlv_4x64      mm512_dintrlv_4x64
-#define extr_lane_4x64    mm512_extr_lane_4x64
-
-#define intrlv_8x32       mm512_intrlv_8x32
-#define dintrlv_8x32      mm512_dintrlv_8x32
-#define extr_lane_8x32    mm512_extr_lane_8x32
-
-#elif defined(__AVX__)
-*/
-#if defined(__AVX__)
-
-#define intrlv_2x128      mm256_intrlv_2x128
-#define dintrlv_2x128     mm256_dintrlv_2x128
-
-#define intrlv_4x64       mm256_intrlv_4x64
-#define dintrlv_4x64      mm256_dintrlv_4x64
-#define extr_lane_4x64    mm256_extr_lane_4x64
-
-#define intrlv_8x32       mm256_intrlv_8x32
-#define dintrlv_8x32      mm256_dintrlv_8x32
-#define extr_lane_8x32    mm256_extr_lane_8x32
-
-#define intrlv_4x32       mm256_intrlv_4x32
-#define dintrlv_4x32      mm256_dintrlv_4x32
-#define extr_lane_4x32    mm256_extr_lane_4x32
-
-#else
-
-#define intrlv_2x128      mm128_intrlv_2x128
-#define dintrlv_2x128     mm128_dintrlv_2x128
-
-#define intrlv_4x64       mm128_intrlv_4x64
-#define dintrlv_4x64      mm128_dintrlv_4x64
-#define extr_lane_4x64    mm128_extr_lane_4x64
-
-#define intrlv_8x32       mm128_intrlv_8x32
-#define dintrlv_8x32      mm128_dintrlv_8x32
-#define extr_lane_8x32    mm128_extr_lane_8x32
-
-#define intrlv_2x64       mm128_intrlv_2x64
-#define dintrlv_2x64      mm128_dintrlv_2x64
-#define extr_lane_2x64    mm128_extr_lane_2x64
-
-#define intrlv_4x32       mm128_intrlv_4x32
-#define dintrlv_4x32      mm128_dintrlv_4x32
-#define extr_lane_4x32    mm128_extr_lane_4x32
-
-#endif
-
-#endif  // INTRLV_SELECTOR_H__

simd-utils/intrlv-sse2.h

@@ -1,192 +0,0 @@
-#if !defined(INTRLV_SSE2_H__)
-#define INTRLV_SSE2_H__ 1
-
-// Don't call __mm_extract_epi32 directly, it needs SSE4.1.
-// Use mm128_extr_32 wrapper instead, it has both SSE4.1 & SSE2 covered.
-
-#if  defined(__SSE2__)
-
-///////////////////////////////////////////////////////////////
-//
-//           SSE2 128 bit vectors
-
-
-// Macros to abstract typecasting
-
-// Interleave lanes 
-#define mm128_put_64( s0, s1) \
-  _mm_set_epi64x( *((const uint64_t*)(s1)), *((const uint64_t*)(s0)) )
-
-#define mm128_put_32( s0, s1, s2, s3 ) \
-  _mm_set_epi32( *((const uint32_t*)(s3)), *((const uint32_t*)(s2)), \
-                 *((const uint32_t*)(s1)), *((const uint32_t*)(s0)) )
-
-// Deinterleave lanes
-#define mm128_get_64( s, i0, i1 ) \
-  _mm_set_epi64x( ((const uint64_t*)(s))[i1], ((const uint64_t*)(s))[i0] )
-
-#define mm128_get_32( s, i0, i1, i2, i3 ) \
-  _mm_set_epi32( ((const uint32_t*)(s))[i3], ((const uint32_t*)(s))[i2], \
-                 ((const uint32_t*)(s))[i1], ((const uint32_t*)(s))[i0] )
-
-// blend 2 vectors while interleaving: { hi[n], lo[n-1], ... hi[1], lo[0] }
-#define mm128_intrlv_blend_64( hi, lo ) \
-                _mm256_blend_epi16( hi, lo, 0x0f )
-#define mm128_intrlv_blend_32( hi, lo ) \
-                _mm6_blend_epi16( hi, lo, 0x33 )
-
-// 1 sse2 block, 16 x 16 bytes
-
-#define mm128_intrlv_4x32_128( d, s0, s1, s2, s3 )\
-do { \
-   casti_m128i( d,0 ) = _mm_set_epi32( \
-                            mm128_extr_32( s3, 0 ), mm128_extr_32( s2, 0 ), \
-                            mm128_extr_32( s1, 0 ), mm128_extr_32( s0, 0 ) ); \
-   casti_m128i( d,1 ) = _mm_set_epi32( \
-                            mm128_extr_32( s3, 1 ), mm128_extr_32( s2, 1 ), \
-                            mm128_extr_32( s1, 1 ), mm128_extr_32( s0, 1 ) ); \
-   casti_m128i( d,2 ) = _mm_set_epi32( \
-                            mm128_extr_32( s3, 2 ), mm128_extr_32( s2, 2 ), \
-                            mm128_extr_32( s1, 2 ), mm128_extr_32( s0, 2 ) ); \
-   casti_m128i( d,3 ) = _mm_set_epi32( \
-                            mm128_extr_32( s3, 3 ), mm128_extr_32( s2, 3 ), \
-                            mm128_extr_32( s1, 3 ), mm128_extr_32( s0, 3 ) ); \
-} while(0)
-
-static inline void mm128_dintrlv_4x32_128( void *d0, void *d1, void *d2,
-                                           void *d3, const void *src )
-{
-   __m128i s0 = *(__m128i*) src;
-   __m128i s1 = *(__m128i*)(src+16);
-   __m128i s2 = *(__m128i*)(src+32);
-   __m128i s3 = *(__m128i*)(src+48);
-
-   *(__m128i*)d0 = _mm_set_epi32(
-                     mm128_extr_32( s3,0 ), mm128_extr_32( s2,0 ),
-                     mm128_extr_32( s1,0 ), mm128_extr_32( s0,0 ) );
-   *(__m128i*)d1 = _mm_set_epi32(
-                     mm128_extr_32( s3,1 ), mm128_extr_32( s2,1 ),
-                     mm128_extr_32( s1,1 ), mm128_extr_32( s0,1 ) );
-   *(__m128i*)d2 = _mm_set_epi32(
-                     mm128_extr_32( s3,2 ), mm128_extr_32( s2,2 ),
-                     mm128_extr_32( s1,2 ), mm128_extr_32( s0,2 ) );
-   *(__m128i*)d3 = _mm_set_epi32(
-                     mm128_extr_32( s3,3 ), mm128_extr_32( s2,3 ),
-                     mm128_extr_32( s1,3 ), mm128_extr_32( s0,3 ) );
-}
-
-static inline void mm128_intrlv_2x64x128( void *d, const void *s0,
-                       const void *s1 )
-{
-  casti_m128i( d,0 ) = mm128_put_64( s0,    s1    );
-  casti_m128i( d,1 ) = mm128_put_64( s0+ 8, s1+ 8 );
-  casti_m128i( d,2 ) = mm128_put_64( s0+16, s1+16 );
-  casti_m128i( d,3 ) = mm128_put_64( s0+24, s1+24 );
-}
-
-#define mm128_bswap_intrlv_4x32_128( d, src ) \
-do { \
-  __m128i ss = mm128_bswap_32( src );\
-  casti_m128i( d,0 ) = _mm_set1_epi32( mm128_extr_32( ss, 0 ) ); \
-  casti_m128i( d,1 ) = _mm_set1_epi32( mm128_extr_32( ss, 1 ) ); \
-  casti_m128i( d,2 ) = _mm_set1_epi32( mm128_extr_32( ss, 2 ) ); \
-  casti_m128i( d,3 ) = _mm_set1_epi32( mm128_extr_32( ss, 3 ) ); \
-} while(0)
-
-
-//
-// User functions.
-
-// interleave 4 arrays of 32 bit elements for 128 bit processing
-// bit_len must be 256, 512 or 640 bits.
-static inline void mm128_intrlv_4x32( void *d, const void *s0,
-               const void *s1, const void *s2, const void *s3, int bit_len )
-{
-   mm128_intrlv_4x32_128( d    , casti_m128i(s0,0), casti_m128i(s1,0),
-                                 casti_m128i(s2,0), casti_m128i(s3,0) );
-   mm128_intrlv_4x32_128( d+ 64, casti_m128i(s0,1), casti_m128i(s1,1),   
-                                 casti_m128i(s2,1), casti_m128i(s3,1) );   
-   if ( bit_len <= 256 ) return;
-   mm128_intrlv_4x32_128( d+128, casti_m128i(s0,2), casti_m128i(s1,2),
-                                 casti_m128i(s2,2), casti_m128i(s3,2) );
-   mm128_intrlv_4x32_128( d+192, casti_m128i(s0,3), casti_m128i(s1,3),
-                                 casti_m128i(s2,3), casti_m128i(s3,3) );
-   if ( bit_len <= 512 ) return;
-   mm128_intrlv_4x32_128( d+256, casti_m128i(s0,4), casti_m128i(s1,4),
-                                 casti_m128i(s2,4), casti_m128i(s3,4) );
-   if ( bit_len <= 640 ) return;
-   mm128_intrlv_4x32_128( d+320, casti_m128i(s0,5), casti_m128i(s1,5),
-                                 casti_m128i(s2,5), casti_m128i(s3,5) );
-   mm128_intrlv_4x32_128( d+384, casti_m128i(s0,6), casti_m128i(s1,6),
-                                 casti_m128i(s2,6), casti_m128i(s3,6) );
-   mm128_intrlv_4x32_128( d+448, casti_m128i(s0,7), casti_m128i(s1,7),
-                                 casti_m128i(s2,7), casti_m128i(s3,7) );
-   // bit_len == 1024
-}
-
-// Still used by decred due to odd data size: 180 bytes
-// bit_len must be multiple of 32
-static inline void mm128_intrlv_4x32x( void *dst, void *src0, void  *src1,
-                                        void *src2, void *src3, int bit_len )
-{
-   uint32_t *d  = (uint32_t*)dst;
-   uint32_t *s0 = (uint32_t*)src0;
-   uint32_t *s1 = (uint32_t*)src1;
-   uint32_t *s2 = (uint32_t*)src2;
-   uint32_t *s3 = (uint32_t*)src3;
-
-   for ( int i = 0; i < bit_len >> 5; i++, d += 4 )
-   {
-      *d     = *(s0+i);
-      *(d+1) = *(s1+i);
-      *(d+2) = *(s2+i);
-      *(d+3) = *(s3+i);
-   }
-}
-
-static inline void mm128_dintrlv_4x32( void *d0, void *d1, void *d2,
-                                       void *d3, const void *s, int bit_len )
-{
-   mm128_dintrlv_4x32_128( d0    , d1    , d2    , d3    , s     );
-   mm128_dintrlv_4x32_128( d0+ 16, d1+ 16, d2+ 16, d3+ 16, s+ 64 );
-   if ( bit_len <= 256 ) return;
-   mm128_dintrlv_4x32_128( d0+ 32, d1+ 32, d2+ 32, d3+ 32, s+128 );
-   mm128_dintrlv_4x32_128( d0+ 48, d1+ 48, d2+ 48, d3+ 48, s+192 );
-   if ( bit_len <= 512 ) return;
-   mm128_dintrlv_4x32_128( d0+ 64, d1+ 64, d2+ 64, d3+ 64, s+256 );
-   if ( bit_len <= 640 ) return;
-   mm128_dintrlv_4x32_128( d0+ 80, d1+ 80, d2+ 80, d3+ 80, s+320 );
-   mm128_dintrlv_4x32_128( d0+ 96, d1+ 96, d2+ 96, d3+ 96, s+384 );
-   mm128_dintrlv_4x32_128( d0+112, d1+112, d2+112, d3+112, s+448 );
-   // bit_len == 1024
-}
-
-// extract and deinterleave specified lane.
-static inline void mm128_extr_lane_4x32( void *d, const void *s,
-                                         const int lane, const int bit_len )
-{
-  casti_m128i( d, 0 ) =
-             mm128_get_32( s, lane   , lane+ 4, lane+ 8, lane+12 );
-  casti_m128i( d, 1 ) =
-             mm128_get_32( s, lane+16, lane+20, lane+24, lane+28 );
-  if ( bit_len <= 256 ) return;
-  casti_m128i( d, 2 ) =
-             mm128_get_32( s, lane+32, lane+36, lane+40, lane+44 );
-  casti_m128i( d, 3 ) =
-             mm128_get_32( s, lane+48, lane+52, lane+56, lane+60 );
-  // bit_len == 512
-}
-
-// Interleave 80 bytes of 32 bit data for 4 lanes.
-static inline void mm128_bswap_intrlv80_4x32( void *d, const void *s )
-{
-   mm128_bswap_intrlv_4x32_128( d    , casti_m128i( s, 0 ) );
-   mm128_bswap_intrlv_4x32_128( d+ 64, casti_m128i( s, 1 ) );
-   mm128_bswap_intrlv_4x32_128( d+128, casti_m128i( s, 2 ) );
-   mm128_bswap_intrlv_4x32_128( d+192, casti_m128i( s, 3 ) );
-   mm128_bswap_intrlv_4x32_128( d+256, casti_m128i( s, 4 ) );
-}
-
-#endif // SSE2
-#endif // INTRLV_SSE2_H__
-

simd-utils/simd-128.h

@@ -1,5 +1,5 @@
-#if !defined(SIMD_SSE2_H__)
-#define SIMD_SSE2_H__ 1
+#if !defined(SIMD_128_H__)
+#define SIMD_128_H__ 1
 
 #if defined(__SSE2__)
 
@@ -15,69 +15,158 @@
 //
 // 128 bit operations are enhanced with uint128 which adds 128 bit integer
 // support for arithmetic and other operations. Casting to uint128_t is not
-// free, it requires a move from mmx to gpr but is often the only way or
-// the more efficient way for certain operations.
-
-// Compile time constant initializers are type agnostic and can have
-// a pointer handle of almost any type. All arguments must be scalar constants.
-// up to 64 bits. These iniitializers should only be used at compile time
-// to initialize vector arrays. All data reside in memory.
+// efficient but is sometimes the only way for certain operations.
 //
-// These are of limited use, it is often simpler to use uint64_t arrays
-// and cast as required.
-
-#define mm128_const_64( x1, x0 ) {{ x1, x0 }}
-#define mm128_const1_64( x )     {{  x,  x }}
-
-#define mm128_const_32( x3, x2, x1, x0 ) {{ x3, x2, x1, x0 }}
-#define mm128_const1_32( x ) {{ x,x,x,x }}
-
-#define mm128_const_16( x7, x6, x5, x4, x3, x2, x1, x0 ) \
-                     {{ x7, x6, x5, x4, x3, x2, x1, x0 }}
-#define mm128_const1_16( x ) {{ x,x,x,x, x,x,x,x }}
-
-#define mm128_const_8( x15, x14, x13, x12, x11, x10, x09, x08, \
-                       x07, x06, x05, x04, x03, x02, x01, x00 ) \
-                    {{ x15, x14, x13, x12, x11, x10, x09, x08, \
-                       x07, x06, x05, x04, x03, x02, x01, x00 }}
-#define mm128_const1_8( x ) {{ x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x }}
-
-// Compile time constants, use only for compile time initializing.
-#define c128_zero      mm128_const1_64( 0ULL )
-#define c128_one_128   mm128_const_64(  0ULL, 1ULL )  
-#define c128_one_64    mm128_const1_64( 1ULL )
-#define c128_one_32    mm128_const1_32( 1UL )
-#define c128_one_16    mm128_const1_16( 1U )
-#define c128_one_8     mm128_const1_8(  1U )
-#define c128_neg1      mm128_const1_64( 0xFFFFFFFFFFFFFFFFULL )
-#define c128_neg1_64   mm128_const1_64( 0xFFFFFFFFFFFFFFFFULL )
-#define c128_neg1_32   mm128_const1_32( 0xFFFFFFFFUL )
-#define c128_neg1_16   mm128_const1_32( 0xFFFFU )
-#define c128_neg1_8    mm128_const1_32( 0xFFU )
-
+// Constants are an issue with simd. Simply put, immediate constants don't
+// exist. All simd constants either reside in memory or a register.
+// The distibction is made below with c128 being memory resident defined
+// at compile time and m128 being register defined at run time.
+//
+// All run time constants must be generated using their components elements
+// incurring significant overhead. The more elements the more overhead
+// both in instructions and in GP register usage. Whenever possible use
+// 64 bit constant elements regardless of the actual element size.
+//
+// Due to the cost of generating constants they should not be regenerated
+// in the same function. Instead, define a local const.
+//
+// Some constant values can be generated using shortcuts. Zero for example
+// is as simple as XORing any register with itself, and is implemented
+// in the setzero instrinsic. These shortcuts must be implemented is asm
+// due to doing things the compiler would complain about. Another single
+// instruction constant is -1, defined below. Others may be added as the need
+// arises. Even single instruction constants are less efficient than local
+// register variables so the advice above stands.
+//
+// One common use for simd constants is as a control index for some simd
+// instructions like blend and shuffle. The utilities below do not take this
+// into account. Those that generate a simd constant should not be used
+// repeatedly. It may be better for the application to reimplement the
+// utility to better suit its usage.
+//
+// More tips:
+//
+// Conversions from integer to vector should be avoided whenever possible.
+// Extract, insert and set and set1 instructions should be avoided.
+// In addition to the issues with constants set is also very inefficient with
+// variables.
+// Converting integer data to perform a couple of vector operations
+// then converting back to integer should be avoided. Converting data in
+// registers should also be avoided. Conversion should be limited to buffers
+// in memory where the data is loaded directly to vector registers, bypassing
+// the integer to vector conversion.
 //
 // Pseudo constants.
 //
 // These can't be used for compile time initialization.
 // These should be used for all simple vectors.
-//
-// _mm_setzero_si128 uses pxor instruction, it's unclear what _mm_set_epi does.
-// Clearly it's faster than reading a memory resident constant. Assume set
-// is also faster.
-// If a pseudo constant is used often in a function it may be preferable
-// to define a register variable to represent that constant.
-// register __m128i zero = mm_setzero_si128().
-// This reduces any references to a move instruction.
+// Repeated usage of any simd pseudo-constant should use a locally defined
+// const rather than recomputing it for every reference.
 
 #define m128_zero      _mm_setzero_si128()
 
-#define m128_one_128   _mm_set_epi64x(  0ULL, 1ULL )
-#define m128_one_64    _mm_set1_epi64x( 1ULL )
-#define m128_one_32    _mm_set1_epi32(  1UL )
-#define m128_one_16    _mm_set1_epi16(  1U )
-#define m128_one_8     _mm_set1_epi8(   1U )
+// As suggested by Intel...
+// Arg passing for simd registers is assumed to be first output arg,
+// then input args, then locals. This is probably wrong, gcc likely picks
+// whichever register is currently holding the variable, or whichever
+// register is available to hold it. Nevertheless, all args are specified
+// by their arg number and local variables use registers starting at 
+// last arg + 1, by type.
+// Output args don't need to be listed as clobbered.
 
-#define m128_neg1      _mm_set1_epi64x( 0xFFFFFFFFFFFFFFFFULL )
+
+static inline __m128i m128_one_64_fn()
+{
+  __m128i a;
+  asm( "pxor %0, %0\n\t"
+       "pcmpeqd %%xmm1, %%xmm1\n\t"
+       "psubq %%xmm1, %0\n\t"
+       : "=x"(a)
+       :
+       : "xmm1" );
+  return a;
+}
+#define m128_one_64    m128_one_64_fn()
+
+static inline __m128i m128_one_32_fn()
+{
+  __m128i a;
+  asm( "pxor %0, %0\n\t"
+       "pcmpeqd %%xmm1, %%xmm1\n\t"
+       "psubd %%xmm1, %0\n\t"
+       : "=x"(a)
+       :
+       : "xmm1" );
+  return a;
+}
+#define m128_one_32    m128_one_32_fn()
+
+static inline __m128i m128_one_16_fn()
+{
+  __m128i a;
+  asm( "pxor %0, %0\n\t"
+       "pcmpeqd %%xmm1, %%xmm1\n\t"
+       "psubw %%xmm1, %0\n\t"
+       : "=x"(a)
+       :
+       : "xmm1" );
+  return a;
+}
+#define m128_one_16    m128_one_16_fn()
+
+static inline __m128i m128_one_8_fn()
+{
+  __m128i a;
+  asm( "pxor %0, %0\n\t"
+       "pcmpeqd %%xmm1, %%xmm1\n\t"
+       "psubb %%xmm1, %0\n\t"
+       : "=x"(a)
+       :
+       : "xmm1" );
+  return a;
+}
+#define m128_one_8    m128_one_8_fn()
+
+static inline __m128i m128_neg1_fn()
+{
+   __m128i a;
+   asm( "pcmpeqd %0, %0\n\t"
+        : "=x"(a) );
+   return a;
+}
+#define m128_neg1    m128_neg1_fn()
+
+#if defined(__SSE41__)
+
+static inline __m128i m128_one_128_fn()
+{
+   __m128i a;
+   asm( "pinsrq $0, $1, %0\n\t"
+        "pinsrq $1, $0, %0\n\t"
+        : "=x"(a) );
+   return a;
+}
+#define m128_one_128    m128_one_128_fn()
+
+// alternative to _mm_set_epi64x, doesn't use mem,
+// cost = 2 pinsrt, estimate 4 clocks.
+static inline __m128i m128_const_64( uint64_t hi, uint64_t lo )
+{
+   __m128i a;
+   asm( "pinsrq $0, %2, %0\n\t"
+        "pinsrq $1, %1, %0\n\t"
+        : "=x"(a)
+        : "r"(hi), "r"(lo) );
+   return a;
+} 
+
+#else
+
+#define m128_one_128   _mm_set_epi64x(  0ULL, 1ULL )
+
+#define m128_const_64 _mm_set_epi64x
+
+#endif
 
 //
 // Basic operations without equivalent SIMD intrinsic
@@ -90,9 +179,21 @@
 #define mm128_negate_32( v )    _mm_sub_epi32( m128_zero, v )  
 #define mm128_negate_16( v )    _mm_sub_epi16( m128_zero, v )  
 
-// Use uint128_t for most arithmetic, bit shift, comparison operations
-// spanning all 128 bits. Some extractions are also more efficient 
-// casting __m128i as uint128_t and usingstandard operators.
+// Add 4 values, fewer dependencies than sequential addition.
+#define mm128_add4_64( a, b, c, d ) \
+   _mm_add_epi64( _mm_add_epi64( a, b ), _mm_add_epi64( c, d ) )
+
+#define mm128_add4_32( a, b, c, d ) \
+   _mm_add_epi32( _mm_add_epi32( a, b ), _mm_add_epi32( c, d ) )
+
+#define mm128_add4_16( a, b, c, d ) \
+   _mm_add_epi16( _mm_add_epi16( a, b ), _mm_add_epi16( c, d ) )
+
+#define mm128_add4_8( a, b, c, d ) \
+   _mm_add_epi8( _mm_add_epi8( a, b ), _mm_add_epi8( c, d ) )
+
+#define mm128_xor4( a, b, c, d ) \
+   _mm_xor_si128( _mm_xor_si128( a, b ), _mm_xor_si128( c, d ) )
 
 // This isn't cheap, not suitable for bulk usage.
 #define mm128_extr_4x32( a0, a1, a2, a3, src ) \
@@ -105,6 +206,16 @@ do { \
 
 // Horizontal vector testing
 
+#if defined(__SSE41__)
+
+#define mm128_allbits0( a )    _mm_testz_si128(   a, a )
+#define mm128_allbits1( a )    _mm_testc_si128(   a, m128_neg1 )
+#define mm128_allbitsne( a )   _mm_testnzc_si128( a, m128_neg1 )
+#define mm128_anybits0         mm128_allbitsne
+#define mm128_anybits1         mm128_allbitsne
+
+#else   // SSE2
+
 // Bit-wise test of entire vector, useful to test results of cmp.
 #define mm128_anybits0( a ) (uint128_t)(a)
 #define mm128_anybits1( a ) (((uint128_t)(a))+1)
@@ -112,6 +223,8 @@ do { \
 #define mm128_allbits0( a ) ( !mm128_anybits1(a) )
 #define mm128_allbits1( a ) ( !mm128_anybits0(a) )
 
+#endif // SSE41 else SSE2
+
 //
 // Vector pointer cast
 
@@ -139,6 +252,7 @@ do { \
 
 #else
 
+// Doesn't work with register variables.
 #define mm128_extr_64(a,n)   (((uint64_t*)&a)[n])
 #define mm128_extr_32(a,n)   (((uint32_t*)&a)[n])
 
@@ -209,7 +323,7 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, int n )
 // Bit rotations
 
 // AVX512 has implemented bit rotation for 128 bit vectors with
-// 64 and 32 bit elements. Not really useful.
+// 64 and 32 bit elements.
 
 //
 // Rotate each element of v by c bits
@@ -233,13 +347,16 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, int n )
    _mm_or_si128( _mm_slli_epi16( v, c ), _mm_srli_epi16( v, 16-(c) ) )
 
 //
-// Rotate elements accross all lanes
+// Rotate vector elements accross all lanes
 
 #define mm128_swap_64( v )    _mm_shuffle_epi32( v, 0x4e )
 
 #define mm128_ror_1x32( v )   _mm_shuffle_epi32( v, 0x39 )
 #define mm128_rol_1x32( v )   _mm_shuffle_epi32( v, 0x93 )
 
+#if defined (__SSE3__)
+// no SSE2 implementation, no current users
+
 #define mm128_ror_1x16( v ) \
    _mm_shuffle_epi8( v, _mm_set_epi8(  1, 0,15,14,13,12,11,10 \
                                        9, 8, 7, 6, 5, 4, 3, 2 ) )
@@ -252,6 +369,7 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, int n )
 #define mm128_rol_1x8( v ) \
    _mm_shuffle_epi8( v, _mm_set_epi8( 14,13,12,11,10, 9, 8, 7, \
                                        6, 5, 4, 3, 2, 1, 0,15 ) )
+#endif  // SSE3
 
 // Rotate 16 byte (128 bit) vector by c bytes.
 // Less efficient using shift but more versatile. Use only for odd number
@@ -262,17 +380,6 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, int n )
 #define mm128_brol( v, c ) \
    _mm_or_si128( _mm_slli_si128( v, c ), _mm_srli_si128( v, 16-(c) ) )
 
-// Invert vector: {3,2,1,0} -> {0,1,2,3}
-#define mm128_invert_32( v ) _mm_shuffle_epi32( a, 0x1b )
-
-#define mm128_invert_16( v ) \
-   _mm_shuffle_epi8( v, _mm_set_epi8( 1, 0,   3, 2,   5, 4,   7, 6, \
-                                      9, 8,  11,10,  13,12,  15,14 ) )
-
-#define mm128_invert_8( v ) \
-   _mm_shuffle_epi8( v, _mm_set_epi8( 0, 1, 2, 3, 4, 5, 6, 7, \
-                                      8, 9,10,11,12,13,14,15 ) )
-
 //
 // Rotate elements within lanes.
 
@@ -283,7 +390,6 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, int n )
 #define mm128_rol16_64( v )   _mm_shuffle_epi8( v, \
               _mm_set_epi8( 13,12,11,10, 9, 8,15,14,  5, 4, 3, 2, 1, 0, 7, 6 )
 
-
 #define mm128_swap16_32( v )  _mm_shuffle_epi8( v, \
                       _mm_set_epi8( 13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2 )
 
@@ -293,17 +399,45 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, int n )
 #if defined(__SSSE3__)
 
 #define mm128_bswap_64( v ) \
-   _mm_shuffle_epi8( v, _mm_set_epi8( 8, 9,10,11,12,13,14,15, \
-                                      0, 1, 2, 3, 4, 5, 6, 7 ) )
+   _mm_shuffle_epi8( v, m128_const64(  0x08090a0b0c0d0e0f, \
+                                       0x0001020304050607 ) )
 
 #define mm128_bswap_32( v ) \
-   _mm_shuffle_epi8( v, _mm_set_epi8( 12,13,14,15,   8, 9,10,11, \
-                                       4, 5, 6, 7,   0, 1, 2, 3 ) )
+   _mm_shuffle_epi8( v, m128_const_64( 0x0c0d0e0f08090a0b, \
+                                       0x0405060700010203 ) )
 
 #define mm128_bswap_16( v ) \
    _mm_shuffle_epi8( v, _mm_set_epi8( 14,15,  12,13,  10,11,   8, 9, \
                                        6, 7,   4, 5,   2, 3,   0, 1 ) )
 
+// 8 byte qword * 8 qwords * 2 lanes = 128 bytes
+#define mm128_block_bswap_64( d, s ) do \
+{ \
+   __m128i ctl = m128_const_64(  0x08090a0b0c0d0e0f, 0x0001020304050607 ); \
+  casti_m128i( d, 0 ) = _mm_shuffle_epi8( casti_m128i( s, 0 ), ctl ); \
+  casti_m128i( d, 1 ) = _mm_shuffle_epi8( casti_m128i( s, 1 ), ctl ); \
+  casti_m128i( d, 2 ) = _mm_shuffle_epi8( casti_m128i( s, 2 ), ctl ); \
+  casti_m128i( d, 3 ) = _mm_shuffle_epi8( casti_m128i( s, 3 ), ctl ); \
+  casti_m128i( d, 4 ) = _mm_shuffle_epi8( casti_m128i( s, 4 ), ctl ); \
+  casti_m128i( d, 5 ) = _mm_shuffle_epi8( casti_m128i( s, 5 ), ctl ); \
+  casti_m128i( d, 6 ) = _mm_shuffle_epi8( casti_m128i( s, 6 ), ctl ); \
+  casti_m128i( d, 7 ) = _mm_shuffle_epi8( casti_m128i( s, 7 ), ctl ); \
+} while(0)
+
+// 4 byte dword * 8 dwords * 4 lanes = 128 bytes
+#define mm128_block_bswap_32( d, s ) do \
+{ \
+   __m128i ctl = m128_const_64( 0x0c0d0e0f08090a0b, 0x0405060700010203 ); \
+  casti_m128i( d, 0 ) = _mm_shuffle_epi8( casti_m128i( s, 0 ), ctl ); \
+  casti_m128i( d, 1 ) = _mm_shuffle_epi8( casti_m128i( s, 1 ), ctl ); \
+  casti_m128i( d, 2 ) = _mm_shuffle_epi8( casti_m128i( s, 2 ), ctl ); \
+  casti_m128i( d, 3 ) = _mm_shuffle_epi8( casti_m128i( s, 3 ), ctl ); \
+  casti_m128i( d, 4 ) = _mm_shuffle_epi8( casti_m128i( s, 4 ), ctl ); \
+  casti_m128i( d, 5 ) = _mm_shuffle_epi8( casti_m128i( s, 5 ), ctl ); \
+  casti_m128i( d, 6 ) = _mm_shuffle_epi8( casti_m128i( s, 6 ), ctl ); \
+  casti_m128i( d, 7 ) = _mm_shuffle_epi8( casti_m128i( s, 7 ), ctl ); \
+} while(0)
+
 #else  // SSE2
 
 // Use inline function instead of macro due to multiple statements.
@@ -326,16 +460,41 @@ static inline __m128i mm128_bswap_16( __m128i v )
   return _mm_or_si128( _mm_slli_epi16( v, 8 ), _mm_srli_epi16( v, 8 ) );
 }
 
+static inline void mm128_block_bswap_64( __m128i *d, __m128i *s )
+{
+   d[0] = mm128_bswap_32( s[0] );
+   d[1] = mm128_bswap_32( s[1] );
+   d[2] = mm128_bswap_32( s[2] );
+   d[3] = mm128_bswap_32( s[3] );
+   d[4] = mm128_bswap_32( s[4] );
+   d[5] = mm128_bswap_32( s[5] );
+   d[6] = mm128_bswap_32( s[6] );
+   d[7] = mm128_bswap_32( s[7] );
+}
+
+static inline void mm128_block_bswap_32( __m128i *d, __m128i *s )
+{
+   d[0] = mm128_bswap_32( s[0] );
+   d[1] = mm128_bswap_32( s[1] );
+   d[2] = mm128_bswap_32( s[2] );
+   d[3] = mm128_bswap_32( s[3] );
+   d[4] = mm128_bswap_32( s[4] );
+   d[5] = mm128_bswap_32( s[5] );
+   d[6] = mm128_bswap_32( s[6] );
+   d[7] = mm128_bswap_32( s[7] );
+}
+
 #endif // SSSE3 else SSE2
+
 //
 // Rotate in place concatenated 128 bit vectors as one 256 bit vector.
 
 // Swap 128 bit vectorse.
 
-#define mm128_swap128_256(v1, v2) \
-   v1 = _mm_xor_si128(v1, v2); \
-   v2 = _mm_xor_si128(v1, v2); \
-   v1 = _mm_xor_si128(v1, v2);
+#define mm128_swap128_256( v1, v2 ) \
+   v1 = _mm_xor_si128( v1, v2 ); \
+   v2 = _mm_xor_si128( v1, v2 ); \
+   v1 = _mm_xor_si128( v1, v2 );
 
 // Concatenate v1 & v2 and rotate as one 256 bit vector.
 #if defined(__SSE4_1__)
@@ -457,4 +616,4 @@ do { \
 #endif  // SSE4.1 else SSE2
 
 #endif // __SSE2__
-#endif // SIMD_SSE2_H__
+#endif // SIMD_128_H__

simd-utils/simd-256.h

@@ -1,44 +1,134 @@
-#if !defined(SIMD_AVX2_H__)
-#define SIMD_AVX2_H__ 1
+#if !defined(SIMD_256_H__)
+#define SIMD_256_H__ 1
 
-#if defined(__AVX2__)
+#if defined(__AVX__)
 
 /////////////////////////////////////////////////////////////////////
 //
 //             AVX2 256 bit vectors
 //
-// AVX2 is required for integer support of 256 bit vectors.
+// Basic support for 256 bit vectors is available with AVX but integer
+// support requires AVX2.
 // Some 256 bit vector utilities require AVX512 or have more efficient
 // AVX512 implementations. They will be selected automatically but their use
 // is limited because 256 bit vectors are less likely to be used when 512
 // is available.
 
-// Vector type overlays used by compile time vector constants.
-// Constants of these types reside in memory.
-
-
 //
-// Basic operations without SIMD equivalent
+// Pseudo constants.
+// These can't be used for compile time initialization but are preferable
+// for simple constant vectors at run time. For repeated use define a local
+// constant to avoid multiple calls to the same macro.
 
-// Bitwise not ( ~x )
-#define mm256_not( x )       _mm256_xor_si256( (x), m256_neg1 ) \
+#define m256_zero         _mm256_setzero_si256()
 
-// Unary negation of each element ( -a )
-#define mm256_negate_64( a ) _mm256_sub_epi64( m256_zero, a )
-#define mm256_negate_32( a ) _mm256_sub_epi32( m256_zero, a )
-#define mm256_negate_16( a ) _mm256_sub_epi16( m256_zero, a )
+#define m256_one_256 \
+   _mm256_insertf128_si256( _mm256_castsi128_si256( m128_one_128 ), \
+                                                    m128_zero, 1 )
 
-/***************************
- *
- * extracti128 (AVX2) vs extractf128 (AVX)???
- 
- 
+#define m256_one_128 \
+   _mm256_insertf128_si256( _mm256_castsi128_si256( m128_one_128 ), \
+                                                    m128_one_128, 1 )
+
+// set instructions load memory resident constants, this avoids mem.
+// cost 4 pinsert + 1 vinsert, estimate 7 clocks.
+#define m256_const_64( i3, i2, i1, i0 ) \
+   _mm256_insertf128_si256( _mm256_castsi128_si256( m128_const_64( i1, i0 ) ), \
+                            m128_const_64( i3, i2 ), 1 )
+#define m256_const1_64( i ) m256_const_64( i, i, i, i )
+
+#if defined(__AVX2__)
+
+// These look like a lot of overhead but the compiler optimizes nicely
+// and puts the asm inline in the calling function. Usage is like any
+// variable expression.
+// __m256i foo = m256_one_64; 
+
+static inline __m256i m256_one_64_fn()
+{
+  __m256i a;
+  asm( "vpxor %0, %0, %0\n\t"
+       "vpcmpeqd %%ymm1, %%ymm1, %%ymm1\n\t"
+       "vpsubq %%ymm1, %0, %0\n\t"
+       :"=x"(a)
+       :
+       : "ymm1" );
+  return a;
+}
+#define m256_one_64    m256_one_64_fn()
+
+static inline __m256i m256_one_32_fn()
+{
+  __m256i a;
+  asm( "vpxor %0, %0, %0\n\t"
+       "vpcmpeqd %%ymm1, %%ymm1, %%ymm1\n\t"
+       "vpsubd %%ymm1, %0, %0\n\t"
+       :"=x"(a)
+       :
+       : "ymm1" );
+  return a;
+}
+#define m256_one_32    m256_one_32_fn()
+
+static inline __m256i m256_one_16_fn()
+{
+  __m256i a;
+  asm( "vpxor %0, %0, %0\n\t"
+       "vpcmpeqd %%ymm1, %%ymm1, %%ymm1\n\t"
+       "vpsubw %%ymm1, %0, %0\n\t"
+       :"=x"(a)
+       :
+       : "ymm1" );
+  return a;
+}
+#define m256_one_16    m256_one_16_fn()
+
+static inline __m256i m256_one_8_fn()
+{
+  __m256i a;
+  asm( "vpxor %0, %0, %0\n\t"
+       "vpcmpeqd %%ymm1, %%ymm1, %%ymm1\n\t"
+       "vpsubb %%ymm1, %0, %0\n\t"
+       :"=x"(a)
+       :
+       : "ymm1" );
+  return a;
+}
+#define m256_one_8    m256_one_8_fn()
+
+static inline __m256i m256_neg1_fn()
+{
+   __m256i a;
+   asm( "vpcmpeqq %0, %0, %0\n\t"
+        :"=x"(a) );
+   return a;
+}
+#define m256_neg1    m256_neg1_fn()
+
+#else  // AVX
+
+#define m256_one_64       _mm256_set1_epi64x( 1ULL )
+#define m256_one_32       _mm256_set1_epi64x( 0x0000000100000001ULL )
+#define m256_one_16       _mm256_set1_epi64x( 0x0001000100010001ULL )
+#define m256_one_8        _mm256_set1_epi64x( 0x0101010101010101ULL )
+
+// AVX doesn't have inserti128 but insertf128 will do. 
+// Ideally this can be done with 2 instructions and no temporary variables.
+static inline __m256i m256_neg1_fn()
+{
+   __m128i a = m128_neg1;
+   return _mm256_insertf128_si256( _mm256_castsi128_si256( a ), a, 1 );
+}
+#define m256_neg1   m256_neg1_fn()
+//#define m256_neg1    _mm256_set1_epi64x( 0xFFFFFFFFFFFFFFFFULL )
+
+#endif  // AVX2 else AVX
 //
 // Vector size conversion.
 //
 // Allows operations on either or both halves of a 256 bit vector serially.
 // Handy for parallel AES.
-// Caveats:
+// Caveats when writing:
 //      _mm256_castsi256_si128 is free and without side effects.
 //      _mm256_castsi128_si256 is also free but leaves the high half
 //      undefined. That's ok if the hi half will be subseqnently assigned.
@@ -78,14 +168,22 @@ do { \
 // Insert b into specified half of a leaving other half of a unchanged.
 #define mm256_ins_lo128_256( a, b )  _mm256_inserti128_si256( a, b, 0 )
 #define mm256_ins_hi128_256( a, b )  _mm256_inserti128_si256( a, b, 1 )
-*/
 
-/*
+
 // concatenate two 128 bit vectors into one 256 bit vector: { hi, lo }
 #define mm256_concat_128( hi, lo ) \
    mm256_ins_hi128_256( _mm256_castsi128_si256( lo ), hi )
 
 // Horizontal vector testing
+#if defined(__AVX2__)
+
+#define mm256_allbits0( a )    _mm256_testz_si256(   a, a )
+#define mm256_allbits1( a )    _mm256_testc_si256(   a, m256_neg1 )
+#define mm256_allbitsne( a )   _mm256_testnzc_si256( a, m256_neg1 )
+#define mm256_anybits0         mm256_allbitsne
+#define mm256_anybits1         mm256_allbitsne
+
+#else  // AVX
 
 // Bit-wise test of entire vector, useful to test results of cmp.
 #define mm256_anybits0( a ) \
@@ -99,35 +197,20 @@ do { \
 #define mm256_allbits0_256( a ) ( !mm256_anybits1(a) )
 #define mm256_allbits1_256( a ) ( !mm256_anybits0(a) )
 
+#endif   // AVX2 else AVX
+
 // Parallel AES, for when x is expected to be in a 256 bit register.
-#define mm256_aesenc_2x128( x ) \
-     mm256_concat_128( \
-     _mm_aesenc_si128( mm128_extr_hi128_256( x ), m128_zero ), \
-          _mm_aesenc_si128( mm128_extr_lo128_256( x ), m128_zero ) )
+// Use same 128 bit key.
+#define mm256_aesenc_2x128( x, k ) \
+   mm256_concat_128( _mm_aesenc_si128( mm128_extr_hi128_256( x ), k ), \
+                     _mm_aesenc_si128( mm128_extr_lo128_256( x ), k ) )
 
-#define mm256_aesenckey_2x128( x, k ) \
-     mm256_concat_128( \
-     _mm_aesenc_si128( mm128_extr_hi128_256( x ), \
-                       mm128_extr_lo128_256( k ) ), \
-     _mm_aesenc_si128( mm128_extr_hi128_256( x ), \
-                       mm128_extr_lo128_256( k ) ) )
-
-#define mm256_paesenc_2x128( y, x ) do \
+#define mm256_paesenc_2x128( y, x, k ) do \
 { \
-  __m256i *X = (__m256i*)x; \
-  __m256i *Y = (__m256i*)y; \
-  y[0] = _mm_aesenc_si128( x[0], m128_zero ); \
-  y[1] = _mm_aesenc_si128( x[1], m128_zero ); \
-} while(0);
-
-// With pointers.
-#define mm256_paesenckey_2x128( y, x, k ) do \
-{ \
-  __m256i *X = (__m256i*)x; \
-  __m256i *Y = (__m256i*)y; \
-  __m256i *K = (__m256i*)ky; \
-  y[0] = _mm_aesenc_si128( x[0], K[0] ); \
-  y[1] = _mm_aesenc_si128( x[1], K[1] ); \
+  __m128i *X = (__m128i*)x; \
+  __m128i *Y = (__m128i*)y; \
+  Y[0] = _mm_aesenc_si128( X[0], k ); \
+  Y[1] = _mm_aesenc_si128( X[1], k ); \
 } while(0);
 
 //
@@ -201,7 +284,41 @@ static inline void memset_256( __m256i *dst, const __m256i a,  int n )
 static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
 {   for ( int i = 0; i < n; i ++ ) dst[i] = src[i]; }
 
-*************************************/
+///////////////////////////////
+//
+//    AVX2 needed from now on.
+//
+
+#if defined(__AVX2__)
+
+//
+// Basic operations without SIMD equivalent
+
+// Bitwise not ( ~x )
+#define mm256_not( x )       _mm256_xor_si256( (x), m256_neg1 ) \
+
+// Unary negation of each element ( -a )
+#define mm256_negate_64( a ) _mm256_sub_epi64( m256_zero, a )
+#define mm256_negate_32( a ) _mm256_sub_epi32( m256_zero, a )
+#define mm256_negate_16( a ) _mm256_sub_epi16( m256_zero, a )
+
+
+// Add 4 values, fewer dependencies than sequential addition.
+
+#define mm256_add4_64( a, b, c, d ) \
+   _mm256_add_epi64( _mm256_add_epi64( a, b ), _mm256_add_epi64( c, d ) )
+
+#define mm256_add4_32( a, b, c, d ) \
+   _mm256_add_epi32( _mm256_add_epi32( a, b ), _mm256_add_epi32( c, d ) )
+
+#define mm256_add4_16( a, b, c, d ) \
+   _mm256_add_epi16( _mm256_add_epi16( a, b ), _mm256_add_epi16( c, d ) )
+
+#define mm256_add4_8( a, b, c, d ) \
+   _mm256_add_epi8( _mm256_add_epi8( a, b ), _mm256_add_epi8( c, d ) )
+
+#define mm256_xor4( a, b, c, d ) \
+   _mm256_xor_si256( _mm256_xor_si256( a, b ), _mm256_xor_si256( c, d ) )
 
 //
 //           Bit rotations.
@@ -241,24 +358,27 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
 // index vector c
 #define mm256_rorv_64( v, c ) \
    _mm256_or_si256( \
-         _mm256_srlv_epi64( v, _mm256_set1_epi64x( c ) ), \
-         _mm256_sllv_epi64( v, _mm256_set1_epi64x( 64-(c) ) ) )
+         _mm256_srlv_epi64( v, c ), \
+         _mm256_sllv_epi64( v, _mm256_sub_epi64( \
+                                   _mm256_set1_epi64x( 64 ), c ) ) )
 
 #define mm256_rolv_64( v, c ) \
    _mm256_or_si256( \
-         _mm256_sllv_epi64( v, _mm256_set1_epi64x( c ) ), \
-         _mm256_srlv_epi64( v, _mm256_set1_epi64x( 64-(c) ) ) )
-
+         _mm256_sllv_epi64( v, c ), \
+         _mm256_srlv_epi64( v, _mm256_sub_epi64( \
+                                   _mm256_set1_epi64x( 64 ), c ) ) )
 
 #define mm256_rorv_32( v, c ) \
    _mm256_or_si256( \
-         _mm256_srlv_epi32( v, _mm256_set1_epi32( c ) ), \
-         _mm256_sllv_epi32( v, _mm256_set1_epi32( 32-(c) ) ) )
+         _mm256_srlv_epi32( v, c ), \
+         _mm256_sllv_epi32( v, _mm256_sub_epi32( \
+                                  _mm256_set1_epi32( 32 ), c ) ) )
 
 #define mm256_rolv_32( v, c ) \
    _mm256_or_si256( \
-         _mm256_sllv_epi32( v, _mm256_set1_epi32( c ) ), \
-         _mm256_srlv_epi32( v, _mm256_set1_epi32( 32-(c) ) ) )
+         _mm256_sllv_epi32( v, c ), \
+         _mm256_srlv_epi32( v, _mm256_sub_epi32( \
+                                     _mm256_set1_epi32( 32 ), c ) ) )
 
 // AVX512 can do 16 bit elements.
 
@@ -275,17 +395,28 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
 #define mm256_ror_1x64( v )     _mm256_permute4x64_epi64( v, 0x39 )
 #define mm256_rol_1x64( v )     _mm256_permute4x64_epi64( v, 0x93 )
 
-// Rotate 256 bit vector by one 32 bit element.
+// A little faster with avx512
+// Rotate 256 bit vector by one 32 bit element. Use 64 bit set, it's faster.
 #define mm256_ror_1x32( v ) \
-    _mm256_permutevar8x32_epi32( v, _mm256_set_epi32( 0,7,6,5, 4,3,2,1 ) )
+    _mm256_permutevar8x32_epi32( v, \
+                     m256_const_64( 0x0000000000000007, 0x0000000600000005, \
+                                    0x0000000400000003, 0x0000000200000001 )
+
 #define mm256_rol_1x32( v ) \
-    _mm256_permutevar8x32_epi32( v, _mm256_set_epi32( 6,5,4,3, 2,1,0,7 ) )
+    _mm256_permutevar8x32_epi32( v, \
+                     m256_const_64( 0x0000000600000005,  0x0000000400000003, \
+                                    0x0000000200000001,  0x0000000000000007 )
 
 // Rotate 256 bit vector by three 32 bit elements (96 bits).
 #define mm256_ror_3x32( v ) \
-    _mm256_permutevar8x32_epi32( v, _mm256_set_epi32( 2,1,0,7, 6,5,4,3 ) )
+    _mm256_permutevar8x32_epi32( v, \
+                     m256_const_64( 0x0000000200000001, 0x0000000000000007, \
+                                    0x0000000600000005, 0x0000000400000003 ) 
+
 #define mm256_rol_3x32( v ) \
-    _mm256_permutevar8x32_epi32( v, _mm256_set_epi32( 4,3,2,1, 0,7,6,5 ) )
+    _mm256_permutevar8x32_epi32( v, \
+                     m256_const_64( 0x0000000400000003, 0x0000000200000001, \
+                                    0x0000000000000007, 0x0000000600000005 )
 
 // AVX512 can do 16 & 8 bit elements.
 #if defined(__AVX512VL__)
@@ -293,7 +424,7 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
 // Rotate 256 bit vector by one 16 bit element.     
 #define mm256_ror_1x16( v ) \
    _mm256_permutexvar_epi16( _mm256_set_epi16( \
-    0,15,14,13,12,11,10, 9,   8, 7, 6, 5, 4, 3, 2, 1 ), v )
+         0,15,14,13,12,11,10, 9,   8, 7, 6, 5, 4, 3, 2, 1 ), v )
 
 #define mm256_rol_1x16( v ) \
    _mm256_permutexvar_epi16( _mm256_set_epi16( \
@@ -303,7 +434,7 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
 #define mm256_ror_1x8( v ) \
    _mm256_permutexvar_epi8( _mm256_set_epi8( \
          0,31,30,29,28,27,26,25,  24,23,22,21,20,19,18,17, \
-   16,15,14,13,12,11,10, 9,   8, 7, 6, 5, 4, 3, 2, 1 ), v )
+        16,15,14,13,12,11,10, 9,   8, 7, 6, 5, 4, 3, 2, 1 ), v )
 
 #define mm256_rol_1x8( v ) \
    _mm256_permutexvar_epi8( _mm256_set_epi8( \
@@ -312,14 +443,6 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
 
 #endif  // AVX512
 
-// Invert vector: {3,2,1,0} -> {0,1,2,3}
-#define mm256_invert_64( v ) _mm256_permute4x64_epi64( a, 0x1b )
-
-#define mm256_invert_32( v ) \
-     _mm256_permutevar8x32_epi32( v, _mm256_set_epi32( 0,1,2,3,4,5,6,7 ) )
-
-// AVX512 can do 16 & 8 bit elements.
-
 //
 // Rotate elements within lanes of 256 bit vector.
 
@@ -332,15 +455,23 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
 
 // Rotate each 128 bit lane by one 16 bit element.
 #define mm256_rol1x16_128( v ) \
-         _mm256_shuffle_epi8( 13,12,11,10, 9,8,7,6, 5,4,3,2, 1,0,15,14 )
+        _mm256_shuffle_epi8( v, _mm256_set_epi16( 6,5,4,3,2,1,0,7, \
+                                                  6,5,4,3,2,1,0,7 ) )
 #define mm256_ror1x16_128( v ) \
-        _mm256_shuffle_epi8( 1,0,15,14, 13,12,11,10, 9,8,7,6, 5,4,3,2 )
+        _mm256_shuffle_epi8( v, _mm256_set_epi16( 0,7,6,5,4,3,2,1, \
+                                                  0,7,6,5,4,3,2,1 ) )
 
 // Rotate each 128 bit lane by one byte
 #define mm256_rol1x8_128( v ) \
-        _mm256_shuffle_epi8( 14, 13,12,11, 10,9,8,7, 6,5,4,3, 2,1,0,15 )
+        _mm256_shuffle_epi8( v, _mm256_set_epi8(14,13,12,11,10, 9, 8, 7, \
+                                                 6, 5, 4, 3, 2, 1, 0,15, \
+                                                14,13,12,11,10, 9, 8, 7, \
+                                                 6, 5, 4, 3, 2, 1, 0,15 ) )
 #define mm256_ror1x8_128( v ) \
-        _mm256_shuffle_epi8( 0,15,14,13, 12,11,10,9, 8,7,6,5, 4,3,2,1 )
+        _mm256_shuffle_epi8( v, _mm256_set_epi8( 0,15,14,13,12,11,10, 9, \
+                                                 8, 7, 6, 5, 4, 3, 2, 1, \
+                                                 0,15,14,13,12,11,10, 9, \
+                                                 8, 7, 6, 5, 4, 3, 2, 1 ) )
 
 // Rotate each 128 bit lane by c bytes.
 #define mm256_bror_128( v, c ) \
@@ -354,28 +485,27 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
 #define mm256_swap32_64( v )    _mm256_shuffle_epi32( v, 0xb1 )
 
 #define mm256_ror16_64( v ) \
-      _mm256_shuffle_epi8(  9, 8,15,14,13,12,11,10,  1, 0, 7, 6, 5, 4, 3, 2 );
+   _mm256_shuffle_epi8( v, _mm256_set_epi16( 4,7,6,5,0,3,2,1, \
+                                             4,7,6,5,0,3,2,1 ) )
 #define mm256_rol16_64( v ) \
-      _mm256_shuffle_epi8( 13,12,11,10, 9, 8,15,14,  5, 4, 3, 2, 1, 0, 7, 6 );
+   _mm256_shuffle_epi8( v, _mm256_set_epi16( 6,5,4,7,2,1,0,3, \
+                                             6,5,4,7,2,1,0,3 ) )
 
 
 // Swap 16 bit elements in each 32 bit lane
-#define mm256_swap16_32( v )  _mm256_shuffle_epi8( v, \
-        _mm_set_epi8( 13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2 )
+#define mm256_swap16_32( v ) \
+   _mm256_shuffle_epi8( v, _mm256_set_epi16( 6,7,4,5,2,3,0,1, \
+                                             6,7,4,5,2,3,0,1 ) )
 
 //
 // Swap bytes in vector elements, endian bswap.
 #define mm256_bswap_64( v ) \
-   _mm256_shuffle_epi8( v, _mm256_set_epi8( 8, 9,10,11,12,13,14,15, \
-                                            0, 1, 2, 3, 4, 5, 6, 7, \
-                                            8, 9,10,11,12,13,14,15, \
-                                            0, 1, 2, 3, 4, 5, 6, 7 ) )
+   _mm256_shuffle_epi8( v, m256_const_64( 0x08090a0b0c0d0e0f, \
+                0x0001020304050607, 0x08090a0b0c0d0e0f, 0x0001020304050607 ) )
 
 #define mm256_bswap_32( v ) \
-   _mm256_shuffle_epi8( v, _mm256_set_epi8( 12,13,14,15,   8, 9,10,11, \
-                                             4, 5, 6, 7,   0, 1, 2, 3, \
-                                            12,13,14,15,   8, 9,10,11, \
-                                             4, 5, 6, 7,   0, 1, 2, 3 ) )
+   _mm256_shuffle_epi8( v, m256_const_64( 0x0c0d0e0f08090a0b, \
+                0x0405060700010203, 0x0c0d0e0f08090a0b, 0x0405060700010203 ) )
 
 #define mm256_bswap_16( v ) \
    _mm256_shuffle_epi8( v, _mm256_set_epi8(  14,15,  12,13,  10,11,   8, 9, \
@@ -383,6 +513,36 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
                                              14,15,  12,13,  10,11,   8, 9, \
                                               6, 7,   4, 5,   2, 3,   0, 1 ) )
 
+// 8 byte qword * 8 qwords * 4 lanes = 256 bytes
+#define mm256_block_bswap_64( d, s ) do \
+{ \
+  __m256i ctl = m256_const_64( 0x08090a0b0c0d0e0f, 0x0001020304050607, \
+                               0x08090a0b0c0d0e0f, 0x0001020304050607 ); \
+  casti_m256i( d, 0 ) = _mm256_shuffle_epi8( casti_m256i( s, 0 ), ctl ); \
+  casti_m256i( d, 1 ) = _mm256_shuffle_epi8( casti_m256i( s, 1 ), ctl ); \
+  casti_m256i( d, 2 ) = _mm256_shuffle_epi8( casti_m256i( s, 2 ), ctl ); \
+  casti_m256i( d, 3 ) = _mm256_shuffle_epi8( casti_m256i( s, 3 ), ctl ); \
+  casti_m256i( d, 4 ) = _mm256_shuffle_epi8( casti_m256i( s, 4 ), ctl ); \
+  casti_m256i( d, 5 ) = _mm256_shuffle_epi8( casti_m256i( s, 5 ), ctl ); \
+  casti_m256i( d, 6 ) = _mm256_shuffle_epi8( casti_m256i( s, 6 ), ctl ); \
+  casti_m256i( d, 7 ) = _mm256_shuffle_epi8( casti_m256i( s, 7 ), ctl ); \
+} while(0)
+
+// 4 byte dword * 8 dwords * 8 lanes = 256 bytes
+#define mm256_block_bswap_32( d, s ) do \
+{ \
+  __m256i ctl = m256_const_64( 0x0c0d0e0f08090a0b, 0x0405060700010203, \
+                               0x0c0d0e0f08090a0b, 0x0405060700010203 ); \
+  casti_m256i( d, 0 ) = _mm256_shuffle_epi8( casti_m256i( s, 0 ), ctl ); \
+  casti_m256i( d, 1 ) = _mm256_shuffle_epi8( casti_m256i( s, 1 ), ctl ); \
+  casti_m256i( d, 2 ) = _mm256_shuffle_epi8( casti_m256i( s, 2 ), ctl ); \
+  casti_m256i( d, 3 ) = _mm256_shuffle_epi8( casti_m256i( s, 3 ), ctl ); \
+  casti_m256i( d, 4 ) = _mm256_shuffle_epi8( casti_m256i( s, 4 ), ctl ); \
+  casti_m256i( d, 5 ) = _mm256_shuffle_epi8( casti_m256i( s, 5 ), ctl ); \
+  casti_m256i( d, 6 ) = _mm256_shuffle_epi8( casti_m256i( s, 6 ), ctl ); \
+  casti_m256i( d, 7 ) = _mm256_shuffle_epi8( casti_m256i( s, 7 ), ctl ); \
+} while(0)
+
 //
 // Rotate two concatenated 256 bit vectors as one 512 bit vector by specified
 // number of elements. Rotate is done in place, source arguments are
@@ -466,5 +626,6 @@ do { \
 } while(0)
 
 #endif // __AVX2__
-#endif // SIMD_AVX2_H__
+#endif // __AVX__
+#endif // SIMD_256_H__
 

simd-utils/simd-512.h

@@ -1,5 +1,5 @@
-#if !defined(SIMD_AVX512_H__)
-#define SIMD_AVX512_H__ 1
+#if !defined(SIMD_512_H__)
+#define SIMD_512_H__ 1
 
 #if defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
 
@@ -41,79 +41,6 @@
 // Experimental, not fully tested.
 
 
-//
-// Compile time vector constants and initializers.
-//
-// The following macro constants and functions should only be used
-// for compile time initialization of constant and variable vector
-// arrays. These constants use memory, use set instruction or pseudo
-// constants at run time to avoid using memory.
-
-// Constant initializers
-#define mm512_const_64( x7, x6, x5, x4, x3, x2, x1, x0 ) \
-                     {{ x7, x6, x5, x4, x3, x2, x1, x0 }}
-
-#define mm512_const1_64( x ) {{ x,x,x,x,x,x,x }}
-
-#define mm512_const_32( x15, x14, x13, x12, x11, x10, x09, x08, \
-                        x07, x06, x05, x04, x03, x02, x01, x00 ) \
-                     {{ x15, x14, x13, x12, x11, x10, x09, x08, }} \
-                        x07, x06, x05, x04, x03, x02, x01, x00 }}
-
-#define mm512_const1_32( x ) {{ x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x }}
-
-#define mm512_const_16( x31, x30, x29, x28, x27, x26, x25, x24, \
-                        x23, x22, x21, x20, x19, x18, x17, x16, \
-                        x15, x14, x13, x12, x11, x10, x09, x08, \
-                        x07, x06, x05, x04, x03, x02, x01, x00 ) \
-                     {{ x31, x30, x29, x28, x27, x26, x25, x24, \
-                        x23, x22, x21, x20, x19, x18, x17, x16, \
-                        x15, x14, x13, x12, x11, x10, x09, x08, \
-                        x07, x06, x05, x04, x03, x02, x01, x00 }}
-
-#define mm512_const1_16( x ) {{ x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x, \
-                                x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x }}
-
-#define mm512_const_8( x63, x62, x61, x60, x59, x58, x57, x56, \
-                       x55, x54, x53, x52, x51, x50, x49, x48, \
-                       x47, x46, x45, x44, x43, x42, x41, x40, \
-                       x39, x38, x37, x36, x35, x34, x33, x32, \
-                       x31, x30, x29, x28, x27, x26, x25, x24, \
-                       x23, x22, x21, x20, x19, x18, x17, x16, \
-                       x15, x14, x13, x12, x11, x10, x09, x08, \
-                       x07, x06, x05, x04, x03, x02, x01, x00 ) \
-                    {{ x63, x62, x61, x60, x59, x58, x57, x56, \
-                       x55, x54, x53, x52, x51, x50, x49, x48, \
-                       x47, x46, x45, x44, x43, x42, x41, x40, \
-                       x39, x38, x37, x36, x35, x34, x33, x32, \
-                       x31, x30, x29, x28, x27, x26, x25, x24, \
-                       x23, x22, x21, x20, x19, x18, x17, x16, \
-                       x15, x14, x13, x12, x11, x10, x09, x08, \
-                       x07, x06, x05, x04, x03, x02, x01, x00 }}
-
-#define mm512_const1_8( x ) {{ x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x, \
-                               x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x, \
-                               x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x, \
-                               x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x }}
-
-// Predefined compile time constant vectors.
-#define c512_zero      mm512_const1_64(   0ULL )
-#define c512_neg1      mm512_const1_64(   0xFFFFFFFFFFFFFFFFULL )
-#define c512_one_512   mm512_const_epi64( 0ULL, 0ULL, 0ULL, 0ULL, \
-                                          0ULL, 0ULL, 0ULL, 1ULL )
-#define c512_one_256   mm512_const_64(    0ULL, 0ULL, 0ULL, 1ULL, \
-                                          0ULL, 0ULL, 0ULL, 1ULL )
-#define c512_one_128   mm512_const_64(    0ULL, 1ULL, 0ULL, 1ULL, \
-                                          0ULL, 1ULL, 0ULL, 1ULL )
-#define c512_one_64    mm512_const1_64(   1ULL )
-#define c512_one_32    mm512_const1_32(   1UL )
-#define c512_one_16    mm512_const1_16(   1U )
-#define c512_one_8     mm512_const1_8(    1U )
-#define c512_neg1_64   mm512_const1_64( 0xFFFFFFFFFFFFFFFFULL )
-#define c512_neg1_32   mm512_const1_32( 0xFFFFFFFFUL )
-#define c512_neg1_16   mm512_const1_32( 0xFFFFU )
-#define c512_neg1_8    mm512_const1_32( 0xFFU )
-
 //
 // Pseudo constants.
 
@@ -127,11 +54,77 @@
                                            0ULL, 0ULL, 0ULL, 1ULL )
 #define m512_one_256    _mm512_set4_epi64( 0ULL, 0ULL, 0ULL, 1ULL )
 #define m512_one_128    _mm512_set4_epi64( 0ULL, 1ULL, 0ULL, 1ULL )
-#define m512_one_64     _mm512_set1_epi64( 1ULL )
-#define m512_one_32     _mm512_set1_epi32( 1UL )
-#define m512_one_16     _mm512_set1_epi16( 1U )
-#define m512_one_8      _mm512_set1_epi8(  1U )
-#define m512_neg1       _mm512_set1_epi64( 0xFFFFFFFFFFFFFFFFULL )
+//#define m512_one_64     _mm512_set1_epi64( 1ULL )
+//#define m512_one_32     _mm512_set1_epi32( 1UL )
+//#define m512_one_16     _mm512_set1_epi16( 1U )
+//#define m512_one_8      _mm512_set1_epi8(  1U )
+//#define m512_neg1       _mm512_set1_epi64( 0xFFFFFFFFFFFFFFFFULL )
+
+#define mi512_const_64( i7, i6, i5, i4, i3, i2, i1, i0 ) \
+  _mm512_inserti64x4( _mm512_castsi512_si256( m256_const_64( i3.i2,i1,i0 ) ), \
+                      m256_const_64( i7,i6,i5,i4 ), 1 )
+#define m512_const1_64( i ) m256_const_64( i, i, i, i, i, i, i, i )
+
+static inline __m512i m512_one_64_fn()
+{
+  __m512i a;
+  asm( "vpxorq %0, %0, %0\n\t"
+       "vpcmpeqd %%zmm1, %%zmm1, %%zmm1\n\t"
+       "vpsubq %%zmm1, %0, %0\n\t"
+       :"=x"(a)
+       :
+       : "zmm1" );
+  return a;
+}
+#define m512_one_64    m512_one_64_fn()
+
+static inline __m512i m512_one_32_fn()
+{
+  __m512i a;
+  asm( "vpxord %0, %0, %0\n\t"
+       "vpcmpeqd %%zmm1, %%zmm1, %%zmm1\n\t"
+       "vpsubd %%zmm1, %0, %0\n\t"
+       :"=x"(a)
+       :
+       : "zmm1" );
+  return a;
+}
+#define m512_one_32    m512_one_32_fn()
+
+static inline __m512i m512_one_16_fn()
+{
+  __m512i a;
+  asm( "vpxord %0, %0, %0\n\t"
+       "vpcmpeqd %%zmm1, %%zmm1, %%zmm1\n\t"
+       "vpsubw %%zmm1, %0, %0\n\t"
+       :"=x"(a)
+       :
+       : "zmm1" );
+  return a;
+}
+#define m512_one_16    m512_one_16_fn()
+
+static inline __m512i m512_one_8_fn()
+{
+  __m512i a;
+  asm( "vpxord %0, %0, %0\n\t"
+       "vpcmpeqd %%zmm1, %%zmm1, %%zmm1\n\t"
+       "vpsubb %%zmm1, %0, %0\n\t"
+       :"=x"(a)
+       :
+       : "zmm1" );
+  return a;
+}
+#define m512_one_8    m512_one_8_fn()
+
+static inline __m512i m512_neg1_fn()
+{
+   __m512i a;
+   asm( "vpcmpeqq %0, %0, %0\n\t"
+        :"=x"(a) );
+   return a;
+}
+#define m512_neg1    m512_neg1_fn()
 
 
 //
@@ -142,6 +135,15 @@
 #define mm512_negate_32( x ) _mm512_sub_epi32( m512_zero, x )  
 #define mm512_negate_16( x ) _mm512_sub_epi16( m512_zero, x )  
 
+
+
+#define mm256_extr_lo256_512( a ) _mm512_castsi512_si256( a )
+#define mm256_extr_hi256_512( a ) _mm512_extracti64x4_epi64( a, 1 )
+
+#define mm128_extr_lo128_512( a ) _mm512_castsi512_si256( a )
+
+
+
 //
 // Pointer casting
 
@@ -225,6 +227,25 @@
    *((uint32_t*)(d06)) = ((uint32_t*)(s))[14]; \
    *((uint32_t*)(d07)) = ((uint32_t*)(s))[15];
 
+// Add 4 values, fewer dependencies than sequential addition.
+
+
+#define mm512_add4_64( a, b, c, d ) \
+   _mm512_add_epi64( _mm512_add_epi64( a, b ), _mm512_add_epi64( c, d ) )
+
+#define mm512_add4_32( a, b, c, d ) \
+   _mm512_add_epi32( _mm512_add_epi32( a, b ), _mm512_add_epi32( c, d ) )
+
+#define mm512_add4_16( a, b, c, d ) \
+   _mm512_add_epi16( _mm512_add_epi16( a, b ), _mm512_add_epi16( c, d ) )
+
+#define mm512_add4_8( a, b, c, d ) \
+   _mm512_add_epi8( _mm512_add_epi8( a, b ), _mm512_add_epi8( c, d ) )
+
+#define mm512_xor4( a, b, c, d ) \
+   _mm512_xor_si512( _mm512_xor_si256( a, b ), _mm512_xor_si256( c, d ) )
+
+
 
 //
 // Bit rotations.
@@ -246,68 +267,58 @@
 //
 // Rotate elements in 512 bit vector.
 
-#define mm512_swap_256( v ) \
-    _mm512_permutexvar_epi64( v, _mm512_set_epi64( 3,2,1,0,  7,6,5,4 ) )
+#define mm512_swap_256( v )        _mm512_alignr_epi64( v, v, 4 )
 
-#define mm512_ror_1x128( v ) \
-    _mm512_permutexvar_epi64( v, _mm512_set_epi64( 1,0,  7,6,  5,4,  3,2 ) )
+#define mm512_ror_1x128( v )       _mm512_alignr_epi64( v, v, 2 )
+#define mm512_rol_1x128( v )       _mm512_alignr_epi64( v, v, 6 )
 
-#define mm512_rol_1x128( v ) \
-    _mm512_permutexvar_epi64( v, _mm512_set_epi64( 5,4,  3,2,  1,0,  7,6 ) )
+#define mm512_ror_1x64( v )        _mm512_alignr_epi64( v, v, 1 )
+#define mm512_rol_1x64( v )        _mm512_alignr_epi64( v, v, 7 )
 
-#define mm512_ror_1x64( v ) \
-    _mm512_permutexvar_epi64( v, _mm512_set_epi64( 0,7,6,5,4,3,2,1 ) )
+#define mm512_ror_1x32( v )        _mm512_alignr_epi32( v, v, 1 )
+#define mm512_rol_1x32( v )        _mm512_alignr_epi32( v, v, 15 )
 
-#define mm512_rol_1x64( v ) \
-    _mm512_permutexvar_epi64( v, _mm512_set_epi64( 6,5,4,3,2,1,0,7 ) )
+// Generic for odd rotations
+#define mm512_ror_x64( v, n )      _mm512_alignr_epi64( v, v, n )
 
-#define mm512_ror_1x32( v ) \
-  _mm512_permutexvar_epi32( v, _mm512_set_epi32( \
-                      0,15,14,13,12,11,10, 9, 8, 7, 6, 5, 4, 3, 2, 1 ) )
+#define mm512_ror_x32( v, n )      _mm512_alignr_epi32( v, v, n )
 
-#define mm512_rol_1x32( v ) \
-  _mm512_permutexvar_epi32( v, _mm512_set_epi32( \
-                     14,13,12,11,10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0, 15 ) )
 
-//  Although documented to exist in AVX512F the _mm512_set_epi8 &
-//  _mm512_set_epi16 intrinsics fail to compile. Seems usefull to have
-//  for endian byte swapping. Workaround by using _mm512_set_epi32.
-//  Ugly but it works.
 
 #define mm512_ror_1x16( v ) \
-   _mm512_permutexvar_epi16( v, _mm512_set_epi32( \
-                       0x0000001F, 0x001E001D, 0x001C001B, 0x001A0019, \
-                       0X00180017, 0X00160015, 0X00140013, 0X00120011, \
-                       0X0010000F, 0X000E000D, 0X000C000B, 0X000A0009, \
-                       0X00080007, 0X00060005, 0X00040003, 0X00020001 ) )
+   _mm512_permutexvar_epi16( v, m512_const_64( \
+                       0x0000001F001E001D, 0x001C001B001A0019, \
+                       0X0018001700160015, 0X0014001300120011, \
+                       0X0010000F000E000D, 0X000C000B000A0009, \
+                       0X0008000700060005, 0X0004000300020001 ) )
 
 #define mm512_rol_1x16( v ) \
-   _mm512_permutexvar_epi16( v, _mm512_set_epi16( \
-                       0x001E001D, 0x001C001B, 0x001A0019, 0x00180017, \
-                       0X00160015, 0X00140013, 0X00120011, 0x0010000F, \
-                       0X000E000D, 0X000C000B, 0X000A0009, 0X00080007, \
-                       0X00060005, 0X00040003, 0X00020001, 0x0000001F ) )
+   _mm512_permutexvar_epi16( v, m512_const_64( \
+                       0x001E001D001C001B, 0x001A001900180017, \
+                       0X0016001500140013, 0X001200110010000F, \
+                       0X000E000D000C000B, 0X000A000900080007, \
+                       0X0006000500040003, 0X000200010000001F ) )
 
 
 #define mm512_ror_1x8( v ) \
-   _mm512_permutexvar_epi8( v, _mm512_set_epi8( \
-                       0x003F3E3D, 0x3C3B3A39, 0x38373635, 0x34333231, \
-                       0x302F2E2D, 0x2C2B2A29, 0x28272625, 0x24232221, \
-                       0x201F1E1D, 0x1C1B1A19. 0x18171615, 0x14131211, \
-                       0x100F0E0D, 0x0C0B0A09, 0x08070605, 0x04030201 ) )
+   _mm512_permutexvar_epi8( v, m512_const_64( \
+                       0x003F3E3D3C3B3A39, 0x3837363534333231, \
+                       0x302F2E2D2C2B2A29, 0x2827262524232221, \
+                       0x201F1E1D1C1B1A19. 0x1817161514131211, \
+                       0x100F0E0D0C0B0A09, 0x0807060504030201 ) )
 
 #define mm512_rol_1x8( v ) \
-   _mm512_permutexvar_epi8( v, _mm512_set_epi8( \
-                       0x3E3D3C3B, 0x3A393837, 0x36353433, 0x3231302F. \
-                       0x2E2D2C2B, 0x2A292827, 0x26252423, 0x2221201F, \
-                       0x1E1D1C1B, 0x1A191817, 0x16151413, 0x1211100F, \
-                       0x0E0D0C0B, 0x0A090807, 0x06050403, 0x0201003F ) )
+   _mm512_permutexvar_epi8( v, m512_const_64( \
+                       0x3E3D3C3B3A393837, 0x363534333231302F. \
+                       0x2E2D2C2B2A292827, 0x262524232221201F, \
+                       0x1E1D1C1B1A191817, 0x161514131211100F, \
+                       0x0E0D0C0B0A090807, 0x060504030201003F ) )
 
 // Invert vector: {3,2,1,0} -> {0,1,2,3}
 #define mm512_invert_128( v ) _mm512_permute4f128_epi32( a, 0x1b )
 
 #define mm512_invert_64( v ) \
-   _mm512_permutex_epi64( v, _mm512_set_epi64( 0,1,2,3,4,5,6,7 ) )
+   _mm512_permutex_epi64( v, m512_const_64( 0,1,2,3,4,5,6,7 ) )
 
 #define mm512_invert_32( v ) \
    _mm512_permutexvar_epi32( v, _mm512_set_epi32( \
@@ -384,32 +395,32 @@
 #define mm512_rol1x32_128( v )   _mm512_shuffle_epi32( v, 0x93 )
 
 #define mm512_ror1x16_128( v ) \
-    _mm512_permutexvar_epi16( v, _mm512_set_epi32( \
-                     0x0018001F, 0x001E001D, 0x001C001B, 0x001A0019, \
-                     0x00100017, 0x00160015, 0x00140013, 0x00120011, \
-                     0x0008000F, 0x000E000D, 0x000C000B, 0x000A0009, \
-                     0x00000007, 0x00060005, 0x00040003, 0x00020001 ) )
+    _mm512_permutexvar_epi16( v, m512_const_64( \
+                     0x0018001F001E001D, 0x001C001B001A0019, \
+                     0x0010001700160015, 0x0014001300120011, \
+                     0x0008000F000E000D, 0x000C000B000A0009, \
+                     0x0000000700060005, 0x0004000300020001 ) )
 
 #define mm512_rol1x16_128( v ) \
-    _mm512_permutexvar_epi16( v, _mm512_set_epi32( \
-                     0x001E001D, 0x001C001B, 0x001A0019, 0x0018001F, \
-                     0x00160015, 0x00140013, 0x00120011, 0x00100017, \
-                     0x000E000D, 0x000C000B, 0x000A0009, 0x0008000F, \
-                     0x00060005, 0x00040003, 0x00020001, 0x00000007 ) )
+    _mm512_permutexvar_epi16( v, m512_const_64( \
+                     0x001E001D001C001B, 0x001A00190018001F, \
+                     0x0016001500140013, 0x0012001100100017, \
+                     0x000E000D000C000B, 0x000A00090008000F, \
+                     0x0006000500040003, 0x0002000100000007 ) )
 
 #define mm512_ror1x8_128( v ) \
-    _mm512_permutexvar_epi8( v, _mm512_set_epi32( \
-                     0x303F3E3D, 0x3C3B3A39, 0x38373635, 0x34333231, \
-                     0x202F2E2D, 0x2C2B2A29, 0x28272625, 0x24232221, \
-                     0x101F1E1D, 0x1C1B1A19, 0x18171615, 0x14131211, \
-                     0x000F0E0D, 0x0C0B0A09, 0x08070605, 0x04030201 ) )
+    _mm512_permutexvar_epi8( v, m512_const_64( \
+                     0x303F3E3D3C3B3A39, 0x3837363534333231, \
+                     0x202F2E2D2C2B2A29, 0x2827262524232221, \
+                     0x101F1E1D1C1B1A19, 0x1817161514131211, \
+                     0x000F0E0D0C0B0A09, 0x0807060504030201 ) )
 
 #define mm512_rol1x8_128( v ) \
-    _mm512_permutexvar_epi8( v, _mm512_set_epi32( \
-                     0x3E3D3C3B, 0x3A393837, 0x36353433. 0x3231303F, \
-                     0x2E2D2C2B, 0x2A292827, 0x26252423, 0x2221202F, \
-                     0x1E1D1C1B, 0x1A191817, 0x16151413, 0x1211101F, \
-                     0x0E0D0C0B, 0x0A090807, 0x06050403, 0x0201000F ) )
+    _mm512_permutexvar_epi8( v, m512_const_64( \
+                     0x3E3D3C3B3A393837, 0x363534333231303F, \
+                     0x2E2D2C2B2A292827, 0x262524232221202F, \
+                     0x1E1D1C1B1A191817, 0x161514131211101F, \
+                     0x0E0D0C0B0A090807, 0x060504030201000F ) )
 
 // Rotate 128 bit lanes by c bytes.  
 #define mm512_bror_128( v, c ) \
@@ -601,4 +612,4 @@ do { \
 } while(0)
 
 #endif // AVX512
-#endif // SIMD_AVX512_H__
+#endif // SIMD_512_H__

simd-utils/simd-64.h

@@ -1,5 +1,5 @@
-#if !defined(SIMD_MMX_H__)
-#define SIMD_MMX_H__ 1
+#if !defined(SIMD_64_H__)
+#define SIMD_64_H__ 1
 
 #if defined(__MMX__)
 
@@ -13,21 +13,20 @@
 
 
 // Pseudo constants
+/*
 #define m64_zero   _mm_setzero_si64()
 #define m64_one_64 _mm_set_pi32(  0UL, 1UL )
 #define m64_one_32 _mm_set1_pi32( 1UL )
 #define m64_one_16 _mm_set1_pi16( 1U )
 #define m64_one_8  _mm_set1_pi8(  1U );
 #define m64_neg1   _mm_set1_pi32( 0xFFFFFFFFUL )
-/* cast also works, which is better?
+*/
 #define m64_zero   ( (__m64)0ULL )
 #define m64_one_64 ( (__m64)1ULL )
 #define m64_one_32 ( (__m64)0x0000000100000001ULL )
 #define m64_one_16 ( (__m64)0x0001000100010001ULL )
 #define m64_one_8  ( (__m64)0x0101010101010101ULL )
 #define m64_neg1   ( (__m64)0xFFFFFFFFFFFFFFFFULL )
-*/
-
 
 #define casti_m64(p,i) (((__m64*)(p))[(i)])
 
@@ -42,6 +41,14 @@
 #define mm64_negate_8(  v ) _mm_sub_pi8(  m64_zero, (__m64)v )
 
 // Rotate bits in packed elements of 64 bit vector
+#define mm64_rol_64( a, n ) \
+   _mm_or_si64( _mm_slli_si64( (__m64)(a), n ), \
+                _mm_srli_si64( (__m64)(a), 64-(n) ) )
+
+#define mm64_ror_64( a, n ) \
+   _mm_or_si64( _mm_srli_si64( (__m64)(a), n ), \
+                _mm_slli_si64( (__m64)(a), 64-(n) ) )
+
 #define mm64_rol_32( a, n ) \
    _mm_or_si64( _mm_slli_pi32( (__m64)(a), n ), \
                 _mm_srli_pi32( (__m64)(a), 32-(n) ) )
@@ -78,22 +85,20 @@
 // Endian byte swap packed elements
 // A vectorized version of the u64 bswap, use when data already in MMX reg.
 #define mm64_bswap_64( v ) \
-    _mm_shuffle_pi8( (__m64)v, _mm_set_pi8( 0,1,2,3,4,5,6,7 ) )
+    _mm_shuffle_pi8( (__m64)v, (__m64)0x0001020304050607 )
 
 #define mm64_bswap_32( v ) \
-    _mm_shuffle_pi8( (__m64)v, _mm_set_pi8( 4,5,6,7,  0,1,2,3 ) )
+    _mm_shuffle_pi8( (__m64)v, (__m64)0x0405060700010203 )
 
-/*
 #define mm64_bswap_16( v ) \
-    _mm_shuffle_pi8( (__m64)v, _mm_set_pi8( 6,7,  4,5,  2,3,  0,1 ) );
-*/
+    _mm_shuffle_pi8( (__m64)v, (__m64)0x0607040502030001 );
 
 #else
 
 #define mm64_bswap_64( v ) \
        (__m64)__builtin_bswap64( (uint64_t)v )
 
-// This exists only for compatibility with CPUs without SSSE3. MMX doesn't
+// These exist only for compatibility with CPUs without SSSE3. MMX doesn't
 // have extract 32 instruction so pointers are needed to access elements.
 // It' more efficient for the caller to use scalar variables and call
 // bswap_32 directly.
@@ -101,20 +106,11 @@
    _mm_set_pi32( __builtin_bswap32( ((uint32_t*)&v)[1] ), \
                  __builtin_bswap32( ((uint32_t*)&v)[0] )  )
 
-#endif
-
-// Invert vector: {3,2,1,0} -> {0,1,2,3}
-// Invert_64 is the same as bswap64
-// Invert_32 is the same as swap32
-
-#define mm64_invert_16( v ) _mm_shuffle_pi16( (__m64)v, 0x1b )
-
-#if defined(__SSSE3__)
-
-// An SSE2 or MMX version of this would be monstrous, shifting, masking and
-// oring each byte individually.
-#define mm64_invert_8(  v ) \
-    _mm_shuffle_pi8( (__m64)v, _mm_set_pi8( 0,1,2,3,4,5,6,7 ) );
+#define mm64_bswap_16( v ) \
+   _mm_set_pi16( __builtin_bswap16( ((uint16_t*)&v)[3] ), \
+                 __builtin_bswap16( ((uint16_t*)&v)[2] ), \
+                 __builtin_bswap16( ((uint16_t*)&v)[1] ), \
+                 __builtin_bswap16( ((uint16_t*)&v)[0] )  )
 
 #endif
 
@@ -131,5 +127,5 @@ static inline void memset_m64( __m64 *dst, const __m64 a,  int n )
 
 #endif // MMX
 
-#endif // SIMD_MMX_H__
+#endif // SIMD_64_H__
 

simd-utils/simd-avx.h

@@ -1,243 +0,0 @@
-#if !defined(SIMD_AVX_H__)
-#define SIMD_AVX_H__ 1
-
-#if defined(__AVX__)
-
-/////////////////////////////////////////////////////////////////////
-//
-//             AVX 256 bit vectors
-//
-//   Basic support for 256 bit vectors. Most of the good stuff needs AVX2.
-
-// Compile time vector constants and initializers.
-//
-// The following macro constants and functions should only be used
-// for compile time initialization of constant and variable vector
-// arrays. These constants use memory, use _mm256_set at run time to
-// avoid using memory.
-
-#define mm256_const_64( x3, x2, x1, x0 ) {{ x3, x2, x1, x0 }}
-#define mm256_const1_64( x ) {{ x,x,x,x }}
-
-#define mm256_const_32( x7, x6, x5, x4, x3, x2, x1, x0 ) \
-                     {{ x7, x6, x5, x4, x3, x2, x1, x0 }}
-#define mm256_const1_32( x ) {{ x,x,x,x, x,x,x,x }}
-
-#define mm256_const_16( x15, x14, x13, x12, x11, x10, x09, x08, \
-                        x07, x06, x05, x04, x03, x02, x01, x00 ) \
-                     {{ x15, x14, x13, x12, x11, x10, x09, x08, \
-                        x07, x06, x05, x04, x03, x02, x01, x00 }}
-#define mm256_const1_16( x ) {{ x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x }}
-
-#define mm256_const_8( x31, x30, x29, x28, x27, x26, x25, x24, \
-                       x23, x22, x21, x20, x19, x18, x17, x16, \
-                       x15, x14, x13, x12, x11, x10, x09, x08, \
-                       x07, x06, x05, x04, x03, x02, x01, x00 ) \
-                    {{ x31, x30, x29, x28, x27, x26, x25, x24, \
-                       x23, x22, x21, x20, x19, x18, x17, x16, \
-                       x15, x14, x13, x12, x11, x10, x09, x08, \
-                       x07, x06, x05, x04, x03, x02, x01, x00 }}
-#define mm256_const1_8( x ) {{ x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x, \
-                               x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x }}
-
-// Predefined compile time constant vectors.
-// Use Pseudo constants at run time for all simple constant vectors.
-#define c256_zero         mm256_const1_64( 0ULL )
-#define c256_one_256      mm256_const_64(  0ULL, 0ULL, 0ULL, 1ULL )
-#define c256_one_128      mm256_const_64(  0ULL, 1ULL, 0ULL, 1ULL )
-#define c256_one_64       mm256_const1_64( 1ULL )
-#define c256_one_32       mm256_const1_32( 1UL )
-#define c256_one_16       mm256_const1_16( 1U )
-#define c256_one_8        mm256_const1_8(  1U )
-#define c256_neg1         mm256_const1_64( 0xFFFFFFFFFFFFFFFFULL )
-#define c256_neg1_64      mm256_const1_64( 0xFFFFFFFFFFFFFFFFULL )
-#define c256_neg1_32      mm256_const1_32( 0xFFFFFFFFUL )
-#define c256_neg1_16      mm256_const1_16( 0xFFFFU )
-#define c256_neg1_8       mm256_const1_8(  0xFFU )
-
-//
-// Pseudo constants.
-// These can't be used for compile time initialization but are preferable
-// for simple constant vectors at run time.
-
-#define m256_zero            _mm256_setzero_si256()
-#define m256_one_256         _mm256_set_epi64x(  0ULL, 0ULL, 0ULL, 1ULL )
-#define m256_one_128         _mm256_set_epi64x(  0ULL, 1ULL, 0ULL, 1ULL )
-#define m256_one_64          _mm256_set1_epi64x( 1ULL )
-#define m256_one_32          _mm256_set1_epi32(  1UL )
-#define m256_one_16          _mm256_set1_epi16(  1U )
-#define m256_one_8           _mm256_set1_epi8(   1U )
-#define m256_neg1            _mm256_set1_epi64x( 0xFFFFFFFFFFFFFFFFULL )
-
-//
-// Vector size conversion.
-//
-// Allows operations on either or both halves of a 256 bit vector serially.
-// Handy for parallel AES.
-// Caveats:
-//      _mm256_castsi256_si128 is free and without side effects.
-//      _mm256_castsi128_si256 is also free but leaves the high half
-//      undefined. That's ok if the hi half will be subseqnently assigned.
-//      If assigning both, do lo first, If assigning only 1, use
-//      _mm256_inserti128_si256.
-//
-// What to do about extractf128 (AVX) and extracti128 (AVX2)?
-#define mm128_extr_lo128_256( a ) _mm256_castsi256_si128( a )
-#define mm128_extr_hi128_256( a ) _mm256_extractf128_si256( a, 1 )
-
-// Extract 4 u64 from 256 bit vector.
-#define mm256_extr_4x64( a0, a1, a2, a3, src ) \
-do { \
-  __m128i hi = _mm256_extractf128_si256( src, 1 ); \
-  a0 = _mm_extract_epi64( _mm256_castsi256_si128( src ), 0 ); \
-  a1 = _mm_extract_epi64( _mm256_castsi256_si128( src ), 1 ); \
-  a2 = _mm_extract_epi64( hi, 0 ); \
-  a3 = _mm_extract_epi64( hi, 1 ); \
-} while(0)
-
-#define mm256_extr_8x32( a0, a1, a2, a3, a4, a5, a6, a7, src ) \
-do { \
-  __m128i hi = _mm256_extractf128_si256( src, 1 ); \
-  a0 = _mm_extract_epi32( _mm256_castsi256_si128( src ), 0 ); \
-  a1 = _mm_extract_epi32( _mm256_castsi256_si128( src ), 1 ); \
-  a2 = _mm_extract_epi32( _mm256_castsi256_si128( src ), 2 ); \
-  a3 = _mm_extract_epi32( _mm256_castsi256_si128( src ), 3 ); \
-  a4 = _mm_extract_epi32( hi, 0 ); \
-  a5 = _mm_extract_epi32( hi, 1 ); \
-  a6 = _mm_extract_epi32( hi, 2 ); \
-  a7 = _mm_extract_epi32( hi, 3 ); \
-} while(0)
-
-// input __m128i, returns __m256i
-// To build a 256 bit vector from 2 128 bit vectors lo must be done first.
-// lo alone leaves hi undefined, hi alone leaves lo unchanged.
-// Both cost one clock while preserving the other half..
-// Insert b into specified half of a leaving other half of a unchanged.
-#define mm256_ins_lo128_256( a, b )  _mm256_insertf128_si256( a, b, 0 )
-#define mm256_ins_hi128_256( a, b )  _mm256_insertf128_si256( a, b, 1 )
-
-// concatenate two 128 bit vectors into one 256 bit vector: { hi, lo }
-#define mm256_concat_128( hi, lo ) \
-   mm256_ins_hi128_256( _mm256_castsi128_si256( lo ), hi )
-
-// Horizontal vector testing
-
-// Needs int128 support
-// Bit-wise test of entire vector, useful to test results of cmp.
-#define mm256_anybits0( a ) \
-         ( (uint128_t)mm128_extr_hi128_256( a ) \
-         | (uint128_t)mm128_extr_lo128_256( a ) )
-
-#define mm256_anybits1( a ) \
-         ( ( (uint128_t)mm128_extr_hi128_256( a ) + 1 ) \
-         | ( (uint128_t)mm128_extr_lo128_256( a ) + 1 ) )
-
-#define mm256_allbits0_256( a ) ( !mm256_anybits1(a) )
-#define mm256_allbits1_256( a ) ( !mm256_anybits0(a) )
-
-// Parallel AES, for when x is expected to be in a 256 bit register.
-#define mm256_aesenc_2x128( x ) \
-     mm256_concat_128( \
-     _mm_aesenc_si128( mm128_extr_hi128_256( x ), m128_zero ), \
-          _mm_aesenc_si128( mm128_extr_lo128_256( x ), m128_zero ) )
-
-#define mm256_aesenckey_2x128( x, k ) \
-     mm256_concat_128( \
-     _mm_aesenc_si128( mm128_extr_hi128_256( x ), \
-                       mm128_extr_lo128_256( k ) ), \
-     _mm_aesenc_si128( mm128_extr_hi128_256( x ), \
-                       mm128_extr_lo128_256( k ) ) )
-
-#define mm256_paesenc_2x128( y, x ) do \
-{ \
-  __m256i *X = (__m256i*)x; \
-  __m256i *Y = (__m256i*)y; \
-  y[0] = _mm_aesenc_si128( x[0], m128_zero ); \
-  y[1] = _mm_aesenc_si128( x[1], m128_zero ); \
-} while(0);
-
-// With pointers.
-#define mm256_paesenckey_2x128( y, x, k ) do \
-{ \
-  __m256i *X = (__m256i*)x; \
-  __m256i *Y = (__m256i*)y; \
-  __m256i *K = (__m256i*)ky; \
-  y[0] = _mm_aesenc_si128( x[0], K[0] ); \
-  y[1] = _mm_aesenc_si128( x[1], K[1] ); \
-} while(0);
-
-//
-// Pointer casting
-
-// p = any aligned pointer
-// returns p as pointer to vector type, not very useful
-#define castp_m256i(p) ((__m256i*)(p))
-
-// p = any aligned pointer
-// returns *p, watch your pointer arithmetic
-#define cast_m256i(p) (*((__m256i*)(p)))
-
-// p = any aligned pointer, i = scaled array index
-// returns value p[i]
-#define casti_m256i(p,i) (((__m256i*)(p))[(i)])
-
-// p = any aligned pointer, o = scaled offset
-// returns pointer p+o
-#define casto_m256i(p,o) (((__m256i*)(p))+(o))
-
-
-// Gather scatter
-
-#define mm256_gather_64( d, s0, s1, s2, s3 ) \
-    ((uint64_t*)(d))[0] = (uint64_t)(s0); \
-    ((uint64_t*)(d))[1] = (uint64_t)(s1); \
-    ((uint64_t*)(d))[2] = (uint64_t)(s2); \
-    ((uint64_t*)(d))[3] = (uint64_t)(s3);
-
-#define mm256_gather_32( d, s0, s1, s2, s3, s4, s5, s6, s7 ) \
-    ((uint32_t*)(d))[0] = (uint32_t)(s0); \
-    ((uint32_t*)(d))[1] = (uint32_t)(s1); \
-    ((uint32_t*)(d))[2] = (uint32_t)(s2); \
-    ((uint32_t*)(d))[3] = (uint32_t)(s3); \
-    ((uint32_t*)(d))[4] = (uint32_t)(s4); \
-    ((uint32_t*)(d))[5] = (uint32_t)(s5); \
-    ((uint32_t*)(d))[6] = (uint32_t)(s6); \
-    ((uint32_t*)(d))[7] = (uint32_t)(s7);
-
-
-// Scatter data from contiguous memory.
-// All arguments are pointers
-#define mm256_scatter_64( d0, d1, d2, d3, s ) \
-   *((uint64_t*)(d0)) = ((uint64_t*)(s))[0]; \
-   *((uint64_t*)(d1)) = ((uint64_t*)(s))[1]; \
-   *((uint64_t*)(d2)) = ((uint64_t*)(s))[2]; \
-   *((uint64_t*)(d3)) = ((uint64_t*)(s))[3];
-
-#define mm256_scatter_32( d0, d1, d2, d3, d4, d5, d6, d7, s ) \
-   *((uint32_t*)(d0)) = ((uint32_t*)(s))[0]; \
-   *((uint32_t*)(d1)) = ((uint32_t*)(s))[1]; \
-   *((uint32_t*)(d2)) = ((uint32_t*)(s))[2]; \
-   *((uint32_t*)(d3)) = ((uint32_t*)(s))[3]; \
-   *((uint32_t*)(d4)) = ((uint32_t*)(s))[4]; \
-   *((uint32_t*)(d5)) = ((uint32_t*)(s))[5]; \
-   *((uint32_t*)(d6)) = ((uint32_t*)(s))[6]; \
-   *((uint32_t*)(d7)) = ((uint32_t*)(s))[7];
-
-
-//
-// Memory functions
-// n = number of 256 bit (32 byte) vectors
-
-static inline void memset_zero_256( __m256i *dst, int n )
-{   for ( int i = 0; i < n; i++ ) dst[i] = m256_zero; }
-
-static inline void memset_256( __m256i *dst, const __m256i a,  int n )
-{   for ( int i = 0; i < n; i++ ) dst[i] = a; }
-
-static inline void memcpy_256( __m256i *dst, const __m256i *src, int n )
-{   for ( int i = 0; i < n; i ++ ) dst[i] = src[i]; }
-
-
-#endif // __AVX__
-#endif // SIMD_AVX_H__
-

simd-utils/simd-int.h

@@ -62,10 +62,16 @@ static inline void memset_64( uint64_t *dst, const uint64_t a,  int n )
 // 
 //      128 bit integers
 //
+//  128 bit integers are inneficient and not a shortcut for __m128i.
 
 // No real need or use.
 //#define u128_neg1        ((uint128_t)(-1))
 
+// usefull for making constants.
+#define mk_uint128( hi, lo ) \
+   ( ( (uint128_t)(hi) << 64 ) | ( (uint128_t)(lo) ) )
+
+
 // Extracting the low bits is a trivial cast.
 // These specialized functions are optimized while providing a
 // consistent interface.