v3.10.0

v3.9.11
v3.9.10
2025-09-17 23:44:27 +00:00 · 2019-12-03 12:26:11 -05:00 · 2019-11-26 09:22:03 -05:00 · 2019-11-22 20:29:18 -05:00
100 changed files with 12876 additions and 3513 deletions
--- a/14
+++ b/14
@@ -24,18 +24,10 @@ be installed manually. There may be others, read the error messages they
 will give a clue as to the missing package.

 The following command should install everything you need on Debian based
-distributions such as Ubuntu:
+distributions such as Ubuntu. Fedora and other distributions may have similar
+but different package names.

-sudo apt-get install build-essential libssl-dev libcurl4-openssl-dev libjansson-dev libgmp-dev automake zlib1g-dev
-
-build-essential  (Development Tools package group on Fedora)
-automake
-libjansson-dev
-libgmp-dev
-libcurl4-openssl-dev
-libssl-dev
-lib-thread
-zlib1g-dev
+sudo apt-get install build-essential libssl-dev libcurl4-openssl-dev libjansson-dev libgmp-dev zlib1g-dev

 SHA support on AMD Ryzen CPUs requires gcc version 5 or higher and
 openssl 1.1.0e or higher. Add one of the following, depending on the
--- a/5
+++ b/5
@@ -22,14 +22,13 @@ Step by step...

 Refer to Linux compile instructions and install required packages.

-Additionally, install mingw-64.
+Additionally, install mingw-w64.

 sudo apt-get install mingw-w64


 2. Create a local library directory for packages to be compiled in the next
-   step. Recommended location is $HOME/usr/lib/
-
+   step. Suggested location is $HOME/usr/lib/

 3. Download and build other packages for mingw that don't have a mingw64
   version available in the repositories.
--- a/Makefile.am
+++ b/Makefile.am
@@ -117,6 +117,7 @@ cpuminer_SOURCES = \
  algo/keccak/keccak-4way.c\
  algo/keccak/keccak-gate.c \
  algo/keccak/sse2/keccak.c \
+  algo/lanehash/lane.c \
  algo/luffa/sph_luffa.c \
  algo/luffa/luffa.c \
  algo/luffa/luffa_for_sse2.c \
@@ -173,7 +174,6 @@ cpuminer_SOURCES = \
  algo/sha/sph_sha2big.c \
  algo/sha/sha256-hash-4way.c \
  algo/sha/sha512-hash-4way.c \
-  algo/sha/sha256_hash_11way.c \
  algo/sha/sha2.c \
  algo/sha/sha256t-gate.c \
  algo/sha/sha256t-4way.c \
@@ -197,9 +197,9 @@ cpuminer_SOURCES = \
  algo/skein/skein-gate.c \
  algo/skein/skein2.c \
  algo/skein/skein2-4way.c \
-  algo/skein/skein2-gate.c \
  algo/sm3/sm3.c \
  algo/sm3/sm3-hash-4way.c \
+  algo/swifftx/swifftx.c \
  algo/tiger/sph_tiger.c \
  algo/whirlpool/sph_whirlpool.c \
  algo/whirlpool/whirlpool-hash-4way.c \
@@ -279,6 +279,11 @@ cpuminer_SOURCES = \
  algo/x17/sonoa-4way.c \
  algo/x17/sonoa.c \
  algo/x20/x20r.c \
+  algo/x22/x22i-4way.c \
+  algo/x22/x22i.c \
+  algo/x22/x22i-gate.c \
+  algo/x22/x25x.c \
+  algo/x22/x25x-4way.c \
  algo/yescrypt/yescrypt.c \
  algo/yescrypt/sha256_Y.c \
  algo/yescrypt/yescrypt-best.c \
--- a/README.md
+++ b/README.md
@@ -129,6 +129,8 @@ Supported Algorithms
                          x16s          Pigeoncoin (PGN)
                          x17
                          x21s
+                          x22i
+                          x25x
                          xevan         Bitsend (BSD)
                          yescrypt      Globalboost-Y (BSTY)
                          yescryptr8    BitZeny (ZNY)
--- a/README.txt
+++ b/README.txt
@@ -29,6 +29,7 @@ cpuminer-sse2.exe      "-msse2"                  Core2, Nehalem
 cpuminer-aes-sse42.exe "-march=westmere"         Westmere
 cpuminer-avx.exe       "-march=corei7-avx"       Sandy-Ivybridge
 cpuminer-avx2.exe      "-march=core-avx2"        Haswell, Sky-Kaby-Coffeelake
+cpuminer-avx512.exe    "-march=skylake-avx512"   Skylake-X, Cascadelake-X
 cpuminer-zen           "-march=znver1"           AMD Ryzen, Threadripper

 If you like this software feel free to donate:
--- a/30
+++ b/30
@@ -31,6 +31,36 @@ FreeBSD YMMV.
 Change Log
 ----------

+v3.10.0
+
+AVX-512 is now supported on selected algos, Windows binary is now available.
+AVX512 optimizations are available for argon2d, blake2s, keccak, keccakc,
+skein & skein2.
+
+Fixed CPU temperature for some CPU models (Linux only).
+
+Fixed a bug that caused some lanes not to submit shares.
+
+Fixed some previously undetected buffer overflows.
+
+Lyra2rev2 3% faster SSE2 and AVX2.
+
+Added "-fno-asynchronous-unwind-tables" to AVX512 build acript for Windows
+to fix known mingw issue.
+
+Changed AVX2 build script to explicitly add AES to address change in
+behaviour in GCC 9. 
+
+v3.9.11
+
+Added x22i & x25x algos.
+Blake2s 2% faster AVX2 with Intel CPU, slower with Ryzen v1, v2 ?
+
+v3.9.10
+
+Faster X* algos with AVX2.
+Small improvements to summary stats report.
+
 v3.9.9.1

 Fixed a day1 bug that could cause the miner to idle for up to 2 minutes
--- a/algo-gate-api.c
+++ b/algo-gate-api.c
@@ -238,6 +238,8 @@ bool register_algo_gate( int algo, algo_gate_t *gate )
    case ALGO_X16S:          register_x16s_algo          ( gate ); break;
    case ALGO_X17:           register_x17_algo           ( gate ); break;
    case ALGO_X21S:          register_x21s_algo          ( gate ); break;
+    case ALGO_X22I:          register_x22i_algo          ( gate ); break;
+    case ALGO_X25X:          register_x25x_algo          ( gate ); break;
    case ALGO_XEVAN:         register_xevan_algo         ( gate ); break;
 /*    case ALGO_YESCRYPT:     register_yescrypt_05_algo     ( gate ); break;
     case ALGO_YESCRYPTR8:   register_yescryptr8_05_algo   ( gate ); break;
--- a/algo-gate-api.h
+++ b/algo-gate-api.h
@@ -118,25 +118,55 @@ void ( *hash )     ( void*, const void*, uint32_t ) ;
 void ( *hash_suw ) ( void*, const void* );

 //optional, safe to use default in most cases
+
+// Allocate thread local buffers and other initialization specific to miner
+// threads.
 bool ( *miner_thread_init )      ( int );
+
+// Generate global blockheader from stratum data.
 void ( *stratum_gen_work )       ( struct stratum_ctx*, struct work* );
+
+// Get thread local copy of blockheader with unique nonce.
 void ( *get_new_work )           ( struct work*, struct work*, int, uint32_t*,
                                   bool );
+
+// Return pointer to nonce in blockheader.
 uint32_t *( *get_nonceptr )      ( uint32_t* );
-void ( *decode_extra_data )      ( struct work*, uint64_t* );
+
+// Decode getwork blockheader
 bool ( *work_decode )            ( const json_t*, struct work* );
+
+// Extra getwork data
+void ( *decode_extra_data )      ( struct work*, uint64_t* );
+
 bool ( *submit_getwork_result )  ( CURL*, struct work* );
+
 void ( *gen_merkle_root )        ( char*, struct stratum_ctx* );
+
+// Increment extranonce
 void ( *build_extraheader )      ( struct work*, struct stratum_ctx* );
+
 void ( *build_block_header )     ( struct work*, uint32_t, uint32_t*,
 	                                uint32_t*, uint32_t, uint32_t );
+// Build mining.submit message
 void ( *build_stratum_request )  ( char*, struct work*, struct stratum_ctx* );
+
 char* ( *malloc_txs_request )    ( struct work* );
+
+// Big or little
 void ( *set_work_data_endian )   ( struct work* );
+
 double ( *calc_network_diff )    ( struct work* );
+
+// Wait for first work
 bool ( *ready_to_mine )          ( struct work*, struct stratum_ctx*, int );
-void ( *resync_threads )         ( struct work* );
+
+// Diverge mining threads
 bool ( *do_this_thread )         ( int );
+
+// After do_this_thread
+void ( *resync_threads )         ( struct work* );
+
 json_t* (*longpoll_rpc_call)     ( CURL*, int*, char* );
 bool ( *stratum_handle_response )( json_t* );
 set_t optimizations;
--- a/algo/blake/blake-hash-4way.h
+++ b/algo/blake/blake-hash-4way.h
@@ -59,7 +59,6 @@ extern "C"{
 typedef struct {
   unsigned char buf[64<<2];
   uint32_t H[8<<2];
-   uint32_t S[4<<2];
 //   __m128i buf[16] __attribute__ ((aligned (64)));
 //   __m128i H[8];
 //   __m128i S[4];    
@@ -93,7 +92,6 @@ void blake256r8_4way_close(void *cc, void *dst);
 typedef struct {
   __m256i buf[16] __attribute__ ((aligned (64)));
   __m256i H[8];
-   __m256i S[4];
   size_t ptr;
   sph_u32 T0, T1;
   int rounds;   // 14 for blake, 8 for blakecoin & vanilla
--- a/algo/blake/blake256-hash-4way.c
+++ b/algo/blake/blake256-hash-4way.c
@@ -304,16 +304,17 @@ static const sph_u32 CS[16] = {

 #endif

+// Blake-256 4 way

 #define GS_4WAY( m0, m1, c0, c1, a, b, c, d ) \
 do { \
-   a = _mm_add_epi32( _mm_add_epi32( _mm_xor_si128( \
-                                   _mm_set1_epi32( c1 ), m0 ), b ), a ); \
+   a = _mm_add_epi32( _mm_add_epi32( a, b ), \
+                      _mm_xor_si128( _mm_set1_epi32( c1 ), m0 ) ); \
   d = mm128_ror_32( _mm_xor_si128( d, a ), 16 ); \
   c = _mm_add_epi32( c, d ); \
   b = mm128_ror_32( _mm_xor_si128( b, c ), 12 ); \
-   a = _mm_add_epi32( _mm_add_epi32( _mm_xor_si128( \
-                                   _mm_set1_epi32( c0 ), m1 ), b ), a ); \
+   a = _mm_add_epi32( _mm_add_epi32( a, b ), \
+                      _mm_xor_si128( _mm_set1_epi32( c0 ), m1 ) ); \
   d = mm128_ror_32( _mm_xor_si128( d, a ), 8 ); \
   c = _mm_add_epi32( c, d ); \
   b = mm128_ror_32( _mm_xor_si128( b, c ), 7 ); \
@@ -321,7 +322,8 @@ do { \

 #if SPH_COMPACT_BLAKE_32

-// Blake-256 4 way
+// Not used
+#if 0

 #define ROUND_S_4WAY(r)   do { \
 	GS_4WAY(M[sigma[r][0x0]], M[sigma[r][0x1]], \
@@ -342,6 +344,8 @@ do { \
 		CS[sigma[r][0xE]], CS[sigma[r][0xF]], V3, V4, V9, VE); \
 } while (0)

+#endif
+
 #else

 #define ROUND_S_4WAY(r)   do { \
@@ -359,7 +363,6 @@ do { \

 #define DECL_STATE32_4WAY \
 	__m128i H0, H1, H2, H3, H4, H5, H6, H7; \
-	__m128i S0, S1, S2, S3; \
        uint32_t T0, T1;

 #define READ_STATE32_4WAY(state)   do { \
@@ -371,10 +374,6 @@ do { \
 		H5 = casti_m128i( state->H, 5 ); \
 		H6 = casti_m128i( state->H, 6 ); \
 		H7 = casti_m128i( state->H, 7 ); \
-		S0 = casti_m128i( state->S, 0 ); \
-		S1 = casti_m128i( state->S, 1 ); \
-		S2 = casti_m128i( state->S, 2 ); \
-		S3 = casti_m128i( state->S, 3 ); \
 		T0 = (state)->T0; \
 		T1 = (state)->T1; \
 	} while (0)
@@ -388,17 +387,13 @@ do { \
 		casti_m128i( state->H, 5 ) = H5; \
 		casti_m128i( state->H, 6 ) = H6; \
 		casti_m128i( state->H, 7 ) = H7; \
-		casti_m128i( state->S, 0 ) = S0; \
-		casti_m128i( state->S, 1 ) = S1; \
-		casti_m128i( state->S, 2 ) = S2; \
-		casti_m128i( state->S, 3 ) = S3; \
 		(state)->T0 = T0; \
 		(state)->T1 = T1; \
 	} while (0)

 #if SPH_COMPACT_BLAKE_32
 // not used
-
+#if 0
 #define COMPRESS32_4WAY( rounds )   do { \
 	__m128i M[16]; \
 	__m128i V0, V1, V2, V3, V4, V5, V6, V7; \
@@ -441,6 +436,7 @@ do { \
        H7 = _mm_xor_si128( _mm_xor_si128( \
                                   _mm_xor_si128( S3, V7 ), VF ), H7 ); \
 	} while (0)
+#endif

 #else

@@ -508,10 +504,10 @@ do { \
   V5 = H5; \
   V6 = H6; \
   V7 = H7; \
-   V8 = _mm_xor_si128( S0, m128_const1_64( 0x243F6A88243F6A88 ) ); \
-   V9 = _mm_xor_si128( S1, m128_const1_64( 0x85A308D385A308D3 ) ); \
-   VA = _mm_xor_si128( S2, m128_const1_64( 0x13198A2E13198A2E ) ); \
-   VB = _mm_xor_si128( S3, m128_const1_64( 0x0370734403707344 ) ); \
+   V8 = m128_const1_64( 0x243F6A88243F6A88 ); \
+   V9 = m128_const1_64( 0x85A308D385A308D3 ); \
+   VA = m128_const1_64( 0x13198A2E13198A2E ); \
+   VB = m128_const1_64( 0x0370734403707344 ); \
   VC = _mm_xor_si128( _mm_set1_epi32( T0 ), \
                           m128_const1_64( 0xA4093822A4093822 ) ); \
   VD = _mm_xor_si128( _mm_set1_epi32( T0 ), \
@@ -538,14 +534,14 @@ do { \
      ROUND_S_4WAY(2); \
      ROUND_S_4WAY(3); \
   } \
-   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 ); \
+   H0 = _mm_xor_si128( _mm_xor_si128( V8, V0 ), H0 ); \
+   H1 = _mm_xor_si128( _mm_xor_si128( V9, V1 ), H1 ); \
+   H2 = _mm_xor_si128( _mm_xor_si128( VA, V2 ), H2 ); \
+   H3 = _mm_xor_si128( _mm_xor_si128( VB, V3 ), H3 ); \
+   H4 = _mm_xor_si128( _mm_xor_si128( VC, V4 ), H4 ); \
+   H5 = _mm_xor_si128( _mm_xor_si128( VD, V5 ), H5 ); \
+   H6 = _mm_xor_si128( _mm_xor_si128( VE, V6 ), H6 ); \
+   H7 = _mm_xor_si128( _mm_xor_si128( VF, V7 ), H7 ); \
 } while (0)

 #endif
@@ -556,13 +552,13 @@ do { \

 #define GS_8WAY( m0, m1, c0, c1, a, b, c, d ) \
 do { \
-   a = _mm256_add_epi32( _mm256_add_epi32( _mm256_xor_si256( \
-                 _mm256_set1_epi32( c1 ), m0 ), b ), a ); \
+   a = _mm256_add_epi32( _mm256_add_epi32( a, b ), \
+                         _mm256_xor_si256( _mm256_set1_epi32( c1 ), m0 ) ); \
   d = mm256_ror_32( _mm256_xor_si256( d, a ), 16 ); \
   c = _mm256_add_epi32( c, d ); \
   b = mm256_ror_32( _mm256_xor_si256( b, c ), 12 ); \
-   a = _mm256_add_epi32( _mm256_add_epi32( _mm256_xor_si256( \
-                 _mm256_set1_epi32( c0 ), m1 ), b ), a ); \
+   a = _mm256_add_epi32( _mm256_add_epi32( a, b ), \
+                         _mm256_xor_si256( _mm256_set1_epi32( c0 ), m1 ) ); \
   d = mm256_ror_32( _mm256_xor_si256( d, a ), 8 ); \
   c = _mm256_add_epi32( c, d ); \
   b = mm256_ror_32( _mm256_xor_si256( b, c ), 7 ); \
@@ -581,7 +577,6 @@ do { \

 #define DECL_STATE32_8WAY \
   __m256i H0, H1, H2, H3, H4, H5, H6, H7; \
-   __m256i S0, S1, S2, S3; \
   sph_u32 T0, T1;

 #define READ_STATE32_8WAY(state) \
@@ -594,10 +589,6 @@ do { \
   H5 = (state)->H[5]; \
   H6 = (state)->H[6]; \
   H7 = (state)->H[7]; \
-   S0 = (state)->S[0]; \
-   S1 = (state)->S[1]; \
-   S2 = (state)->S[2]; \
-   S3 = (state)->S[3]; \
   T0 = (state)->T0; \
   T1 = (state)->T1; \
 } while (0)
@@ -612,10 +603,6 @@ do { \
   (state)->H[5] = H5; \
   (state)->H[6] = H6; \
   (state)->H[7] = H7; \
-   (state)->S[0] = S0; \
-   (state)->S[1] = S1; \
-   (state)->S[2] = S2; \
-   (state)->S[3] = S3; \
   (state)->T0 = T0; \
   (state)->T1 = T1; \
 } while (0)
@@ -635,10 +622,10 @@ do { \
   V5 = H5; \
   V6 = H6; \
   V7 = H7; \
-   V8 = _mm256_xor_si256( S0, m256_const1_64( 0x243F6A88243F6A88 ) ); \
-   V9 = _mm256_xor_si256( S1, m256_const1_64( 0x85A308D385A308D3 ) ); \
-   VA = _mm256_xor_si256( S2, m256_const1_64( 0x13198A2E13198A2E ) ); \
-   VB = _mm256_xor_si256( S3, m256_const1_64( 0x0370734403707344 ) ); \
+   V8 = m256_const1_64( 0x243F6A88243F6A88 ); \
+   V9 = m256_const1_64( 0x85A308D385A308D3 ); \
+   VA = m256_const1_64( 0x13198A2E13198A2E ); \
+   VB = m256_const1_64( 0x0370734403707344 ); \
   VC = _mm256_xor_si256( _mm256_set1_epi32( T0 ),\
                              m256_const1_64( 0xA4093822A4093822 ) ); \
   VD = _mm256_xor_si256( _mm256_set1_epi32( T0 ),\
@@ -682,14 +669,14 @@ do { \
      ROUND_S_8WAY(2); \
      ROUND_S_8WAY(3); \
   } \
-   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 ); \
+   H0 = _mm256_xor_si256( _mm256_xor_si256( V8, V0 ), H0 ); \
+   H1 = _mm256_xor_si256( _mm256_xor_si256( V9, V1 ), H1 ); \
+   H2 = _mm256_xor_si256( _mm256_xor_si256( VA, V2 ), H2 ); \
+   H3 = _mm256_xor_si256( _mm256_xor_si256( VB, V3 ), H3 ); \
+   H4 = _mm256_xor_si256( _mm256_xor_si256( VC, V4 ), H4 ); \
+   H5 = _mm256_xor_si256( _mm256_xor_si256( VD, V5 ), H5 ); \
+   H6 = _mm256_xor_si256( _mm256_xor_si256( VE, V6 ), H6 ); \
+   H7 = _mm256_xor_si256( _mm256_xor_si256( VF, V7 ), H7 ); \
 } while (0)


@@ -703,7 +690,6 @@ 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 ) = m128_const1_64( 0x6A09E6676A09E667 );
   casti_m128i( ctx->H, 1 ) = m128_const1_64( 0xBB67AE85BB67AE85 );
   casti_m128i( ctx->H, 2 ) = m128_const1_64( 0x3C6EF3723C6EF372 );
@@ -712,11 +698,6 @@ blake32_4way_init( blake_4way_small_context *ctx, const uint32_t *iv,
   casti_m128i( ctx->H, 5 ) = m128_const1_64( 0x9B05688C9B05688C );
   casti_m128i( ctx->H, 6 ) = m128_const1_64( 0x1F83D9AB1F83D9AB );
   casti_m128i( ctx->H, 7 ) = m128_const1_64( 0x5BE0CD195BE0CD19 );
-
-   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;
@@ -824,7 +805,6 @@ static void
 blake32_8way_init( blake_8way_small_context *sc, const sph_u32 *iv,
                   const sph_u32 *salt, int rounds )
 {
-   __m256i zero = m256_zero;
   casti_m256i( sc->H, 0 ) = m256_const1_64( 0x6A09E6676A09E667 );
   casti_m256i( sc->H, 1 ) = m256_const1_64( 0xBB67AE85BB67AE85 );
   casti_m256i( sc->H, 2 ) = m256_const1_64( 0x3C6EF3723C6EF372 );
@@ -833,10 +813,6 @@ blake32_8way_init( blake_8way_small_context *sc, const sph_u32 *iv,
   casti_m256i( sc->H, 5 ) = m256_const1_64( 0x9B05688C9B05688C );
   casti_m256i( sc->H, 6 ) = m256_const1_64( 0x1F83D9AB1F83D9AB );
   casti_m256i( sc->H, 7 ) = m256_const1_64( 0x5BE0CD195BE0CD19 );
-   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;
--- a/algo/blake/blake2b-4way.c
+++ b/algo/blake/blake2b-4way.c
@@ -4,13 +4,59 @@
 */

 #include "blake2b-gate.h"
-
-#if defined(BLAKE2B_4WAY)
-
 #include <string.h>
 #include <stdint.h>
 #include "blake2b-hash-4way.h"

+#if defined(BLAKE2B_8WAY)
+
+int scanhash_blake2b_8way( struct work *work, uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t hash[8*8] __attribute__ ((aligned (128)));;
+   uint32_t vdata[20*8] __attribute__ ((aligned (64)));;
+   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
+   blake2b_8way_ctx ctx __attribute__ ((aligned (64)));
+   uint32_t *hash7 = &(hash[25]);   // 3*8+1
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   int thr_id = mythr->id;
+   __m512i  *noncev = (__m512i*)vdata + 9;   // aligned
+   const uint32_t Htarg = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+
+   uint32_t n = first_nonce;
+
+   mm512_bswap32_intrlv80_8x64( vdata, pdata );
+
+   do {
+      *noncev = mm512_intrlv_blend_32( mm512_bswap_32(
+                _mm512_set_epi32( n+7, 0, n+6, 0, n+5, 0, n+4, 0,
+                                  n+3, 0, n+2, 0, n+1, 0, n  , 0 ) ), *noncev );
+
+      blake2b_8way_init( &ctx );
+      blake2b_8way_update( &ctx, vdata, 80 );
+      blake2b_8way_final( &ctx, hash );
+
+      for ( int lane = 0; lane < 8; lane++ )
+      if ( hash7[ lane<<1 ] < Htarg )
+      {
+          extr_lane_8x64( lane_hash, hash, lane, 256 );
+          if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
+          {
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
+          }
+      }
+      n += 8;
+   } while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+#elif defined(BLAKE2B_4WAY)
+
 // Function not used, code inlined.
 void blake2b_4way_hash(void *output, const void *input)
 {
--- a/algo/blake/blake2b-gate.c
+++ b/algo/blake/blake2b-gate.c
@@ -1,15 +1,19 @@
 #include "blake2b-gate.h"

+
 bool register_blake2b_algo( algo_gate_t* gate )
 {
-#if defined(BLAKE2B_4WAY)
+#if defined(BLAKE2B_8WAY)
+  gate->scanhash  = (void*)&scanhash_blake2b_8way;
+//  gate->hash      = (void*)&blake2b_8way_hash;
+#elif defined(BLAKE2B_4WAY)
  gate->scanhash  = (void*)&scanhash_blake2b_4way;
  gate->hash      = (void*)&blake2b_4way_hash;
 #else
  gate->scanhash  = (void*)&scanhash_blake2b;
  gate->hash      = (void*)&blake2b_hash;
 #endif
-  gate->optimizations =  AVX2_OPT;
+  gate->optimizations =  AVX2_OPT | AVX512_OPT;
  return true;
 };

--- a/algo/blake/blake2b-gate.h
+++ b/algo/blake/blake2b-gate.h
@@ -4,13 +4,21 @@
 #include <stdint.h>
 #include "algo-gate-api.h"

-#if defined(__AVX2__)
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define BLAKE2B_8WAY
+#elif defined(__AVX2__)
  #define BLAKE2B_4WAY
 #endif

 bool register_blake2b_algo( algo_gate_t* gate );

-#if defined(BLAKE2B_4WAY)
+#if defined(BLAKE2B_8WAY)
+
+//void blake2b_8way_hash( void *state, const void *input );
+int scanhash_blake2b_8way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+
+#elif defined(BLAKE2B_4WAY)

 void blake2b_4way_hash( void *state, const void *input );
 int scanhash_blake2b_4way( struct work *work, uint32_t max_nonce,
--- a/algo/blake/blake2b-hash-4way.c
+++ b/algo/blake/blake2b-hash-4way.c
@@ -33,6 +33,178 @@

 #include "blake2b-hash-4way.h"

+static const uint8_t sigma[12][16] =
+{
+      { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
+      { 14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 },
+      { 11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4 },
+      { 7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8 },
+      { 9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13 },
+      { 2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9 },
+      { 12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11 },
+      { 13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10 },
+      { 6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5 },
+      { 10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13, 0 },
+      { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
+      { 14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 }
+};
+
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define B2B8W_G(a, b, c, d, x, y) \
+{ \
+   v[a] = _mm512_add_epi64( _mm512_add_epi64( v[a], v[b] ), x ); \
+   v[d] = mm512_ror_64( _mm512_xor_si512( v[d], v[a] ), 32 ); \
+   v[c] = _mm512_add_epi64( v[c], v[d] ); \
+   v[b] = mm512_ror_64( _mm512_xor_si512( v[b], v[c] ), 24 ); \
+   v[a] = _mm512_add_epi64( _mm512_add_epi64( v[a], v[b] ), y ); \
+   v[d] = mm512_ror_64( _mm512_xor_si512( v[d], v[a] ), 16 ); \
+   v[c] = _mm512_add_epi64( v[c], v[d] ); \
+   v[b] = mm512_ror_64( _mm512_xor_si512( v[b], v[c] ), 63 ); \
+}
+
+static void blake2b_8way_compress( blake2b_8way_ctx *ctx, int last )
+{  
+   __m512i v[16], m[16];
+
+   v[ 0] = ctx->h[0];
+   v[ 1] = ctx->h[1];
+   v[ 2] = ctx->h[2];
+   v[ 3] = ctx->h[3];
+   v[ 4] = ctx->h[4];
+   v[ 5] = ctx->h[5];
+   v[ 6] = ctx->h[6];
+   v[ 7] = ctx->h[7];
+   v[ 8] = m512_const1_64( 0x6A09E667F3BCC908 );
+   v[ 9] = m512_const1_64( 0xBB67AE8584CAA73B );
+   v[10] = m512_const1_64( 0x3C6EF372FE94F82B );
+   v[11] = m512_const1_64( 0xA54FF53A5F1D36F1 );
+   v[12] = m512_const1_64( 0x510E527FADE682D1 );
+   v[13] = m512_const1_64( 0x9B05688C2B3E6C1F );
+   v[14] = m512_const1_64( 0x1F83D9ABFB41BD6B );
+   v[15] = m512_const1_64( 0x5BE0CD19137E2179 );
+
+   v[12] = _mm512_xor_si512( v[12], _mm512_set1_epi64( ctx->t[0] ) );
+   v[13] = _mm512_xor_si512( v[13], _mm512_set1_epi64( ctx->t[1] ) );
+
+   if ( last )
+      v[14] = mm512_not( v[14] );
+
+   m[ 0] = ctx->b[ 0];
+   m[ 1] = ctx->b[ 1];
+   m[ 2] = ctx->b[ 2];
+   m[ 3] = ctx->b[ 3];
+   m[ 4] = ctx->b[ 4];
+   m[ 5] = ctx->b[ 5];
+   m[ 6] = ctx->b[ 6];
+   m[ 7] = ctx->b[ 7];
+   m[ 8] = ctx->b[ 8];
+   m[ 9] = ctx->b[ 9];
+   m[10] = ctx->b[10];
+   m[11] = ctx->b[11];
+   m[12] = ctx->b[12];
+   m[13] = ctx->b[13];
+   m[14] = ctx->b[14];
+   m[15] = ctx->b[15];
+
+   for ( int i = 0; i < 12; i++ )
+   {
+      B2B8W_G( 0, 4,  8, 12, m[ sigma[i][ 0] ], m[ sigma[i][ 1] ] );
+      B2B8W_G( 1, 5,  9, 13, m[ sigma[i][ 2] ], m[ sigma[i][ 3] ] );
+      B2B8W_G( 2, 6, 10, 14, m[ sigma[i][ 4] ], m[ sigma[i][ 5] ] );
+      B2B8W_G( 3, 7, 11, 15, m[ sigma[i][ 6] ], m[ sigma[i][ 7] ] );
+      B2B8W_G( 0, 5, 10, 15, m[ sigma[i][ 8] ], m[ sigma[i][ 9] ] );
+      B2B8W_G( 1, 6, 11, 12, m[ sigma[i][10] ], m[ sigma[i][11] ] );
+      B2B8W_G( 2, 7,  8, 13, m[ sigma[i][12] ], m[ sigma[i][13] ] );
+      B2B8W_G( 3, 4,  9, 14, m[ sigma[i][14] ], m[ sigma[i][15] ] );
+   }
+
+   ctx->h[0] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[0], v[0] ), v[ 8] );
+   ctx->h[1] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[1], v[1] ), v[ 9] );
+   ctx->h[2] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[2], v[2] ), v[10] );
+   ctx->h[3] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[3], v[3] ), v[11] );
+   ctx->h[4] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[4], v[4] ), v[12] );
+   ctx->h[5] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[5], v[5] ), v[13] );
+   ctx->h[6] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[6], v[6] ), v[14] );
+   ctx->h[7] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[7], v[7] ), v[15] );
+}
+
+int blake2b_8way_init( blake2b_8way_ctx *ctx )
+{
+   size_t i;
+
+   ctx->h[0] = m512_const1_64( 0x6A09E667F3BCC908 );
+   ctx->h[1] = m512_const1_64( 0xBB67AE8584CAA73B );
+   ctx->h[2] = m512_const1_64( 0x3C6EF372FE94F82B );
+   ctx->h[3] = m512_const1_64( 0xA54FF53A5F1D36F1 );
+   ctx->h[4] = m512_const1_64( 0x510E527FADE682D1 );
+   ctx->h[5] = m512_const1_64( 0x9B05688C2B3E6C1F );
+   ctx->h[6] = m512_const1_64( 0x1F83D9ABFB41BD6B );
+   ctx->h[7] = m512_const1_64( 0x5BE0CD19137E2179 );
+
+   ctx->h[0] = _mm512_xor_si512( ctx->h[0], m512_const1_64( 0x01010020 ) );
+
+   ctx->t[0] = 0;
+   ctx->t[1] = 0;
+   ctx->c = 0;
+   ctx->outlen = 32;
+
+   for ( i = 0; i < 16; i++ )
+     ctx->b[i] = m512_zero;
+
+   return 0;
+}
+
+
+void blake2b_8way_update( blake2b_8way_ctx *ctx, const void *input,
+                          size_t inlen )
+{
+   __m512i* in =(__m512i*)input;
+
+   size_t i, c;
+   c = ctx->c >> 3;
+
+   for ( i = 0; i < (inlen >> 3); i++ )
+   {
+      if ( ctx->c == 128 )
+      {
+         ctx->t[0] += ctx->c;
+         if ( ctx->t[0] < ctx->c )
+            ctx->t[1]++;
+         blake2b_8way_compress( ctx, 0 );
+         ctx->c = 0;
+      }
+      ctx->b[ c++ ] = in[i];
+      ctx->c += 8;
+   }
+}
+
+void blake2b_8way_final( blake2b_8way_ctx *ctx, void *out )
+{
+   size_t c;
+   c = ctx->c >> 3;
+
+   ctx->t[0] += ctx->c;
+   if ( ctx->t[0] < ctx->c )
+      ctx->t[1]++;
+
+   while ( ctx->c < 128 )
+   {
+      ctx->b[c++] = m512_zero;
+      ctx->c += 8;
+   }
+
+   blake2b_8way_compress( ctx, 1 );           // final block flag = 1
+
+   casti_m512i( out, 0 ) = ctx->h[0];
+   casti_m512i( out, 1 ) = ctx->h[1];
+   casti_m512i( out, 2 ) = ctx->h[2];
+   casti_m512i( out, 3 ) = ctx->h[3];
+}
+
+#endif
+
 #if defined(__AVX2__)

 // G Mixing function.
@@ -61,21 +233,6 @@ static const uint64_t blake2b_iv[8] = {

 static void blake2b_4way_compress( blake2b_4way_ctx *ctx, int last )
 {
-	const uint8_t sigma[12][16] = {
-		{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
-		{ 14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 },
-		{ 11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4 },
-		{ 7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8 },
-		{ 9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13 },
-		{ 2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9 },
-		{ 12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11 },
-		{ 13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10 },
-		{ 6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5 },
-		{ 10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13, 0 },
-		{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 },
-		{ 14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3 }
-	};
-	int i;
 	__m256i v[16], m[16];

   v[ 0] = ctx->h[0];
@@ -118,7 +275,7 @@ static void blake2b_4way_compress( blake2b_4way_ctx *ctx, int last )
   m[14] = ctx->b[14];
   m[15] = ctx->b[15];
   
-	for ( i = 0; i < 12; i++ )
+	for ( int i = 0; i < 12; i++ )
   { 
 		B2B_G( 0, 4,  8, 12, m[ sigma[i][ 0] ], m[ sigma[i][ 1] ] );
 		B2B_G( 1, 5,  9, 13, m[ sigma[i][ 2] ], m[ sigma[i][ 3] ] );
--- a/algo/blake/blake2b-hash-4way.h
+++ b/algo/blake/blake2b-hash-4way.h
@@ -2,8 +2,6 @@
 #ifndef __BLAKE2B_HASH_4WAY_H__
 #define __BLAKE2B_HASH_4WAY_H__

-#if defined(__AVX2__)
-
 #include "simd-utils.h"
 #include <stddef.h>
 #include <stdint.h>
@@ -16,14 +14,34 @@
 #define ALIGN(x) __attribute__((aligned(x)))
 #endif

+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+ALIGN(128) typedef struct {
+   __m512i b[16]; // input buffer
+   __m512i h[8];  // chained state
+   uint64_t t[2];  // total number of bytes
+   size_t c;       // pointer for b[]
+   size_t outlen;  // digest size
+} blake2b_8way_ctx;
+
+int blake2b_8way_init( blake2b_8way_ctx *ctx );
+void blake2b_8way_update( blake2b_8way_ctx *ctx, const void *input,
+                          size_t inlen );
+void blake2b_8way_final( blake2b_8way_ctx *ctx, void *out );
+
+#endif
+
+#if defined(__AVX2__)
+
 // state context
-ALIGN(64) typedef struct {
+ALIGN(128) typedef struct {
 	__m256i b[16]; // input buffer
 	__m256i h[8];  // chained state
 	uint64_t t[2];  // total number of bytes
 	size_t c;       // pointer for b[]
 	size_t outlen;  // digest size
-} blake2b_4way_ctx __attribute__((aligned(64)));
+} blake2b_4way_ctx;

 int blake2b_4way_init( blake2b_4way_ctx *ctx );
 void blake2b_4way_update( blake2b_4way_ctx *ctx, const void *input,
--- a/algo/blake/blake2s-4way.c
+++ b/algo/blake/blake2s-4way.c
@@ -3,22 +3,72 @@
 #include <string.h>
 #include <stdint.h>

-#if defined(BLAKE2S_8WAY)
+#if defined(BLAKE2S_16WAY)
+
+static __thread blake2s_16way_state blake2s_16w_ctx;
+
+void blake2s_16way_hash( void *output, const void *input )
+{
+   blake2s_16way_state ctx;
+   memcpy( &ctx, &blake2s_16w_ctx, sizeof ctx );
+   blake2s_16way_update( &ctx, input + (64<<4), 16 );
+   blake2s_16way_final( &ctx, output, BLAKE2S_OUTBYTES );
+}
+
+int scanhash_blake2s_16way( struct work *work, uint32_t max_nonce,
+                            uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t vdata[20*16] __attribute__ ((aligned (128)));
+   uint32_t hash[8*16] __attribute__ ((aligned (64)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
+   uint32_t *hash7 = &(hash[7<<4]);
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t Htarg = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   __m512i  *noncev = (__m512i*)vdata + 19;   // aligned
+   uint32_t n = first_nonce;
+   int thr_id = mythr->id;  
+
+   mm512_bswap32_intrlv80_16x32( vdata, pdata );
+   blake2s_16way_init( &blake2s_16w_ctx, BLAKE2S_OUTBYTES );
+   blake2s_16way_update( &blake2s_16w_ctx, vdata, 64 );
+
+   do {
+      *noncev = mm512_bswap_32( _mm512_set_epi32(
+	                  n+15, n+14, n+13, n+12, n+11, n+10, n+ 9, n+ 8,
+	                  n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n+ 1, n ) );
+      pdata[19] = n;
+
+      blake2s_16way_hash( hash, vdata );
+
+      for ( int lane = 0; lane < 16; lane++ )
+      if ( unlikely( hash7[lane] <= Htarg ) )
+      {
+         extr_lane_16x32( lane_hash, hash, lane, 256 );
+         if ( likely( fulltest( lane_hash, ptarget ) && !opt_benchmark ) )
+         {
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
+         }
+      }
+      n += 16;
+   } while ( (n < max_nonce-16) && !work_restart[thr_id].restart );
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+#elif defined(BLAKE2S_8WAY)

 static __thread blake2s_8way_state blake2s_8w_ctx;

 void blake2s_8way_hash( void *output, const void *input )
 {
-   uint32_t vhash[8*8] __attribute__ ((aligned (64)));
   blake2s_8way_state ctx;
   memcpy( &ctx, &blake2s_8w_ctx, sizeof ctx );
-
   blake2s_8way_update( &ctx, input + (64<<3), 16 );
-   blake2s_8way_final( &ctx, vhash, BLAKE2S_OUTBYTES );
-
-   dintrlv_8x32( output,     output+ 32, output+ 64, output+ 96,
-                 output+128, output+160, output+192, output+224,
-                 vhash, 256 );
+   blake2s_8way_final( &ctx, output, BLAKE2S_OUTBYTES );
 }

 int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
@@ -26,13 +76,15 @@ int scanhash_blake2s_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 lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash7 = &(hash[7<<3]);
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
   const uint32_t Htarg = ptarget[7];
   const uint32_t first_nonce = pdata[19];
   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
   uint32_t n = first_nonce;
-   int thr_id = mythr->id;  // thr_id arg is deprecated
+   int thr_id = mythr->id; 

   mm256_bswap32_intrlv80_8x32( vdata, pdata );
   blake2s_8way_init( &blake2s_8w_ctx, BLAKE2S_OUTBYTES );
@@ -45,16 +97,17 @@ int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,

      blake2s_8way_hash( hash, vdata );

-
-      for ( int i = 0; i < 8; i++ )
-      if (  (hash+(i<<3))[7] <= Htarg )
-      if ( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark )
+      for ( int lane = 0; lane < 8; lane++ )
+      if ( unlikely( hash7[lane] <= Htarg ) )
      {
-          pdata[19] = n+i;
-          submit_lane_solution( work, hash+(i<<3), mythr, i );
+         extr_lane_8x32( lane_hash, hash, lane, 256 );
+         if ( likely( fulltest( lane_hash, ptarget ) && !opt_benchmark ) )
+         {
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
+         }
      }
      n += 8;
-
   } while ( (n < max_nonce) && !work_restart[thr_id].restart );

   *hashes_done = n - first_nonce + 1;
@@ -67,15 +120,10 @@ static __thread blake2s_4way_state blake2s_4w_ctx;

 void blake2s_4way_hash( void *output, const void *input )
 {
-   uint32_t vhash[8*4] __attribute__ ((aligned (64)));
   blake2s_4way_state ctx;
   memcpy( &ctx, &blake2s_4w_ctx, sizeof ctx );
-
   blake2s_4way_update( &ctx, input + (64<<2), 16 );
-   blake2s_4way_final( &ctx, vhash, BLAKE2S_OUTBYTES );
-
-   dintrlv_4x32( output, output+32, output+64, output+96,
-		            vhash, 256 );
+   blake2s_4way_final( &ctx, output, BLAKE2S_OUTBYTES );
 }

 int scanhash_blake2s_4way( struct work *work, uint32_t max_nonce,
@@ -83,13 +131,15 @@ int scanhash_blake2s_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 lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash7 = &(hash[7<<2]);
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
   const uint32_t Htarg = ptarget[7];
   const uint32_t first_nonce = pdata[19];
   __m128i  *noncev = (__m128i*)vdata + 19;   // aligned
   uint32_t n = first_nonce;
-   int thr_id = mythr->id;  // thr_id arg is deprecated
+   int thr_id = mythr->id; 

   mm128_bswap32_intrlv80_4x32( vdata, pdata );
   blake2s_4way_init( &blake2s_4w_ctx, BLAKE2S_OUTBYTES );
@@ -101,15 +151,16 @@ int scanhash_blake2s_4way( struct work *work, uint32_t max_nonce,

      blake2s_4way_hash( hash, vdata );

-      for ( int i = 0; i < 4; i++ )
-      if ( (hash+(i<<3))[7] <= Htarg )
-      if ( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark )
+      for ( int lane = 0; lane < 4; lane++ ) if ( hash7[lane] <= Htarg )
      {
-          pdata[19] = n+i;
-          submit_lane_solution( work, hash+(i<<3), mythr, i );
+         extr_lane_4x32( lane_hash, hash, lane, 256 );
+         if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
+         {
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
+              }
      }
      n += 4;
-
   } while ( (n < max_nonce) && !work_restart[thr_id].restart );

   *hashes_done = n - first_nonce + 1;
--- a/algo/blake/blake2s-gate.c
+++ b/algo/blake/blake2s-gate.c
@@ -2,7 +2,11 @@

 bool register_blake2s_algo( algo_gate_t* gate )
 {
-#if defined(BLAKE2S_8WAY)
+#if defined(BLAKE2S_16WAY)
+  gate->scanhash  = (void*)&scanhash_blake2s_16way;
+  gate->hash      = (void*)&blake2s_16way_hash;
+#elif defined(BLAKE2S_8WAY)
+//#if defined(BLAKE2S_8WAY)
  gate->scanhash  = (void*)&scanhash_blake2s_8way;
  gate->hash      = (void*)&blake2s_8way_hash;
 #elif defined(BLAKE2S_4WAY)
@@ -12,7 +16,7 @@ bool register_blake2s_algo( algo_gate_t* gate )
  gate->scanhash  = (void*)&scanhash_blake2s;
  gate->hash      = (void*)&blake2s_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AVX2_OPT;
+  gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
  return true;
 };

--- a/algo/blake/blake2s-gate.h
+++ b/algo/blake/blake2s-gate.h
@@ -8,13 +8,26 @@
 #if defined(__SSE2__)
  #define BLAKE2S_4WAY
 #endif
+
 #if defined(__AVX2__)
  #define BLAKE2S_8WAY
 #endif

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define BLAKE2S_16WAY
+#endif
+
 bool register_blake2s_algo( algo_gate_t* gate );

-#if defined(BLAKE2S_8WAY)
+#if defined(BLAKE2S_16WAY)
+
+void blake2s_16way_hash( void *state, const void *input );
+int scanhash_blake2s_16way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+
+#elif defined (BLAKE2S_8WAY)
+
+//#if defined(BLAKE2S_8WAY)

 void blake2s_8way_hash( void *state, const void *input );
 int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
--- a/algo/blake/blake2s-hash-4way.c
+++ b/algo/blake/blake2s-hash-4way.c
@@ -20,12 +20,13 @@
 //#if defined(__SSE4_2__)
 #if defined(__SSE2__)

-
+/*
 static const uint32_t blake2s_IV[8] =
 {
 	0x6A09E667UL, 0xBB67AE85UL, 0x3C6EF372UL, 0xA54FF53AUL,
 	0x510E527FUL, 0x9B05688CUL, 0x1F83D9ABUL, 0x5BE0CD19UL
 };
+*/

 static const uint8_t blake2s_sigma[10][16] =
 {
@@ -41,6 +42,7 @@ static const uint8_t blake2s_sigma[10][16] =
 	{ 10,  2,  8,  4,  7,  6,  1,  5, 15, 11,  9, 14,  3, 12, 13 , 0 } ,
 };

+
 // define a constant for initial param.

 int blake2s_4way_init( blake2s_4way_state *S, const uint8_t outlen )
@@ -88,41 +90,45 @@ int blake2s_4way_compress( blake2s_4way_state *S, const __m128i* block )
   memcpy_128( m, block, 16 );
   memcpy_128( v, S->h, 8 );

-   v[ 8] = _mm_set1_epi32( blake2s_IV[0] );
-   v[ 9] = _mm_set1_epi32( blake2s_IV[1] );
-   v[10] = _mm_set1_epi32( blake2s_IV[2] );
-   v[11] = _mm_set1_epi32( blake2s_IV[3] );
+   v[ 8] = m128_const1_64( 0x6A09E6676A09E667ULL );
+   v[ 9] = m128_const1_64( 0xBB67AE85BB67AE85ULL );
+   v[10] = m128_const1_64( 0x3C6EF3723C6EF372ULL );
+   v[11] = m128_const1_64( 0xA54FF53AA54FF53AULL );
   v[12] = _mm_xor_si128( _mm_set1_epi32( S->t[0] ),
-                          _mm_set1_epi32( blake2s_IV[4] ) );
+                          m128_const1_64( 0x510E527F510E527FULL ) );
   v[13] = _mm_xor_si128( _mm_set1_epi32( S->t[1] ),
-                          _mm_set1_epi32( blake2s_IV[5] ) );
+                          m128_const1_64( 0x9B05688C9B05688CULL ) );
   v[14] = _mm_xor_si128( _mm_set1_epi32( S->f[0] ),
-                          _mm_set1_epi32( blake2s_IV[6] ) );
+                          m128_const1_64( 0x1F83D9AB1F83D9ABULL ) );
   v[15] = _mm_xor_si128( _mm_set1_epi32( S->f[1] ),
-                          _mm_set1_epi32( blake2s_IV[7] ) );
+                          m128_const1_64( 0x5BE0CD195BE0CD19ULL ) );

-#define G4W(r,i,a,b,c,d) \
+#define G4W( sigma0, sigma1, a, b, c, d ) \
 do { \
-   a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ blake2s_sigma[r][2*i+0] ] ); \
+   uint8_t s0 = sigma0; \
+   uint8_t s1 = sigma1; \
+   a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ s0 ] ); \
   d = mm128_ror_32( _mm_xor_si128( d, a ), 16 ); \
   c = _mm_add_epi32( c, d ); \
   b = mm128_ror_32( _mm_xor_si128( b, c ), 12 ); \
-   a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ blake2s_sigma[r][2*i+1] ] ); \
+   a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ s1 ] ); \
   d = mm128_ror_32( _mm_xor_si128( d, a ),  8 ); \
   c = _mm_add_epi32( c, d ); \
   b = mm128_ror_32( _mm_xor_si128( b, c ),  7 ); \
 } while(0)

+
 #define ROUND4W(r)  \
 do { \
-   G4W( r, 0, v[ 0], v[ 4], v[ 8], v[12] ); \
-   G4W( r, 1, v[ 1], v[ 5], v[ 9], v[13] ); \
-   G4W( r, 2, v[ 2], v[ 6], v[10], v[14] ); \
-   G4W( r, 3, v[ 3], v[ 7], v[11], v[15] ); \
-   G4W( r, 4, v[ 0], v[ 5], v[10], v[15] ); \
-   G4W( r, 5, v[ 1], v[ 6], v[11], v[12] ); \
-   G4W( r, 6, v[ 2], v[ 7], v[ 8], v[13] ); \
-   G4W( r, 7, v[ 3], v[ 4], v[ 9], v[14] ); \
+   uint8_t *sigma = (uint8_t*)&blake2s_sigma[r]; \
+   G4W( sigma[ 0], sigma[ 1], v[ 0], v[ 4], v[ 8], v[12] ); \
+   G4W( sigma[ 2], sigma[ 3], v[ 1], v[ 5], v[ 9], v[13] ); \
+   G4W( sigma[ 4], sigma[ 5], v[ 2], v[ 6], v[10], v[14] ); \
+   G4W( sigma[ 6], sigma[ 7], v[ 3], v[ 7], v[11], v[15] ); \
+   G4W( sigma[ 8], sigma[ 9], v[ 0], v[ 5], v[10], v[15] ); \
+   G4W( sigma[10], sigma[11], v[ 1], v[ 6], v[11], v[12] ); \
+   G4W( sigma[12], sigma[13], v[ 2], v[ 7], v[ 8], v[13] ); \
+   G4W( sigma[14], sigma[15], v[ 3], v[ 4], v[ 9], v[14] ); \
 } while(0)

   ROUND4W( 0 );
@@ -144,26 +150,47 @@ do { \
   return 0;
 }

+// There is a problem that can't be resolved internally.
+// If the last block is a full 64 bytes it should not be compressed in
+// update but left for final. However, when streaming, it isn't known
+// which block is last. There may be a subsequent call to update to add
+// more data.
+//
+// The reference code handled this by juggling 2 blocks at a time at
+// a significant performance penalty.
+//
+// Instead a new function is introduced called full_blocks which combines
+// update and final and is to be used in non-streaming mode where the data
+// is a multiple of 64 bytes.
+// 
+// Supported:
+//    64 + 16 bytes  (blake2s with midstate optimization)
+//    80 bytes       (blake2s without midstate optimization)
+//    Any multiple of 64 bytes in one shot (x25x)
+//
+// Unsupported:
+//    Stream of full 64 byte blocks one at a time.   
+
+// use only when streaming more data or final block not full.
 int blake2s_4way_update( blake2s_4way_state *S, const void *in,
                         uint64_t inlen )
 {
-  __m128i *input = (__m128i*)in;
-  __m128i *buf = (__m128i*)S->buf;
-  const int bsize = BLAKE2S_BLOCKBYTES;
+   __m128i *input = (__m128i*)in;
+   __m128i *buf = (__m128i*)S->buf;

   while( inlen > 0 )
   {
      size_t left = S->buflen;
-      if( inlen >= bsize - left )
+      if( inlen >= BLAKE2S_BLOCKBYTES - left )
      {
-         memcpy_128( buf + (left>>2), input, (bsize - left) >> 2 );
-         S->buflen += bsize - left;
+         memcpy_128( buf + (left>>2), input, (BLAKE2S_BLOCKBYTES - left) >> 2 );
+         S->buflen += BLAKE2S_BLOCKBYTES - left;
         S->t[0] += BLAKE2S_BLOCKBYTES;
         S->t[1] += ( S->t[0] < BLAKE2S_BLOCKBYTES );
         blake2s_4way_compress( S, buf ); 
         S->buflen = 0;
-         input += ( bsize >> 2 );
-         inlen -= bsize;
+         input += ( BLAKE2S_BLOCKBYTES >> 2 );
+         inlen -= BLAKE2S_BLOCKBYTES;
      }
      else
      {
@@ -195,8 +222,45 @@ int blake2s_4way_final( blake2s_4way_state *S, void *out, uint8_t outlen )
   return 0;
 }

+// Update and final when inlen is a multiple of 64 bytes
+int blake2s_4way_full_blocks( blake2s_4way_state *S, void *out,
+                              const void *input, uint64_t inlen )
+{
+    __m128i *in = (__m128i*)input;
+    __m128i *buf = (__m128i*)S->buf;
+
+    while( inlen > BLAKE2S_BLOCKBYTES )
+    {
+       memcpy_128( buf, in, BLAKE2S_BLOCKBYTES >> 2 );
+       S->buflen = BLAKE2S_BLOCKBYTES;
+       inlen -= BLAKE2S_BLOCKBYTES;
+       S->t[0] += BLAKE2S_BLOCKBYTES;
+       S->t[1] += ( S->t[0] < BLAKE2S_BLOCKBYTES );
+       blake2s_4way_compress( S, buf );
+       S->buflen = 0;
+       in += ( BLAKE2S_BLOCKBYTES >> 2 );
+    }
+
+    // last block
+    memcpy_128( buf, in, BLAKE2S_BLOCKBYTES >> 2 );
+    S->buflen = BLAKE2S_BLOCKBYTES;
+    S->t[0] += S->buflen;
+    S->t[1] += ( S->t[0] < S->buflen );
+    if ( S->last_node )  S->f[1] = ~0U;
+    S->f[0] = ~0U;
+    blake2s_4way_compress( S, buf );
+
+    for ( int i = 0; i < 8; ++i )
+      casti_m128i( out, i ) = S->h[ i ];
+    return 0;
+}
+
 #if defined(__AVX2__)

+// The commented code below is slower on Intel but faster on
+// Zen1 AVX2. It's also faster than Zen1 AVX.
+// Ryzen gen2 is unknown at this time.
+ 
 int blake2s_8way_compress( blake2s_8way_state *S, const __m256i *block )
 {
   __m256i m[16];
@@ -205,6 +269,23 @@ int blake2s_8way_compress( blake2s_8way_state *S, const __m256i *block )
   memcpy_256( m, block, 16 );
   memcpy_256( v, S->h, 8 );

+   v[ 8] = m256_const1_64( 0x6A09E6676A09E667ULL );
+   v[ 9] = m256_const1_64( 0xBB67AE85BB67AE85ULL );
+   v[10] = m256_const1_64( 0x3C6EF3723C6EF372ULL );
+   v[11] = m256_const1_64( 0xA54FF53AA54FF53AULL );
+   v[12] = _mm256_xor_si256( _mm256_set1_epi32( S->t[0] ),
+                          m256_const1_64( 0x510E527F510E527FULL ) );
+
+   v[13] = _mm256_xor_si256( _mm256_set1_epi32( S->t[1] ),
+                          m256_const1_64( 0x9B05688C9B05688CULL ) );
+
+   v[14] = _mm256_xor_si256( _mm256_set1_epi32( S->f[0] ),
+                          m256_const1_64( 0x1F83D9AB1F83D9ABULL ) );
+
+   v[15] = _mm256_xor_si256( _mm256_set1_epi32( S->f[1] ),
+                          m256_const1_64( 0x5BE0CD195BE0CD19ULL ) );
+
+/*
   v[ 8] = _mm256_set1_epi32( blake2s_IV[0] );
   v[ 9] = _mm256_set1_epi32( blake2s_IV[1] );
   v[10] = _mm256_set1_epi32( blake2s_IV[2] );
@@ -218,6 +299,7 @@ int blake2s_8way_compress( blake2s_8way_state *S, const __m256i *block )
   v[15] = _mm256_xor_si256( _mm256_set1_epi32( S->f[1] ),
                             _mm256_set1_epi32( blake2s_IV[7] ) );

+
 #define G8W(r,i,a,b,c,d) \
 do { \
   a = _mm256_add_epi32( _mm256_add_epi32( a, b ), \
@@ -231,7 +313,36 @@ do { \
   c = _mm256_add_epi32( c, d ); \
   b = mm256_ror_32( _mm256_xor_si256( b, c ),  7 ); \
 } while(0)
+*/

+#define G8W( sigma0, sigma1, a, b, c, d) \
+do { \
+   uint8_t s0 = sigma0; \
+   uint8_t s1 = sigma1; \
+   a = _mm256_add_epi32( _mm256_add_epi32( a, b ), m[ s0 ] ); \
+   d = mm256_ror_32( _mm256_xor_si256( d, a ), 16 ); \
+   c = _mm256_add_epi32( c, d ); \
+   b = mm256_ror_32( _mm256_xor_si256( b, c ), 12 ); \
+   a = _mm256_add_epi32( _mm256_add_epi32( a, b ), m[ s1 ] ); \
+   d = mm256_ror_32( _mm256_xor_si256( d, a ),  8 ); \
+   c = _mm256_add_epi32( c, d ); \
+   b = mm256_ror_32( _mm256_xor_si256( b, c ),  7 ); \
+} while(0)
+
+#define ROUND8W(r)  \
+do { \
+   uint8_t *sigma = (uint8_t*)&blake2s_sigma[r]; \
+   G8W( sigma[ 0], sigma[ 1], v[ 0], v[ 4], v[ 8], v[12] ); \
+   G8W( sigma[ 2], sigma[ 3], v[ 1], v[ 5], v[ 9], v[13] ); \
+   G8W( sigma[ 4], sigma[ 5], v[ 2], v[ 6], v[10], v[14] ); \
+   G8W( sigma[ 6], sigma[ 7], v[ 3], v[ 7], v[11], v[15] ); \
+   G8W( sigma[ 8], sigma[ 9], v[ 0], v[ 5], v[10], v[15] ); \
+   G8W( sigma[10], sigma[11], v[ 1], v[ 6], v[11], v[12] ); \
+   G8W( sigma[12], sigma[13], v[ 2], v[ 7], v[ 8], v[13] ); \
+   G8W( sigma[14], sigma[15], v[ 3], v[ 4], v[ 9], v[14] ); \
+} while(0)
+
+/*
 #define ROUND8W(r)  \
 do { \
   G8W( r, 0, v[ 0], v[ 4], v[ 8], v[12] ); \
@@ -243,6 +354,7 @@ do { \
   G8W( r, 6, v[ 2], v[ 7], v[ 8], v[13] ); \
   G8W( r, 7, v[ 3], v[ 4], v[ 9], v[14] ); \
 } while(0)
+*/

   ROUND8W( 0 );
   ROUND8W( 1 );
@@ -354,6 +466,168 @@ int blake2s_8way_final( blake2s_8way_state *S, void *out, uint8_t outlen )

 #endif // __AVX2__

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// Blake2s-256 16 way
+
+int blake2s_16way_compress( blake2s_16way_state *S, const __m512i *block )
+{
+   __m512i m[16];
+   __m512i v[16];
+
+   memcpy_512( m, block, 16 );
+   memcpy_512( v, S->h, 8 );
+
+   v[ 8] = m512_const1_64( 0x6A09E6676A09E667ULL );
+   v[ 9] = m512_const1_64( 0xBB67AE85BB67AE85ULL );
+   v[10] = m512_const1_64( 0x3C6EF3723C6EF372ULL );
+   v[11] = m512_const1_64( 0xA54FF53AA54FF53AULL );
+   v[12] = _mm512_xor_si512( _mm512_set1_epi32( S->t[0] ),
+                          m512_const1_64( 0x510E527F510E527FULL ) );
+
+   v[13] = _mm512_xor_si512( _mm512_set1_epi32( S->t[1] ),
+                          m512_const1_64( 0x9B05688C9B05688CULL ) );
+
+   v[14] = _mm512_xor_si512( _mm512_set1_epi32( S->f[0] ),
+                          m512_const1_64( 0x1F83D9AB1F83D9ABULL ) );
+
+   v[15] = _mm512_xor_si512( _mm512_set1_epi32( S->f[1] ),
+                          m512_const1_64( 0x5BE0CD195BE0CD19ULL ) );
+
+
+#define G16W( sigma0, sigma1, a, b, c, d) \
+do { \
+   uint8_t s0 = sigma0; \
+   uint8_t s1 = sigma1; \
+   a = _mm512_add_epi32( _mm512_add_epi32( a, b ), m[ s0 ] ); \
+   d = mm512_ror_32( _mm512_xor_si512( d, a ), 16 ); \
+   c = _mm512_add_epi32( c, d ); \
+   b = mm512_ror_32( _mm512_xor_si512( b, c ), 12 ); \
+   a = _mm512_add_epi32( _mm512_add_epi32( a, b ), m[ s1 ] ); \
+   d = mm512_ror_32( _mm512_xor_si512( d, a ),  8 ); \
+   c = _mm512_add_epi32( c, d ); \
+   b = mm512_ror_32( _mm512_xor_si512( b, c ),  7 ); \
+} while(0)
+
+#define ROUND16W(r)  \
+do { \
+   uint8_t *sigma = (uint8_t*)&blake2s_sigma[r]; \
+   G16W( sigma[ 0], sigma[ 1], v[ 0], v[ 4], v[ 8], v[12] ); \
+   G16W( sigma[ 2], sigma[ 3], v[ 1], v[ 5], v[ 9], v[13] ); \
+   G16W( sigma[ 4], sigma[ 5], v[ 2], v[ 6], v[10], v[14] ); \
+   G16W( sigma[ 6], sigma[ 7], v[ 3], v[ 7], v[11], v[15] ); \
+   G16W( sigma[ 8], sigma[ 9], v[ 0], v[ 5], v[10], v[15] ); \
+   G16W( sigma[10], sigma[11], v[ 1], v[ 6], v[11], v[12] ); \
+   G16W( sigma[12], sigma[13], v[ 2], v[ 7], v[ 8], v[13] ); \
+   G16W( sigma[14], sigma[15], v[ 3], v[ 4], v[ 9], v[14] ); \
+} while(0)
+
+   ROUND16W( 0 );
+   ROUND16W( 1 );
+   ROUND16W( 2 );
+   ROUND16W( 3 );
+   ROUND16W( 4 );
+   ROUND16W( 5 );
+   ROUND16W( 6 );
+   ROUND16W( 7 );
+   ROUND16W( 8 );
+   ROUND16W( 9 );
+
+   for( size_t i = 0; i < 8; ++i )
+      S->h[i] = _mm512_xor_si512( _mm512_xor_si512( S->h[i], v[i] ), v[i + 8] );
+
+#undef G16W
+#undef ROUND16W
+   return 0;
+}
+
+int blake2s_16way_init( blake2s_16way_state *S, const uint8_t outlen )
+{
+   blake2s_nway_param P[1];
+
+   P->digest_length = outlen;
+   P->key_length    = 0;
+   P->fanout        = 1;
+   P->depth         = 1;
+   P->leaf_length   = 0;
+   *((uint64_t*)(P->node_offset)) = 0;
+   P->node_depth    = 0;
+   P->inner_length  = 0;
+   memset( P->salt,     0, sizeof( P->salt ) );
+   memset( P->personal, 0, sizeof( P->personal ) );
+
+   memset( S, 0, sizeof( blake2s_16way_state ) );
+   S->h[0] = m512_const1_64( 0x6A09E6676A09E667ULL );
+   S->h[1] = m512_const1_64( 0xBB67AE85BB67AE85ULL );
+   S->h[2] = m512_const1_64( 0x3C6EF3723C6EF372ULL );
+   S->h[3] = m512_const1_64( 0xA54FF53AA54FF53AULL );
+   S->h[4] = m512_const1_64( 0x510E527F510E527FULL );
+   S->h[5] = m512_const1_64( 0x9B05688C9B05688CULL );
+   S->h[6] = m512_const1_64( 0x1F83D9AB1F83D9ABULL );
+   S->h[7] = m512_const1_64( 0x5BE0CD195BE0CD19ULL );
+
+   uint32_t *p = ( uint32_t * )( P );
+
+   /* IV XOR ParamBlock */
+   for ( size_t i = 0; i < 8; ++i )
+      S->h[i] = _mm512_xor_si512( S->h[i], _mm512_set1_epi32( p[i] ) );
+   return 0;
+}
+
+int blake2s_16way_update( blake2s_16way_state *S, const void *in,
+                         uint64_t inlen )
+{
+  __m512i *input = (__m512i*)in;
+  __m512i *buf = (__m512i*)S->buf;
+  const int bsize = BLAKE2S_BLOCKBYTES;
+
+   while( inlen > 0 )
+   {
+      size_t left = S->buflen;
+      if( inlen >= bsize - left )
+      {
+         memcpy_512( buf + (left>>2), input, (bsize - left) >> 2 );
+         S->buflen += bsize - left;
+         S->t[0] += BLAKE2S_BLOCKBYTES;
+         S->t[1] += ( S->t[0] < BLAKE2S_BLOCKBYTES );
+         blake2s_16way_compress( S, buf );
+         S->buflen = 0;
+         input += ( bsize >> 2 );
+         inlen -= bsize;
+      }
+      else
+      {
+          memcpy_512( buf + ( left>>2 ), input, inlen>>2 );
+          S->buflen += (size_t) inlen;
+          input += ( inlen>>2 );
+          inlen -= inlen;
+      }
+   }
+   return 0;
+}
+
+int blake2s_16way_final( blake2s_16way_state *S, void *out, uint8_t outlen )
+{
+   __m512i *buf = (__m512i*)S->buf;
+
+   S->t[0] += S->buflen;
+   S->t[1] += ( S->t[0] < S->buflen );
+   if ( S->last_node )
+      S->f[1] = ~0U;
+   S->f[0] = ~0U;
+
+   memset_zero_512( buf + ( S->buflen>>2 ),
+                    ( BLAKE2S_BLOCKBYTES - S->buflen ) >> 2 );
+   blake2s_16way_compress( S, buf );
+
+   for ( int i = 0; i < 8; ++i )
+      casti_m512i( out, i ) = S->h[ i ];
+   return 0;
+}
+
+#endif   // AVX512
+
+
 #if 0
 int blake2s( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen )
 {
--- a/algo/blake/blake2s-hash-4way.h
+++ b/algo/blake/blake2s-hash-4way.h
@@ -75,6 +75,9 @@ int blake2s_4way_init( blake2s_4way_state *S, const uint8_t outlen );
 int blake2s_4way_update( blake2s_4way_state *S, const void *in,
                         uint64_t inlen );
 int blake2s_4way_final( blake2s_4way_state *S, void *out, uint8_t outlen );
+int blake2s_4way_full_blocks( blake2s_4way_state *S, void *out,
+                              const void *input, uint64_t inlen );
+

 #if defined(__AVX2__)

@@ -92,6 +95,27 @@ int blake2s_8way_init( blake2s_8way_state *S, const uint8_t outlen );
 int blake2s_8way_update( blake2s_8way_state *S, const void *in,
                         uint64_t inlen );
 int blake2s_8way_final( blake2s_8way_state *S, void *out, uint8_t outlen );
+//int blake2s_8way_full_blocks( blake2s_8way_state *S, void *out,
+//                              const void *input, uint64_t inlen );
+
+#endif
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+ALIGN( 128 ) typedef struct __blake2s_16way_state
+{
+   __m512i h[8];
+   uint8_t  buf[ BLAKE2S_BLOCKBYTES * 16 ];
+   uint32_t t[2];
+   uint32_t f[2];
+   size_t   buflen;
+   uint8_t  last_node;
+} blake2s_16way_state ;
+
+int blake2s_16way_init( blake2s_16way_state *S, const uint8_t outlen );
+int blake2s_16way_update( blake2s_16way_state *S, const void *in,
+                         uint64_t inlen );
+int blake2s_16way_final( blake2s_16way_state *S, void *out, uint8_t outlen );

 #endif

--- a/algo/blake/blake512-hash-4way.c
+++ b/algo/blake/blake512-hash-4way.c
@@ -403,7 +403,9 @@ static const sph_u64 CB[16] = {
 	__m256i M[16]; \
 	__m256i V0, V1, V2, V3, V4, V5, V6, V7; \
 	__m256i V8, V9, VA, VB, VC, VD, VE, VF; \
-	unsigned r; \
+   const __m256i shuff_bswap64 = m256_const2_64( 0x08090a0b0c0d0e0f, \
+                                                 0x0001020304050607 ) \
+   unsigned r; \
 	V0 = H0; \
 	V1 = H1; \
 	V2 = H2; \
@@ -412,53 +414,53 @@ 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 ) ); \
-	M[0x0] = mm256_bswap_64( *(buf+0) ); \
-	M[0x1] = mm256_bswap_64( *(buf+1) ); \
-	M[0x2] = mm256_bswap_64( *(buf+2) ); \
-	M[0x3] = mm256_bswap_64( *(buf+3) ); \
-	M[0x4] = mm256_bswap_64( *(buf+4) ); \
-	M[0x5] = mm256_bswap_64( *(buf+5) ); \
-	M[0x6] = mm256_bswap_64( *(buf+6) ); \
-	M[0x7] = mm256_bswap_64( *(buf+7) ); \
-	M[0x8] = mm256_bswap_64( *(buf+8) ); \
-	M[0x9] = mm256_bswap_64( *(buf+9) ); \
-	M[0xA] = mm256_bswap_64( *(buf+10) ); \
-	M[0xB] = mm256_bswap_64( *(buf+11) ); \
-	M[0xC] = mm256_bswap_64( *(buf+12) ); \
-	M[0xD] = mm256_bswap_64( *(buf+13) ); \
-	M[0xE] = mm256_bswap_64( *(buf+14) ); \
-	M[0xF] = mm256_bswap_64( *(buf+15) ); \
+   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, CB7, CB7, CB7 ) ); \
+   M[0x0] = _mm256_shuffle_epi8( *(buf+ 0), shuff_bswap64 ); \
+	M[0x1] = _mm256_shuffle_epi8( *(buf+ 1), shuff_bswap64 ); \
+	M[0x2] = _mm256_shuffle_epi8( *(buf+ 2), shuff_bswap64 ); \
+	M[0x3] = _mm256_shuffle_epi8( *(buf+ 3), shuff_bswap64 ); \
+	M[0x4] = _mm256_shuffle_epi8( *(buf+ 4), shuff_bswap64 ); \
+	M[0x5] = _mm256_shuffle_epi8( *(buf+ 5), shuff_bswap64 ); \
+	M[0x6] = _mm256_shuffle_epi8( *(buf+ 6), shuff_bswap64 ); \
+	M[0x7] = _mm256_shuffle_epi8( *(buf+ 7), shuff_bswap64 ); \
+	M[0x8] = _mm256_shuffle_epi8( *(buf+ 8), shuff_bswap64 ); \
+	M[0x9] = _mm256_shuffle_epi8( *(buf+ 9), shuff_bswap64 ); \
+	M[0xA] = _mm256_shuffle_epi8( *(buf+10), shuff_bswap64 ); \
+	M[0xB] = _mm256_shuffle_epi8( *(buf+11), shuff_bswap64 ); \
+	M[0xC] = _mm256_shuffle_epi8( *(buf+12), shuff_bswap64 ); \
+	M[0xD] = _mm256_shuffle_epi8( *(buf+13), shuff_bswap64 ); \
+	M[0xE] = _mm256_shuffle_epi8( *(buf+14), shuff_bswap64 ); \
+	M[0xF] = _mm256_shuffle_epi8( *(buf+15), shuff_bswap64 ); \
 	for (r = 0; r < 16; r ++) \
 		ROUND_B_4WAY(r); \
-        H0 = _mm256_xor_si256( _mm256_xor_si256( \
+   H0 = _mm256_xor_si256( _mm256_xor_si256( \
                    _mm256_xor_si256( S0, V0 ), V8 ), H0 ); \
-        H1 = _mm256_xor_si256( _mm256_xor_si256( \
+   H1 = _mm256_xor_si256( _mm256_xor_si256( \
                    _mm256_xor_si256( S1, V1 ), V9 ), H1 ); \
-        H2 = _mm256_xor_si256( _mm256_xor_si256( \
+   H2 = _mm256_xor_si256( _mm256_xor_si256( \
                    _mm256_xor_si256( S2, V2 ), VA ), H2 ); \
-        H3 = _mm256_xor_si256( _mm256_xor_si256( \
+   H3 = _mm256_xor_si256( _mm256_xor_si256( \
                    _mm256_xor_si256( S3, V3 ), VB ), H3 ); \
-        H4 = _mm256_xor_si256( _mm256_xor_si256( \
+   H4 = _mm256_xor_si256( _mm256_xor_si256( \
                    _mm256_xor_si256( S0, V4 ), VC ), H4 ); \
-        H5 = _mm256_xor_si256( _mm256_xor_si256( \
+   H5 = _mm256_xor_si256( _mm256_xor_si256( \
                    _mm256_xor_si256( S1, V5 ), VD ), H5 ); \
-        H6 = _mm256_xor_si256( _mm256_xor_si256( \
+   H6 = _mm256_xor_si256( _mm256_xor_si256( \
                    _mm256_xor_si256( S2, V6 ), VE ), H6 ); \
-        H7 = _mm256_xor_si256( _mm256_xor_si256( \
+   H7 = _mm256_xor_si256( _mm256_xor_si256( \
                    _mm256_xor_si256( S3, V7 ), VF ), H7 ); \
-	} while (0)
+} while (0)

 #else

@@ -491,8 +493,7 @@ static const sph_u64 CB[16] = {
                         m256_const1_64( CB6 ) );  \
  VF = _mm256_xor_si256( _mm256_set1_epi64x( T1 ), \
                         m256_const1_64( CB7 ) );  \
-  shuf_bswap64 = m256_const_64( 0x08090a0b0c0d0e0f, 0x0001020304050607, \
-                                0x08090a0b0c0d0e0f, 0x0001020304050607 ); \
+  shuf_bswap64 = m256_const2_64( 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 ); \
@@ -620,7 +621,7 @@ blake64_4way_close( blake_4way_big_context *sc,
   bit_len = ((unsigned)ptr << 3);
   z = 0x80 >> n;
   zz = ((ub & -z) | z) & 0xFF;
-   buf[ptr>>3] = _mm256_set_epi64x( zz, zz, zz, zz );
+   buf[ptr>>3] = _mm256_set1_epi64x( zz );
   tl = sc->T0 + bit_len;
   th = sc->T1;
   if (ptr == 0 )
--- a/algo/bmw/bmw-hash-4way.h
+++ b/algo/bmw/bmw-hash-4way.h
@@ -78,7 +78,7 @@ void bmw256_4way_addbits_and_close(
 // BMW-256 8 way 32

 typedef struct {
-   __m256i buf[64];
+   __m256i buf[16];
   __m256i H[16];
   size_t ptr;
   uint32_t bit_count;  // assume bit_count fits in 32 bits
@@ -121,7 +121,7 @@ typedef struct {
   __m256i H[16];
   size_t ptr;
   sph_u64 bit_count;
-} bmw_4way_big_context;
+} bmw_4way_big_context __attribute__((aligned(128)));

 typedef bmw_4way_big_context bmw512_4way_context;

@@ -137,6 +137,22 @@ void bmw512_4way_addbits_and_close(

 #endif  // __AVX2__

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+typedef struct {
+   __m512i buf[16];
+   __m512i H[16];
+   size_t ptr;
+   uint64_t bit_count;
+} bmw512_8way_context __attribute__((aligned(128)));
+
+void bmw512_8way_init( bmw512_8way_context *ctx );
+void bmw512_8way_update( bmw512_8way_context *ctx, const void *data,
+                         size_t len );
+void bmw512_8way_close( bmw512_8way_context *ctx, void *dst );
+
+#endif // AVX512
+
 #ifdef __cplusplus
 }
 #endif
--- a/algo/bmw/bmw256-hash-4way.c
+++ b/algo/bmw/bmw256-hash-4way.c
@@ -137,165 +137,151 @@ static const uint32_t IV256[] = {
                           ss4( qt[ (i)- 2 ] ), ss5( qt[ (i)- 1 ] ) ) ), \
      add_elt_s( M, H, (i)-16 ) )

+// Expressions are grouped using associativity to reduce CPU depenedencies,
+// resulting in some sign changes compared to the reference code.
+
 #define Ws0 \
   _mm_add_epi32( \
-       _mm_add_epi32( \
-          _mm_add_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 5], H[ 5] ), \
-                            _mm_xor_si128( M[ 7], H[ 7] ) ), \
-             _mm_xor_si128( M[10], H[10] ) ), \
-          _mm_xor_si128( M[13], H[13] ) ), \
-       _mm_xor_si128( M[14], H[14] ) )
+      _mm_add_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 5], H[ 5] ), \
+                        _mm_xor_si128( M[ 7], H[ 7] ) ), \
+         _mm_xor_si128( M[10], H[10] ) ), \
+      _mm_add_epi32( _mm_xor_si128( M[13], H[13] ), \
+                     _mm_xor_si128( M[14], H[14] ) ) )

 #define Ws1 \
-   _mm_sub_epi32( \
+   _mm_add_epi32( \
       _mm_add_epi32( \
-          _mm_add_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 6], H[ 6] ), \
-                            _mm_xor_si128( M[ 8], H[ 8] ) ), \
-             _mm_xor_si128( M[11], H[11] ) ), \
-          _mm_xor_si128( M[14], H[14] ) ), \
-       _mm_xor_si128( M[15], H[15] ) )
+          _mm_sub_epi32( _mm_xor_si128( M[ 6], H[ 6] ), \
+                         _mm_xor_si128( M[ 8], H[ 8] ) ), \
+          _mm_xor_si128( M[11], H[11] ) ), \
+       _mm_sub_epi32( _mm_xor_si128( M[14], H[14] ), \
+                      _mm_xor_si128( M[15], H[15] ) ) )

 #define Ws2 \
-   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_add_epi32( \
-             _mm_add_epi32( _mm_xor_si128( M[ 0], H[ 0] ), \
-                            _mm_xor_si128( M[ 7], H[ 7] ) ), \
-             _mm_xor_si128( M[ 9], H[ 9] ) ), \
-          _mm_xor_si128( M[12], H[12] ) ), \
-       _mm_xor_si128( M[15], H[15] ) )
+   _mm_sub_epi32( \
+      _mm_add_epi32( \
+         _mm_add_epi32( _mm_xor_si128( M[ 0], H[ 0] ), \
+                        _mm_xor_si128( M[ 7], H[ 7] ) ), \
+         _mm_xor_si128( M[ 9], H[ 9] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[12], H[12] ), \
+                     _mm_xor_si128( M[15], H[15] ) ) )

 #define Ws3 \
-   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_add_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 0], H[ 0] ), \
-                            _mm_xor_si128( M[ 1], H[ 1] ) ), \
-             _mm_xor_si128( M[ 8], H[ 8] ) ), \
-          _mm_xor_si128( M[10], H[10] ) ), \
-       _mm_xor_si128( M[13], H[13] ) )
+   _mm_sub_epi32( \
+      _mm_add_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 0], H[ 0] ), \
+                        _mm_xor_si128( M[ 1], H[ 1] ) ), \
+         _mm_xor_si128( M[ 8], H[ 8] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[10], H[10] ), \
+                     _mm_xor_si128( M[13], H[13] ) ) )

 #define Ws4 \
   _mm_sub_epi32( \
-       _mm_sub_epi32( \
-          _mm_add_epi32( \
-             _mm_add_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
-                            _mm_xor_si128( M[ 2], H[ 2] ) ), \
-             _mm_xor_si128( M[ 9], H[ 9] ) ), \
-          _mm_xor_si128( M[11], H[11] ) ), \
-       _mm_xor_si128( M[14], H[14] ) )
+      _mm_add_epi32( \
+         _mm_add_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
+                        _mm_xor_si128( M[ 2], H[ 2] ) ), \
+         _mm_xor_si128( M[ 9], H[ 9] ) ), \
+      _mm_add_epi32( _mm_xor_si128( M[11], H[11] ), \
+                     _mm_xor_si128( M[14], H[14] ) ) )

 #define Ws5 \
-   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_add_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 3], H[ 3] ), \
-                            _mm_xor_si128( M[ 2], H[ 2] ) ), \
-             _mm_xor_si128( M[10], H[10] ) ), \
-          _mm_xor_si128( M[12], H[12] ) ), \
-       _mm_xor_si128( M[15], H[15] ) )
+   _mm_sub_epi32( \
+      _mm_add_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 3], H[ 3] ), \
+                        _mm_xor_si128( M[ 2], H[ 2] ) ), \
+         _mm_xor_si128( M[10], H[10] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[12], H[12] ), \
+                     _mm_xor_si128( M[15], H[15] ) ) )

 #define Ws6 \
-   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_sub_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 4], H[ 4] ), \
-                            _mm_xor_si128( M[ 0], H[ 0] ) ), \
-             _mm_xor_si128( M[ 3], H[ 3] ) ), \
-          _mm_xor_si128( M[11], H[11] ) ), \
-       _mm_xor_si128( M[13], H[13] ) )
+   _mm_sub_epi32( \
+      _mm_sub_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 4], H[ 4] ), \
+                        _mm_xor_si128( M[ 0], H[ 0] ) ), \
+         _mm_xor_si128( M[ 3], H[ 3] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[11], H[11] ), \
+                     _mm_xor_si128( M[13], H[13] ) ) )

 #define Ws7 \
   _mm_sub_epi32( \
-       _mm_sub_epi32( \
-          _mm_sub_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
-                            _mm_xor_si128( M[ 4], H[ 4] ) ), \
-             _mm_xor_si128( M[ 5], H[ 5] ) ), \
-          _mm_xor_si128( M[12], H[12] ) ), \
-       _mm_xor_si128( M[14], H[14] ) )
+      _mm_sub_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
+                        _mm_xor_si128( M[ 4], H[ 4] ) ), \
+         _mm_xor_si128( M[ 5], H[ 5] ) ), \
+      _mm_add_epi32( _mm_xor_si128( M[12], H[12] ), \
+                     _mm_xor_si128( M[14], H[14] ) ) )

 #define Ws8 \
-   _mm_sub_epi32( \
-       _mm_add_epi32( \
-          _mm_sub_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 2], H[ 2] ), \
-                            _mm_xor_si128( M[ 5], H[ 5] ) ), \
-             _mm_xor_si128( M[ 6], H[ 6] ) ), \
-          _mm_xor_si128( M[13], H[13] ) ), \
-       _mm_xor_si128( M[15], H[15] ) )
-
-#define Ws9 \
   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_add_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 0], H[ 0] ), \
-                            _mm_xor_si128( M[ 3], H[ 3] ) ), \
-             _mm_xor_si128( M[ 6], H[ 6] ) ), \
-          _mm_xor_si128( M[ 7], H[ 7] ) ), \
-       _mm_xor_si128( M[14], H[14] ) )
+      _mm_sub_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 2], H[ 2] ), \
+                        _mm_xor_si128( M[ 5], H[ 5] ) ), \
+         _mm_xor_si128( M[ 6], H[ 6] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[13], H[13] ), \
+                     _mm_xor_si128( M[15], H[15] ) ) )
+#define Ws9 \
+   _mm_sub_epi32( \
+      _mm_add_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 0], H[ 0] ), \
+                        _mm_xor_si128( M[ 3], H[ 3] ) ), \
+         _mm_xor_si128( M[ 6], H[ 6] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[ 7], H[ 7] ), \
+                     _mm_xor_si128( M[14], H[14] ) ) )

 #define Ws10 \
-   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_sub_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 8], H[ 8] ), \
-                            _mm_xor_si128( M[ 1], H[ 1] ) ), \
-             _mm_xor_si128( M[ 4], H[ 4] ) ), \
-          _mm_xor_si128( M[ 7], H[ 7] ) ), \
-       _mm_xor_si128( M[15], H[15] ) )
+   _mm_sub_epi32( \
+      _mm_sub_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 8], H[ 8] ), \
+                        _mm_xor_si128( M[ 1], H[ 1] ) ), \
+         _mm_xor_si128( M[ 4], H[ 4] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[ 7], H[ 7] ), \
+                     _mm_xor_si128( M[15], H[15] ) ) )

 #define Ws11 \
-   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_sub_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 8], H[ 8] ), \
-                            _mm_xor_si128( M[ 0], H[ 0] ) ), \
-             _mm_xor_si128( M[ 2], H[ 2] ) ), \
-          _mm_xor_si128( M[ 5], H[ 5] ) ), \
-       _mm_xor_si128( M[ 9], H[ 9] ) )
+   _mm_sub_epi32( \
+      _mm_sub_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 8], H[ 8] ), \
+                        _mm_xor_si128( M[ 0], H[ 0] ) ), \
+         _mm_xor_si128( M[ 2], H[ 2] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[ 5], H[ 5] ), \
+                     _mm_xor_si128( M[ 9], H[ 9] ) ) )

 #define Ws12 \
-   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_sub_epi32( \
-             _mm_add_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
-                            _mm_xor_si128( M[ 3], H[ 3] ) ), \
-             _mm_xor_si128( M[ 6], H[ 6] ) ), \
-          _mm_xor_si128( M[ 9], H[ 9] ) ), \
-       _mm_xor_si128( M[10], H[10] ) )
+   _mm_sub_epi32( \
+      _mm_sub_epi32( \
+         _mm_add_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
+                        _mm_xor_si128( M[ 3], H[ 3] ) ), \
+         _mm_xor_si128( M[ 6], H[ 6] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[ 9], H[ 9] ), \
+                     _mm_xor_si128( M[10], H[10] ) ) )

 #define Ws13 \
   _mm_add_epi32( \
-       _mm_add_epi32( \
-          _mm_add_epi32( \
-             _mm_add_epi32( _mm_xor_si128( M[ 2], H[ 2] ), \
-                            _mm_xor_si128( M[ 4], H[ 4] ) ), \
-             _mm_xor_si128( M[ 7], H[ 7] ) ), \
-          _mm_xor_si128( M[10], H[10] ) ), \
-       _mm_xor_si128( M[11], H[11] ) )
+      _mm_add_epi32( \
+         _mm_add_epi32( _mm_xor_si128( M[ 2], H[ 2] ), \
+                        _mm_xor_si128( M[ 4], H[ 4] ) ), \
+         _mm_xor_si128( M[ 7], H[ 7] ) ), \
+      _mm_add_epi32( _mm_xor_si128( M[10], H[10] ), \
+                     _mm_xor_si128( M[11], H[11] ) ) )

 #define Ws14 \
   _mm_sub_epi32( \
-       _mm_sub_epi32( \
-          _mm_add_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[ 3], H[ 3] ), \
-                               _mm_xor_si128( M[ 5], H[ 5] ) ), \
-             _mm_xor_si128( M[ 8], H[ 8] ) ), \
-          _mm_xor_si128( M[11], H[11] ) ), \
-       _mm_xor_si128( M[12], H[12] ) )
+      _mm_add_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[ 3], H[ 3] ), \
+                        _mm_xor_si128( M[ 5], H[ 5] ) ), \
+         _mm_xor_si128( M[ 8], H[ 8] ) ), \
+      _mm_add_epi32( _mm_xor_si128( M[11], H[11] ), \
+                     _mm_xor_si128( M[12], H[12] ) ) )

 #define Ws15 \
-   _mm_add_epi32( \
-       _mm_sub_epi32( \
-          _mm_sub_epi32( \
-             _mm_sub_epi32( _mm_xor_si128( M[12], H[12] ), \
-                            _mm_xor_si128( M[ 4], H[ 4] ) ), \
-             _mm_xor_si128( M[ 6], H[ 6] ) ), \
-          _mm_xor_si128( M[ 9], H[ 9] ) ), \
-       _mm_xor_si128( M[13], H[13] ) )
+   _mm_sub_epi32( \
+      _mm_sub_epi32( \
+         _mm_sub_epi32( _mm_xor_si128( M[12], H[12] ), \
+                        _mm_xor_si128( M[ 4], H[4] ) ), \
+         _mm_xor_si128( M[ 6], H[ 6] ) ), \
+      _mm_sub_epi32( _mm_xor_si128( M[ 9], H[ 9] ), \
+                     _mm_xor_si128( M[13], H[13] ) ) )


 void compress_small( const __m128i *M, const __m128i H[16], __m128i dH[16] )
@@ -700,163 +686,148 @@ bmw256_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)

 #define W8s0 \
   _mm256_add_epi32( \
-       _mm256_add_epi32( \
-          _mm256_add_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 5], H[ 5] ), \
-                               _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-             _mm256_xor_si256( M[10], H[10] ) ), \
-          _mm256_xor_si256( M[13], H[13] ) ), \
-       _mm256_xor_si256( M[14], H[14] ) )
+      _mm256_add_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 5], H[ 5] ), \
+                           _mm256_xor_si256( M[ 7], H[ 7] ) ), \
+         _mm256_xor_si256( M[10], H[10] ) ), \
+      _mm256_add_epi32( _mm256_xor_si256( M[13], H[13] ), \
+                        _mm256_xor_si256( M[14], H[14] ) ) )

 #define W8s1 \
-   _mm256_sub_epi32( \
+   _mm256_add_epi32( \
       _mm256_add_epi32( \
-          _mm256_add_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 6], H[ 6] ), \
-                               _mm256_xor_si256( M[ 8], H[ 8] ) ), \
-             _mm256_xor_si256( M[11], H[11] ) ), \
-          _mm256_xor_si256( M[14], H[14] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+          _mm256_sub_epi32( _mm256_xor_si256( M[ 6], H[ 6] ), \
+                            _mm256_xor_si256( M[ 8], H[ 8] ) ), \
+          _mm256_xor_si256( M[11], H[11] ) ), \
+       _mm256_sub_epi32( _mm256_xor_si256( M[14], H[14] ), \
+                         _mm256_xor_si256( M[15], H[15] ) ) )

 #define W8s2 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_add_epi32( \
-             _mm256_add_epi32( _mm256_xor_si256( M[ 0], H[ 0] ), \
-                               _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-             _mm256_xor_si256( M[ 9], H[ 9] ) ), \
-          _mm256_xor_si256( M[12], H[12] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+   _mm256_sub_epi32( \
+      _mm256_add_epi32( \
+         _mm256_add_epi32( _mm256_xor_si256( M[ 0], H[ 0] ), \
+                           _mm256_xor_si256( M[ 7], H[ 7] ) ), \
+         _mm256_xor_si256( M[ 9], H[ 9] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[12], H[12] ), \
+                        _mm256_xor_si256( M[15], H[15] ) ) )

 #define W8s3 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_add_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 0], H[ 0] ), \
-                               _mm256_xor_si256( M[ 1], H[ 1] ) ), \
-             _mm256_xor_si256( M[ 8], H[ 8] ) ), \
-          _mm256_xor_si256( M[10], H[10] ) ), \
-       _mm256_xor_si256( M[13], H[13] ) )
+   _mm256_sub_epi32( \
+      _mm256_add_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 0], H[ 0] ), \
+                           _mm256_xor_si256( M[ 1], H[ 1] ) ), \
+         _mm256_xor_si256( M[ 8], H[ 8] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[10], H[10] ), \
+                        _mm256_xor_si256( M[13], H[13] ) ) )

 #define W8s4 \
   _mm256_sub_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_add_epi32( \
-             _mm256_add_epi32( _mm256_xor_si256( M[ 1], H[ 1] ), \
-                               _mm256_xor_si256( M[ 2], H[ 2] ) ), \
-             _mm256_xor_si256( M[ 9], H[ 9] ) ), \
-          _mm256_xor_si256( M[11], H[11] ) ), \
-       _mm256_xor_si256( M[14], H[14] ) )
+      _mm256_add_epi32( \
+         _mm256_add_epi32( _mm256_xor_si256( M[ 1], H[ 1] ), \
+                           _mm256_xor_si256( M[ 2], H[ 2] ) ), \
+         _mm256_xor_si256( M[ 9], H[ 9] ) ), \
+      _mm256_add_epi32( _mm256_xor_si256( M[11], H[11] ), \
+                        _mm256_xor_si256( M[14], H[14] ) ) )

 #define W8s5 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_add_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 3], H[ 3] ), \
-                               _mm256_xor_si256( M[ 2], H[ 2] ) ), \
-             _mm256_xor_si256( M[10], H[10] ) ), \
-          _mm256_xor_si256( M[12], H[12] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+   _mm256_sub_epi32( \
+      _mm256_add_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 3], H[ 3] ), \
+                           _mm256_xor_si256( M[ 2], H[ 2] ) ), \
+         _mm256_xor_si256( M[10], H[10] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[12], H[12] ), \
+                        _mm256_xor_si256( M[15], H[15] ) ) )

 #define W8s6 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_sub_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 4], H[ 4] ), \
-                               _mm256_xor_si256( M[ 0], H[ 0] ) ), \
-             _mm256_xor_si256( M[ 3], H[ 3] ) ), \
-          _mm256_xor_si256( M[11], H[11] ) ), \
-       _mm256_xor_si256( M[13], H[13] ) )
+   _mm256_sub_epi32( \
+      _mm256_sub_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 4], H[ 4] ), \
+                           _mm256_xor_si256( M[ 0], H[ 0] ) ), \
+         _mm256_xor_si256( M[ 3], H[ 3] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[11], H[11] ), \
+                        _mm256_xor_si256( M[13], H[13] ) ) )

 #define W8s7 \
   _mm256_sub_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_sub_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 1], H[ 1] ), \
-                               _mm256_xor_si256( M[ 4], H[ 4] ) ), \
-             _mm256_xor_si256( M[ 5], H[ 5] ) ), \
-          _mm256_xor_si256( M[12], H[12] ) ), \
-       _mm256_xor_si256( M[14], H[14] ) )
+      _mm256_sub_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 1], H[ 1] ), \
+                           _mm256_xor_si256( M[ 4], H[ 4] ) ), \
+         _mm256_xor_si256( M[ 5], H[ 5] ) ), \
+      _mm256_add_epi32( _mm256_xor_si256( M[12], H[12] ), \
+                        _mm256_xor_si256( M[14], H[14] ) ) )

 #define W8s8 \
-   _mm256_sub_epi32( \
-       _mm256_add_epi32( \
-          _mm256_sub_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 2], H[ 2] ), \
-                               _mm256_xor_si256( M[ 5], H[ 5] ) ), \
-             _mm256_xor_si256( M[ 6], H[ 6] ) ), \
-          _mm256_xor_si256( M[13], H[13] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+   _mm256_add_epi32( \
+      _mm256_sub_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 2], H[ 2] ), \
+                           _mm256_xor_si256( M[ 5], H[ 5] ) ), \
+         _mm256_xor_si256( M[ 6], H[ 6] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[13], H[13] ), \
+                        _mm256_xor_si256( M[15], H[15] ) ) )

 #define W8s9 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_add_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 0], H[ 0] ), \
-                               _mm256_xor_si256( M[ 3], H[ 3] ) ), \
-             _mm256_xor_si256( M[ 6], H[ 6] ) ), \
-          _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-       _mm256_xor_si256( M[14], H[14] ) )
+   _mm256_sub_epi32( \
+      _mm256_add_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 0], H[ 0] ), \
+                           _mm256_xor_si256( M[ 3], H[ 3] ) ), \
+         _mm256_xor_si256( M[ 6], H[ 6] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[ 7], H[ 7] ), \
+                        _mm256_xor_si256( M[14], H[14] ) ) )

 #define W8s10 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_sub_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 8], H[ 8] ), \
-                               _mm256_xor_si256( M[ 1], H[ 1] ) ), \
-             _mm256_xor_si256( M[ 4], H[ 4] ) ), \
-          _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+   _mm256_sub_epi32( \
+      _mm256_sub_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 8], H[ 8] ), \
+                           _mm256_xor_si256( M[ 1], H[ 1] ) ), \
+         _mm256_xor_si256( M[ 4], H[ 4] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[ 7], H[ 7] ), \
+                        _mm256_xor_si256( M[15], H[15] ) ) )

 #define W8s11 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_sub_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 8], H[ 8] ), \
-                               _mm256_xor_si256( M[ 0], H[ 0] ) ), \
-             _mm256_xor_si256( M[ 2], H[ 2] ) ), \
-          _mm256_xor_si256( M[ 5], H[ 5] ) ), \
-       _mm256_xor_si256( M[ 9], H[ 9] ) )
+   _mm256_sub_epi32( \
+      _mm256_sub_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 8], H[ 8] ), \
+                           _mm256_xor_si256( M[ 0], H[ 0] ) ), \
+         _mm256_xor_si256( M[ 2], H[ 2] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[ 5], H[ 5] ), \
+                        _mm256_xor_si256( M[ 9], H[ 9] ) ) )

 #define W8s12 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_sub_epi32( \
-             _mm256_add_epi32( _mm256_xor_si256( M[ 1], H[ 1] ), \
-                               _mm256_xor_si256( M[ 3], H[ 3] ) ), \
-             _mm256_xor_si256( M[ 6], H[ 6] ) ), \
-          _mm256_xor_si256( M[ 9], H[ 9] ) ), \
-       _mm256_xor_si256( M[10], H[10] ) )
+   _mm256_sub_epi32( \
+      _mm256_sub_epi32( \
+         _mm256_add_epi32( _mm256_xor_si256( M[ 1], H[ 1] ), \
+                           _mm256_xor_si256( M[ 3], H[ 3] ) ), \
+         _mm256_xor_si256( M[ 6], H[ 6] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[ 9], H[ 9] ), \
+                        _mm256_xor_si256( M[10], H[10] ) ) )

 #define W8s13 \
   _mm256_add_epi32( \
-       _mm256_add_epi32( \
-          _mm256_add_epi32( \
-             _mm256_add_epi32( _mm256_xor_si256( M[ 2], H[ 2] ), \
-                               _mm256_xor_si256( M[ 4], H[ 4] ) ), \
-             _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-          _mm256_xor_si256( M[10], H[10] ) ), \
-       _mm256_xor_si256( M[11], H[11] ) )
+      _mm256_add_epi32( \
+         _mm256_add_epi32( _mm256_xor_si256( M[ 2], H[ 2] ), \
+                           _mm256_xor_si256( M[ 4], H[ 4] ) ), \
+         _mm256_xor_si256( M[ 7], H[ 7] ) ), \
+      _mm256_add_epi32( _mm256_xor_si256( M[10], H[10] ), \
+                        _mm256_xor_si256( M[11], H[11] ) ) )

 #define W8s14 \
   _mm256_sub_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_add_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[ 3], H[ 3] ), \
-                               _mm256_xor_si256( M[ 5], H[ 5] ) ), \
-             _mm256_xor_si256( M[ 8], H[ 8] ) ), \
-          _mm256_xor_si256( M[11], H[11] ) ), \
-       _mm256_xor_si256( M[12], H[12] ) )
+      _mm256_add_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[ 3], H[ 3] ), \
+                           _mm256_xor_si256( M[ 5], H[ 5] ) ), \
+         _mm256_xor_si256( M[ 8], H[ 8] ) ), \
+      _mm256_add_epi32( _mm256_xor_si256( M[11], H[11] ), \
+                        _mm256_xor_si256( M[12], H[12] ) ) )

 #define W8s15 \
-   _mm256_add_epi32( \
-       _mm256_sub_epi32( \
-          _mm256_sub_epi32( \
-             _mm256_sub_epi32( _mm256_xor_si256( M[12], H[12] ), \
-                               _mm256_xor_si256( M[ 4], H[4] ) ), \
-             _mm256_xor_si256( M[ 6], H[ 6] ) ), \
-          _mm256_xor_si256( M[ 9], H[ 9] ) ), \
-       _mm256_xor_si256( M[13], H[13] ) )
+   _mm256_sub_epi32( \
+      _mm256_sub_epi32( \
+         _mm256_sub_epi32( _mm256_xor_si256( M[12], H[12] ), \
+                           _mm256_xor_si256( M[ 4], H[4] ) ), \
+         _mm256_xor_si256( M[ 6], H[ 6] ) ), \
+      _mm256_sub_epi32( _mm256_xor_si256( M[ 9], H[ 9] ), \
+                        _mm256_xor_si256( M[13], H[13] ) ) )
+

 void compress_small_8way( const __m256i *M, const __m256i H[16],
 	                  __m256i dH[16] )
--- a/algo/bmw/bmw512-4way.c
+++ b/algo/bmw/bmw512-4way.c
@@ -1,13 +1,66 @@
 #include "bmw512-gate.h"
-
-#ifdef BMW512_4WAY
-
 #include <stdlib.h>
 #include <string.h>
 #include <stdint.h>
 //#include "sph_keccak.h"
 #include "bmw-hash-4way.h"

+#if defined(BMW512_8WAY)
+
+void bmw512hash_8way(void *state, const void *input)
+{
+    bmw512_8way_context ctx;
+    bmw512_8way_init( &ctx );
+    bmw512_8way_update( &ctx, input, 80 );
+    bmw512_8way_close( &ctx, state );
+}
+
+int scanhash_bmw512_8way( struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t vdata[24*8] __attribute__ ((aligned (64)));
+   uint32_t hash[16*8] __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash7 = &(hash[49]);   // 3*16+1
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   uint32_t n = pdata[19];
+   const uint32_t first_nonce = pdata[19];
+   __m512i  *noncev = (__m512i*)vdata + 9;   // aligned
+//   const uint32_t Htarg = ptarget[7];
+   int thr_id = mythr->id;
+
+   mm512_bswap32_intrlv80_8x64( vdata, pdata );
+   do {
+       *noncev = mm512_intrlv_blend_32( mm512_bswap_32(
+                _mm512_set_epi32( n+7, 0, n+6, 0, n+5, 0, n+4, 0 ,
+                                  n+3, 0, n+2, 0, n+1, 0, n  , 0 ) ), *noncev );
+
+      bmw512hash_8way( hash, vdata );
+
+      for ( int lane = 0; lane < 8; lane++ )
+      if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
+      {
+          extr_lane_8x64( lane_hash, hash, lane, 256 );
+          if ( fulltest( lane_hash, ptarget ) )
+          {
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
+          }
+      }
+      n += 4;
+
+   } while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+   
+
+#elif defined(BMW512_4WAY)
+
+//#ifdef BMW512_4WAY
+
 void bmw512hash_4way(void *state, const void *input)
 {
    bmw512_4way_context ctx;
--- a/algo/bmw/bmw512-gate.c
+++ b/algo/bmw/bmw512-gate.c
@@ -2,9 +2,12 @@

 bool register_bmw512_algo( algo_gate_t* gate )
 {
-  gate->optimizations = AVX2_OPT;
+  gate->optimizations = AVX2_OPT | AVX512_OPT;
  opt_target_factor = 256.0;
-#if defined (BMW512_4WAY)
+#if defined (BMW512_8WAY)
+  gate->scanhash  = (void*)&scanhash_bmw512_8way;
+  gate->hash      = (void*)&bmw512hash_8way;
+#elif defined (BMW512_4WAY)
  gate->scanhash  = (void*)&scanhash_bmw512_4way;
  gate->hash      = (void*)&bmw512hash_4way;
 #else
--- a/algo/bmw/bmw512-gate.h
+++ b/algo/bmw/bmw512-gate.h
@@ -1,23 +1,33 @@
 #ifndef BMW512_GATE_H__
-#define BMW512_GATE_H__
+#define BMW512_GATE_H__ 1

 #include "algo-gate-api.h"
 #include <stdint.h>

-#if defined(__AVX2__)
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define BMW512_8WAY 1
+#elif defined(__AVX2__)
  #define BMW512_4WAY 1
 #endif

-#if defined(BMW512_4WAY)
+#if defined(BMW512_8WAY)
+
+void bmw512hash_8way( void *state, const void *input );
+int scanhash_bmw512_8way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+
+#elif defined(BMW512_4WAY)

 void bmw512hash_4way( void *state, const void *input );
 int scanhash_bmw512_4way( struct work *work, uint32_t max_nonce,
                         uint64_t *hashes_done, struct thr_info *mythr );

-#endif
+#else

 void bmw512hash( void *state, const void *input );
 int scanhash_bmw512( struct work *work, uint32_t max_nonce,
                    uint64_t *hashes_done, struct thr_info *mythr );

 #endif
+
+#endif
--- a/algo/bmw/bmw512-hash-4way.c
+++ b/algo/bmw/bmw512-hash-4way.c
@@ -556,7 +556,7 @@ void bmw512_2way_close( bmw_2way_big_context *ctx, void *dst )
   compress_big_2way( buf, h, h2 );
   memcpy_128( buf, h2, 16 );
   compress_big_2way( buf, final_b2, h1 );
-   memcpy( (__m128i*)dst, h1+16, 8 );
+   memcpy( (__m128i*)dst, h1+8, 8 );
 }

 #endif  // __SSE2__
@@ -636,165 +636,152 @@ void bmw512_2way_close( bmw_2way_big_context *ctx, void *dst )
                     sb4( qt[ (i)- 2 ] ), sb5( qt[ (i)- 1 ] ) ) ), \
      add_elt_b( M, H, (i)-16 ) )

+
+
 #define Wb0 \
   _mm256_add_epi64( \
-       _mm256_add_epi64( \
-          _mm256_add_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 5], H[ 5] ), \
-                               _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-             _mm256_xor_si256( M[10], H[10] ) ), \
-          _mm256_xor_si256( M[13], H[13] ) ), \
-       _mm256_xor_si256( M[14], H[14] ) )
+      _mm256_add_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 5], H[ 5] ), \
+                           _mm256_xor_si256( M[ 7], H[ 7] ) ), \
+         _mm256_xor_si256( M[10], H[10] ) ), \
+      _mm256_add_epi64( _mm256_xor_si256( M[13], H[13] ), \
+                        _mm256_xor_si256( M[14], H[14] ) ) )

 #define Wb1 \
-   _mm256_sub_epi64( \
+   _mm256_add_epi64( \
       _mm256_add_epi64( \
-          _mm256_add_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 6], H[ 6] ), \
-                               _mm256_xor_si256( M[ 8], H[ 8] ) ), \
-             _mm256_xor_si256( M[11], H[11] ) ), \
-          _mm256_xor_si256( M[14], H[14] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+          _mm256_sub_epi64( _mm256_xor_si256( M[ 6], H[ 6] ), \
+                            _mm256_xor_si256( M[ 8], H[ 8] ) ), \
+          _mm256_xor_si256( M[11], H[11] ) ), \
+       _mm256_sub_epi64( _mm256_xor_si256( M[14], H[14] ), \
+                         _mm256_xor_si256( M[15], H[15] ) ) )

 #define Wb2 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_add_epi64( \
-             _mm256_add_epi64( _mm256_xor_si256( M[ 0], H[ 0] ), \
-                               _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-             _mm256_xor_si256( M[ 9], H[ 9] ) ), \
-          _mm256_xor_si256( M[12], H[12] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+   _mm256_sub_epi64( \
+      _mm256_add_epi64( \
+         _mm256_add_epi64( _mm256_xor_si256( M[ 0], H[ 0] ), \
+                           _mm256_xor_si256( M[ 7], H[ 7] ) ), \
+         _mm256_xor_si256( M[ 9], H[ 9] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[12], H[12] ), \
+                        _mm256_xor_si256( M[15], H[15] ) ) )

 #define Wb3 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_add_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 0], H[ 0] ), \
-                               _mm256_xor_si256( M[ 1], H[ 1] ) ), \
-             _mm256_xor_si256( M[ 8], H[ 8] ) ), \
-          _mm256_xor_si256( M[10], H[10] ) ), \
-       _mm256_xor_si256( M[13], H[13] ) )
+   _mm256_sub_epi64( \
+      _mm256_add_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 0], H[ 0] ), \
+                           _mm256_xor_si256( M[ 1], H[ 1] ) ), \
+         _mm256_xor_si256( M[ 8], H[ 8] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[10], H[10] ), \
+                        _mm256_xor_si256( M[13], H[13] ) ) )

 #define Wb4 \
   _mm256_sub_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_add_epi64( \
-             _mm256_add_epi64( _mm256_xor_si256( M[ 1], H[ 1] ), \
-                               _mm256_xor_si256( M[ 2], H[ 2] ) ), \
-             _mm256_xor_si256( M[ 9], H[ 9] ) ), \
-          _mm256_xor_si256( M[11], H[11] ) ), \
-       _mm256_xor_si256( M[14], H[14] ) )
+      _mm256_add_epi64( \
+         _mm256_add_epi64( _mm256_xor_si256( M[ 1], H[ 1] ), \
+                           _mm256_xor_si256( M[ 2], H[ 2] ) ), \
+         _mm256_xor_si256( M[ 9], H[ 9] ) ), \
+      _mm256_add_epi64( _mm256_xor_si256( M[11], H[11] ), \
+                        _mm256_xor_si256( M[14], H[14] ) ) )

 #define Wb5 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_add_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 3], H[ 3] ), \
-                               _mm256_xor_si256( M[ 2], H[ 2] ) ), \
-             _mm256_xor_si256( M[10], H[10] ) ), \
-          _mm256_xor_si256( M[12], H[12] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+   _mm256_sub_epi64( \
+      _mm256_add_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 3], H[ 3] ), \
+                           _mm256_xor_si256( M[ 2], H[ 2] ) ), \
+         _mm256_xor_si256( M[10], H[10] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[12], H[12] ), \
+                        _mm256_xor_si256( M[15], H[15] ) ) )

 #define Wb6 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_sub_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 4], H[ 4] ), \
-                               _mm256_xor_si256( M[ 0], H[ 0] ) ), \
-             _mm256_xor_si256( M[ 3], H[ 3] ) ), \
-          _mm256_xor_si256( M[11], H[11] ) ), \
-       _mm256_xor_si256( M[13], H[13] ) )
+   _mm256_sub_epi64( \
+      _mm256_sub_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 4], H[ 4] ), \
+                           _mm256_xor_si256( M[ 0], H[ 0] ) ), \
+         _mm256_xor_si256( M[ 3], H[ 3] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[11], H[11] ), \
+                        _mm256_xor_si256( M[13], H[13] ) ) )

 #define Wb7 \
   _mm256_sub_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_sub_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 1], H[ 1] ), \
-                               _mm256_xor_si256( M[ 4], H[ 4] ) ), \
-             _mm256_xor_si256( M[ 5], H[ 5] ) ), \
-          _mm256_xor_si256( M[12], H[12] ) ), \
-       _mm256_xor_si256( M[14], H[14] ) )
+      _mm256_sub_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 1], H[ 1] ), \
+                           _mm256_xor_si256( M[ 4], H[ 4] ) ), \
+         _mm256_xor_si256( M[ 5], H[ 5] ) ), \
+      _mm256_add_epi64( _mm256_xor_si256( M[12], H[12] ), \
+                        _mm256_xor_si256( M[14], H[14] ) ) )

 #define Wb8 \
-   _mm256_sub_epi64( \
-       _mm256_add_epi64( \
-          _mm256_sub_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 2], H[ 2] ), \
-                               _mm256_xor_si256( M[ 5], H[ 5] ) ), \
-             _mm256_xor_si256( M[ 6], H[ 6] ) ), \
-          _mm256_xor_si256( M[13], H[13] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+   _mm256_add_epi64( \
+      _mm256_sub_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 2], H[ 2] ), \
+                           _mm256_xor_si256( M[ 5], H[ 5] ) ), \
+         _mm256_xor_si256( M[ 6], H[ 6] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[13], H[13] ), \
+                        _mm256_xor_si256( M[15], H[15] ) ) )

 #define Wb9 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_add_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 0], H[ 0] ), \
-                               _mm256_xor_si256( M[ 3], H[ 3] ) ), \
-             _mm256_xor_si256( M[ 6], H[ 6] ) ), \
-          _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-       _mm256_xor_si256( M[14], H[14] ) )
+   _mm256_sub_epi64( \
+      _mm256_add_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 0], H[ 0] ), \
+                           _mm256_xor_si256( M[ 3], H[ 3] ) ), \
+         _mm256_xor_si256( M[ 6], H[ 6] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[ 7], H[ 7] ), \
+                        _mm256_xor_si256( M[14], H[14] ) ) )

 #define Wb10 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_sub_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 8], H[ 8] ), \
-                               _mm256_xor_si256( M[ 1], H[ 1] ) ), \
-             _mm256_xor_si256( M[ 4], H[ 4] ) ), \
-          _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-       _mm256_xor_si256( M[15], H[15] ) )
+   _mm256_sub_epi64( \
+      _mm256_sub_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 8], H[ 8] ), \
+                           _mm256_xor_si256( M[ 1], H[ 1] ) ), \
+         _mm256_xor_si256( M[ 4], H[ 4] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[ 7], H[ 7] ), \
+                        _mm256_xor_si256( M[15], H[15] ) ) )

 #define Wb11 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_sub_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 8], H[ 8] ), \
-                               _mm256_xor_si256( M[ 0], H[ 0] ) ), \
-             _mm256_xor_si256( M[ 2], H[ 2] ) ), \
-          _mm256_xor_si256( M[ 5], H[ 5] ) ), \
-       _mm256_xor_si256( M[ 9], H[ 9] ) )
+   _mm256_sub_epi64( \
+      _mm256_sub_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 8], H[ 8] ), \
+                           _mm256_xor_si256( M[ 0], H[ 0] ) ), \
+         _mm256_xor_si256( M[ 2], H[ 2] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[ 5], H[ 5] ), \
+                        _mm256_xor_si256( M[ 9], H[ 9] ) ) )

 #define Wb12 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_sub_epi64( \
-             _mm256_add_epi64( _mm256_xor_si256( M[ 1], H[ 1] ), \
-                               _mm256_xor_si256( M[ 3], H[ 3] ) ), \
-             _mm256_xor_si256( M[ 6], H[ 6] ) ), \
-          _mm256_xor_si256( M[ 9], H[ 9] ) ), \
-       _mm256_xor_si256( M[10], H[10] ) )
+   _mm256_sub_epi64( \
+      _mm256_sub_epi64( \
+         _mm256_add_epi64( _mm256_xor_si256( M[ 1], H[ 1] ), \
+                           _mm256_xor_si256( M[ 3], H[ 3] ) ), \
+         _mm256_xor_si256( M[ 6], H[ 6] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[ 9], H[ 9] ), \
+                        _mm256_xor_si256( M[10], H[10] ) ) )

 #define Wb13 \
   _mm256_add_epi64( \
-       _mm256_add_epi64( \
-          _mm256_add_epi64( \
-             _mm256_add_epi64( _mm256_xor_si256( M[ 2], H[ 2] ), \
-                               _mm256_xor_si256( M[ 4], H[ 4] ) ), \
-             _mm256_xor_si256( M[ 7], H[ 7] ) ), \
-          _mm256_xor_si256( M[10], H[10] ) ), \
-       _mm256_xor_si256( M[11], H[11] ) )
+      _mm256_add_epi64( \
+         _mm256_add_epi64( _mm256_xor_si256( M[ 2], H[ 2] ), \
+                           _mm256_xor_si256( M[ 4], H[ 4] ) ), \
+         _mm256_xor_si256( M[ 7], H[ 7] ) ), \
+      _mm256_add_epi64( _mm256_xor_si256( M[10], H[10] ), \
+                        _mm256_xor_si256( M[11], H[11] ) ) )

 #define Wb14 \
   _mm256_sub_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_add_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[ 3], H[ 3] ), \
-                               _mm256_xor_si256( M[ 5], H[ 5] ) ), \
-             _mm256_xor_si256( M[ 8], H[ 8] ) ), \
-          _mm256_xor_si256( M[11], H[11] ) ), \
-       _mm256_xor_si256( M[12], H[12] ) )
+      _mm256_add_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[ 3], H[ 3] ), \
+                           _mm256_xor_si256( M[ 5], H[ 5] ) ), \
+         _mm256_xor_si256( M[ 8], H[ 8] ) ), \
+      _mm256_add_epi64( _mm256_xor_si256( M[11], H[11] ), \
+                        _mm256_xor_si256( M[12], H[12] ) ) )

 #define Wb15 \
-   _mm256_add_epi64( \
-       _mm256_sub_epi64( \
-          _mm256_sub_epi64( \
-             _mm256_sub_epi64( _mm256_xor_si256( M[12], H[12] ), \
-                               _mm256_xor_si256( M[ 4], H[4] ) ), \
-             _mm256_xor_si256( M[ 6], H[ 6] ) ), \
-          _mm256_xor_si256( M[ 9], H[ 9] ) ), \
-       _mm256_xor_si256( M[13], H[13] ) )
+   _mm256_sub_epi64( \
+      _mm256_sub_epi64( \
+         _mm256_sub_epi64( _mm256_xor_si256( M[12], H[12] ), \
+                           _mm256_xor_si256( M[ 4], H[4] ) ), \
+         _mm256_xor_si256( M[ 6], H[ 6] ) ), \
+      _mm256_sub_epi64( _mm256_xor_si256( M[ 9], H[ 9] ), \
+                        _mm256_xor_si256( M[13], H[13] ) ) )
+

 void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
 {
@@ -1079,6 +1066,477 @@ bmw512_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)

 #endif  // __AVX2__

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// BMW-512 8 WAY
+
+#define s8b0(x) \
+   mm512_xor4( _mm512_srli_epi64( (x), 1), _mm512_slli_epi64( (x), 3), \
+                mm512_rol_64(     (x), 4),  mm512_rol_64(     (x),37) )
+
+#define s8b1(x) \
+   mm512_xor4( _mm512_srli_epi64( (x), 1), _mm512_slli_epi64( (x), 2), \
+                mm512_rol_64(     (x),13),  mm512_rol_64(     (x),43) )
+
+#define s8b2(x) \
+   mm512_xor4( _mm512_srli_epi64( (x), 2), _mm512_slli_epi64( (x), 1), \
+                mm512_rol_64(     (x),19),  mm512_rol_64(     (x),53) )
+
+#define s8b3(x) \
+   mm512_xor4( _mm512_srli_epi64( (x), 2), _mm512_slli_epi64( (x), 2), \
+                mm512_rol_64(     (x),28),  mm512_rol_64(     (x),59) )
+
+#define s8b4(x) \
+  _mm512_xor_si512( (x), _mm512_srli_epi64( (x), 1 ) )
+
+#define s8b5(x) \
+  _mm512_xor_si512( (x), _mm512_srli_epi64( (x), 2 ) )
+
+#define r8b1(x)    mm512_rol_64( x,  5 )
+#define r8b2(x)    mm512_rol_64( x, 11 )
+#define r8b3(x)    mm512_rol_64( x, 27 )
+#define r8b4(x)    mm512_rol_64( x, 32 )
+#define r8b5(x)    mm512_rol_64( x, 37 )
+#define r8b6(x)    mm512_rol_64( x, 43 )
+#define r8b7(x)    mm512_rol_64( x, 53 )
+
+#define rol8w_off_64( M, j, off ) \
+   mm512_rol_64( M[ ( (j) + (off) ) & 0xF ] , \
+                  ( ( (j) + (off) ) & 0xF ) + 1 )
+
+#define add_elt_b8( M, H, j ) \
+   _mm512_xor_si512( \
+      _mm512_add_epi64( \
+            _mm512_sub_epi64( _mm512_add_epi64( rol8w_off_64( M, j, 0 ), \
+                                                rol8w_off_64( M, j, 3 ) ), \
+                             rol8w_off_64( M, j, 10 ) ), \
+            _mm512_set1_epi64( ( (j) + 16 ) * 0x0555555555555555ULL ) ), \
+       H[ ( (j)+7 ) & 0xF ] )
+
+#define expand1b8( qt, M, H, i ) \
+   _mm512_add_epi64( mm512_add4_64( \
+      mm512_add4_64( s8b1( qt[ (i)-16 ] ), s8b2( qt[ (i)-15 ] ), \
+                     s8b3( qt[ (i)-14 ] ), s8b0( qt[ (i)-13 ] )), \
+      mm512_add4_64( s8b1( qt[ (i)-12 ] ), s8b2( qt[ (i)-11 ] ), \
+                     s8b3( qt[ (i)-10 ] ), s8b0( qt[ (i)- 9 ] )), \
+      mm512_add4_64( s8b1( qt[ (i)- 8 ] ), s8b2( qt[ (i)- 7 ] ), \
+                     s8b3( qt[ (i)- 6 ] ), s8b0( qt[ (i)- 5 ] )), \
+      mm512_add4_64( s8b1( qt[ (i)- 4 ] ), s8b2( qt[ (i)- 3 ] ), \
+                     s8b3( qt[ (i)- 2 ] ), s8b0( qt[ (i)- 1 ] ) ) ), \
+      add_elt_b8( M, H, (i)-16 ) )
+
+#define expand2b8( qt, M, H, i) \
+   _mm512_add_epi64( mm512_add4_64( \
+      mm512_add4_64( qt[ (i)-16 ], r8b1( qt[ (i)-15 ] ), \
+                     qt[ (i)-14 ], r8b2( qt[ (i)-13 ] ) ), \
+      mm512_add4_64( qt[ (i)-12 ], r8b3( qt[ (i)-11 ] ), \
+                     qt[ (i)-10 ], r8b4( qt[ (i)- 9 ] ) ), \
+      mm512_add4_64( qt[ (i)- 8 ], r8b5( qt[ (i)- 7 ] ), \
+                     qt[ (i)- 6 ], r8b6( qt[ (i)- 5 ] ) ), \
+      mm512_add4_64( qt[ (i)- 4 ], r8b7( qt[ (i)- 3 ] ), \
+                     s8b4( qt[ (i)- 2 ] ), s8b5( qt[ (i)- 1 ] ) ) ), \
+      add_elt_b8( M, H, (i)-16 ) )
+
+
+
+#define W8b0 \
+   _mm512_add_epi64( \
+      _mm512_add_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 5], H[ 5] ), \
+                           _mm512_xor_si512( M[ 7], H[ 7] ) ), \
+         _mm512_xor_si512( M[10], H[10] ) ), \
+      _mm512_add_epi64( _mm512_xor_si512( M[13], H[13] ), \
+                        _mm512_xor_si512( M[14], H[14] ) ) )
+
+#define W8b1 \
+   _mm512_add_epi64( \
+       _mm512_add_epi64( \
+          _mm512_sub_epi64( _mm512_xor_si512( M[ 6], H[ 6] ), \
+                            _mm512_xor_si512( M[ 8], H[ 8] ) ), \
+          _mm512_xor_si512( M[11], H[11] ) ), \
+       _mm512_sub_epi64( _mm512_xor_si512( M[14], H[14] ), \
+                         _mm512_xor_si512( M[15], H[15] ) ) )
+
+#define W8b2 \
+   _mm512_sub_epi64( \
+      _mm512_add_epi64( \
+         _mm512_add_epi64( _mm512_xor_si512( M[ 0], H[ 0] ), \
+                           _mm512_xor_si512( M[ 7], H[ 7] ) ), \
+         _mm512_xor_si512( M[ 9], H[ 9] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[12], H[12] ), \
+                        _mm512_xor_si512( M[15], H[15] ) ) )
+
+#define W8b3 \
+   _mm512_sub_epi64( \
+      _mm512_add_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 0], H[ 0] ), \
+                           _mm512_xor_si512( M[ 1], H[ 1] ) ), \
+         _mm512_xor_si512( M[ 8], H[ 8] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[10], H[10] ), \
+                        _mm512_xor_si512( M[13], H[13] ) ) )
+
+#define W8b4 \
+   _mm512_sub_epi64( \
+      _mm512_add_epi64( \
+         _mm512_add_epi64( _mm512_xor_si512( M[ 1], H[ 1] ), \
+                           _mm512_xor_si512( M[ 2], H[ 2] ) ), \
+         _mm512_xor_si512( M[ 9], H[ 9] ) ), \
+      _mm512_add_epi64( _mm512_xor_si512( M[11], H[11] ), \
+                        _mm512_xor_si512( M[14], H[14] ) ) )
+
+#define W8b5 \
+   _mm512_sub_epi64( \
+      _mm512_add_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 3], H[ 3] ), \
+                           _mm512_xor_si512( M[ 2], H[ 2] ) ), \
+         _mm512_xor_si512( M[10], H[10] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[12], H[12] ), \
+                        _mm512_xor_si512( M[15], H[15] ) ) )
+
+#define W8b6 \
+   _mm512_sub_epi64( \
+      _mm512_sub_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 4], H[ 4] ), \
+                           _mm512_xor_si512( M[ 0], H[ 0] ) ), \
+         _mm512_xor_si512( M[ 3], H[ 3] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[11], H[11] ), \
+                        _mm512_xor_si512( M[13], H[13] ) ) )
+
+#define W8b7 \
+   _mm512_sub_epi64( \
+      _mm512_sub_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 1], H[ 1] ), \
+                           _mm512_xor_si512( M[ 4], H[ 4] ) ), \
+         _mm512_xor_si512( M[ 5], H[ 5] ) ), \
+      _mm512_add_epi64( _mm512_xor_si512( M[12], H[12] ), \
+                        _mm512_xor_si512( M[14], H[14] ) ) )
+
+#define W8b8 \
+   _mm512_add_epi64( \
+      _mm512_sub_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 2], H[ 2] ), \
+                           _mm512_xor_si512( M[ 5], H[ 5] ) ), \
+         _mm512_xor_si512( M[ 6], H[ 6] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[13], H[13] ), \
+                        _mm512_xor_si512( M[15], H[15] ) ) )
+
+#define W8b9 \
+   _mm512_sub_epi64( \
+      _mm512_add_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 0], H[ 0] ), \
+                           _mm512_xor_si512( M[ 3], H[ 3] ) ), \
+         _mm512_xor_si512( M[ 6], H[ 6] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[ 7], H[ 7] ), \
+                        _mm512_xor_si512( M[14], H[14] ) ) )
+
+#define W8b10 \
+   _mm512_sub_epi64( \
+      _mm512_sub_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 8], H[ 8] ), \
+                           _mm512_xor_si512( M[ 1], H[ 1] ) ), \
+         _mm512_xor_si512( M[ 4], H[ 4] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[ 7], H[ 7] ), \
+                        _mm512_xor_si512( M[15], H[15] ) ) )
+
+#define W8b11 \
+   _mm512_sub_epi64( \
+      _mm512_sub_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 8], H[ 8] ), \
+                           _mm512_xor_si512( M[ 0], H[ 0] ) ), \
+         _mm512_xor_si512( M[ 2], H[ 2] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[ 5], H[ 5] ), \
+                        _mm512_xor_si512( M[ 9], H[ 9] ) ) )
+
+#define W8b12 \
+   _mm512_sub_epi64( \
+      _mm512_sub_epi64( \
+         _mm512_add_epi64( _mm512_xor_si512( M[ 1], H[ 1] ), \
+                           _mm512_xor_si512( M[ 3], H[ 3] ) ), \
+         _mm512_xor_si512( M[ 6], H[ 6] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[ 9], H[ 9] ), \
+                        _mm512_xor_si512( M[10], H[10] ) ) )
+
+#define W8b13 \
+   _mm512_add_epi64( \
+      _mm512_add_epi64( \
+         _mm512_add_epi64( _mm512_xor_si512( M[ 2], H[ 2] ), \
+                           _mm512_xor_si512( M[ 4], H[ 4] ) ), \
+         _mm512_xor_si512( M[ 7], H[ 7] ) ), \
+      _mm512_add_epi64( _mm512_xor_si512( M[10], H[10] ), \
+                        _mm512_xor_si512( M[11], H[11] ) ) )
+
+#define W8b14 \
+   _mm512_sub_epi64( \
+      _mm512_add_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[ 3], H[ 3] ), \
+                           _mm512_xor_si512( M[ 5], H[ 5] ) ), \
+         _mm512_xor_si512( M[ 8], H[ 8] ) ), \
+      _mm512_add_epi64( _mm512_xor_si512( M[11], H[11] ), \
+                        _mm512_xor_si512( M[12], H[12] ) ) )
+
+#define W8b15 \
+   _mm512_sub_epi64( \
+      _mm512_sub_epi64( \
+         _mm512_sub_epi64( _mm512_xor_si512( M[12], H[12] ), \
+                           _mm512_xor_si512( M[ 4], H[4] ) ), \
+         _mm512_xor_si512( M[ 6], H[ 6] ) ), \
+      _mm512_sub_epi64( _mm512_xor_si512( M[ 9], H[ 9] ), \
+                        _mm512_xor_si512( M[13], H[13] ) ) )
+
+void compress_big_8way( const __m512i *M, const __m512i H[16],
+                        __m512i dH[16] )
+{
+   __m512i qt[32], xl, xh;
+
+   qt[ 0] = _mm512_add_epi64( s8b0( W8b0 ), H[ 1] );
+   qt[ 1] = _mm512_add_epi64( s8b1( W8b1 ), H[ 2] );
+   qt[ 2] = _mm512_add_epi64( s8b2( W8b2 ), H[ 3] );
+   qt[ 3] = _mm512_add_epi64( s8b3( W8b3 ), H[ 4] );
+   qt[ 4] = _mm512_add_epi64( s8b4( W8b4 ), H[ 5] );
+   qt[ 5] = _mm512_add_epi64( s8b0( W8b5 ), H[ 6] );
+   qt[ 6] = _mm512_add_epi64( s8b1( W8b6 ), H[ 7] );
+   qt[ 7] = _mm512_add_epi64( s8b2( W8b7 ), H[ 8] );
+   qt[ 8] = _mm512_add_epi64( s8b3( W8b8 ), H[ 9] );
+   qt[ 9] = _mm512_add_epi64( s8b4( W8b9 ), H[10] );
+   qt[10] = _mm512_add_epi64( s8b0( W8b10), H[11] );
+   qt[11] = _mm512_add_epi64( s8b1( W8b11), H[12] );
+   qt[12] = _mm512_add_epi64( s8b2( W8b12), H[13] );
+   qt[13] = _mm512_add_epi64( s8b3( W8b13), H[14] );
+   qt[14] = _mm512_add_epi64( s8b4( W8b14), H[15] );
+   qt[15] = _mm512_add_epi64( s8b0( W8b15), H[ 0] );
+   qt[16] = expand1b8( qt, M, H, 16 );
+   qt[17] = expand1b8( qt, M, H, 17 );
+   qt[18] = expand2b8( qt, M, H, 18 );
+   qt[19] = expand2b8( qt, M, H, 19 );
+   qt[20] = expand2b8( qt, M, H, 20 );
+   qt[21] = expand2b8( qt, M, H, 21 );
+   qt[22] = expand2b8( qt, M, H, 22 );
+   qt[23] = expand2b8( qt, M, H, 23 );
+   qt[24] = expand2b8( qt, M, H, 24 );
+   qt[25] = expand2b8( qt, M, H, 25 );
+   qt[26] = expand2b8( qt, M, H, 26 );
+   qt[27] = expand2b8( qt, M, H, 27 );
+   qt[28] = expand2b8( qt, M, H, 28 );
+   qt[29] = expand2b8( qt, M, H, 29 );
+   qt[30] = expand2b8( qt, M, H, 30 );
+   qt[31] = expand2b8( qt, M, H, 31 );
+
+   xl = _mm512_xor_si512(
+           mm512_xor4( qt[16], qt[17], qt[18], qt[19] ),
+           mm512_xor4( qt[20], qt[21], qt[22], qt[23] ) );
+   xh = _mm512_xor_si512( xl, _mm512_xor_si512(
+           mm512_xor4( qt[24], qt[25], qt[26], qt[27] ),
+           mm512_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
+
+#define DH1( m, sl, sr, a, b, c ) \
+   _mm512_add_epi64( \
+               _mm512_xor_si512( M[m], \
+                  _mm512_xor_si512( _mm512_slli_epi64( xh, sl ), \
+                                    _mm512_srli_epi64( qt[a], sr ) ) ), \
+               _mm512_xor_si512( _mm512_xor_si512( xl, qt[b] ), qt[c] ) )
+
+#define DHL( m, rl, sl, h, a, b, c ) \
+   _mm512_add_epi64( _mm512_add_epi64( \
+       mm512_rol_64( dH[h], rl ), \
+          _mm512_xor_si512( _mm512_xor_si512( xh, qt[a] ), M[m] )), \
+                 _mm512_xor_si512( _mm512_slli_epi64( xl, sl ), \
+                                   _mm512_xor_si512( qt[b], qt[c] ) ) );
+   
+#define DHR( m, rl, sr, h, a, b, c ) \
+   _mm512_add_epi64( _mm512_add_epi64( \
+       mm512_rol_64( dH[h], rl ), \
+          _mm512_xor_si512( _mm512_xor_si512( xh, qt[a] ), M[m] )), \
+                 _mm512_xor_si512( _mm512_srli_epi64( xl, sr ), \
+                                   _mm512_xor_si512( qt[b], qt[c] ) ) );
+
+
+   dH[ 0] = DH1(  0,  5,  5, 16, 24, 0 );
+   dH[ 1] = DH1(  1,  7,  8, 17, 25, 1 );
+   dH[ 2] = DH1(  2,  5,  5, 18, 26, 2 );
+   dH[ 3] = DH1(  3,  1,  5, 19, 27, 3 );
+   dH[ 4] = DH1(  4,  3,  0, 20, 28, 4 );
+   dH[ 5] = DH1(  5,  6,  6, 21, 29, 5 );
+   dH[ 6] = DH1(  6,  4,  6, 22, 30, 6 );
+   dH[ 7] = DH1(  7, 11,  2, 23, 31, 7 );
+   dH[ 8] = DHL(  8,  9,  8,  4, 24, 23,  8 );
+   dH[ 9] = DHR(  9, 10,  6,  5, 25, 16,  9 );
+   dH[10] = DHL( 10, 11,  6,  6, 26, 17, 10 );
+   dH[11] = DHL( 11, 12,  4,  7, 27, 18, 11 );
+   dH[12] = DHR( 12, 13,  3,  0, 28, 19, 12 );
+   dH[13] = DHR( 13, 14,  4,  1, 29, 20, 13 );
+   dH[14] = DHR( 14, 15,  7,  2, 30, 21, 14 );
+   dH[15] = DHR( 15, 16,  2,  3, 31, 22, 15 );
+
+#undef DH1
+#undef DHL
+#undef DHR
+         
+}
+
+static const __m512i final_b8[16] =
+{
+   { 0xaaaaaaaaaaaaaaa0, 0xaaaaaaaaaaaaaaa0,
+     0xaaaaaaaaaaaaaaa0, 0xaaaaaaaaaaaaaaa0,
+     0xaaaaaaaaaaaaaaa0, 0xaaaaaaaaaaaaaaa0,
+     0xaaaaaaaaaaaaaaa0, 0xaaaaaaaaaaaaaaa0 },
+   { 0xaaaaaaaaaaaaaaa1, 0xaaaaaaaaaaaaaaa1,
+     0xaaaaaaaaaaaaaaa1, 0xaaaaaaaaaaaaaaa1,
+     0xaaaaaaaaaaaaaaa1, 0xaaaaaaaaaaaaaaa1,
+     0xaaaaaaaaaaaaaaa1, 0xaaaaaaaaaaaaaaa1 },
+   { 0xaaaaaaaaaaaaaaa2, 0xaaaaaaaaaaaaaaa2,
+     0xaaaaaaaaaaaaaaa2, 0xaaaaaaaaaaaaaaa2,
+     0xaaaaaaaaaaaaaaa2, 0xaaaaaaaaaaaaaaa2,
+     0xaaaaaaaaaaaaaaa2, 0xaaaaaaaaaaaaaaa2 },
+   { 0xaaaaaaaaaaaaaaa3, 0xaaaaaaaaaaaaaaa3,
+     0xaaaaaaaaaaaaaaa3, 0xaaaaaaaaaaaaaaa3,
+     0xaaaaaaaaaaaaaaa3, 0xaaaaaaaaaaaaaaa3,
+     0xaaaaaaaaaaaaaaa3, 0xaaaaaaaaaaaaaaa3 },
+   { 0xaaaaaaaaaaaaaaa4, 0xaaaaaaaaaaaaaaa4,
+     0xaaaaaaaaaaaaaaa4, 0xaaaaaaaaaaaaaaa4,
+     0xaaaaaaaaaaaaaaa4, 0xaaaaaaaaaaaaaaa4,
+     0xaaaaaaaaaaaaaaa4, 0xaaaaaaaaaaaaaaa4 },
+   { 0xaaaaaaaaaaaaaaa5, 0xaaaaaaaaaaaaaaa5,
+     0xaaaaaaaaaaaaaaa5, 0xaaaaaaaaaaaaaaa5,
+     0xaaaaaaaaaaaaaaa5, 0xaaaaaaaaaaaaaaa5,
+     0xaaaaaaaaaaaaaaa5, 0xaaaaaaaaaaaaaaa5 },
+   { 0xaaaaaaaaaaaaaaa6, 0xaaaaaaaaaaaaaaa6,
+     0xaaaaaaaaaaaaaaa6, 0xaaaaaaaaaaaaaaa6,
+     0xaaaaaaaaaaaaaaa6, 0xaaaaaaaaaaaaaaa6,
+     0xaaaaaaaaaaaaaaa6, 0xaaaaaaaaaaaaaaa6 },
+   { 0xaaaaaaaaaaaaaaa7, 0xaaaaaaaaaaaaaaa7,
+     0xaaaaaaaaaaaaaaa7, 0xaaaaaaaaaaaaaaa7,
+     0xaaaaaaaaaaaaaaa7, 0xaaaaaaaaaaaaaaa7,
+     0xaaaaaaaaaaaaaaa7, 0xaaaaaaaaaaaaaaa7 },
+   { 0xaaaaaaaaaaaaaaa8, 0xaaaaaaaaaaaaaaa8,
+     0xaaaaaaaaaaaaaaa8, 0xaaaaaaaaaaaaaaa8,
+     0xaaaaaaaaaaaaaaa8, 0xaaaaaaaaaaaaaaa8,
+     0xaaaaaaaaaaaaaaa8, 0xaaaaaaaaaaaaaaa8 },
+   { 0xaaaaaaaaaaaaaaa9, 0xaaaaaaaaaaaaaaa9,
+     0xaaaaaaaaaaaaaaa9, 0xaaaaaaaaaaaaaaa9,
+     0xaaaaaaaaaaaaaaa9, 0xaaaaaaaaaaaaaaa9,
+     0xaaaaaaaaaaaaaaa9, 0xaaaaaaaaaaaaaaa9 },
+   { 0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa,
+     0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa,
+     0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa,
+     0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa },
+   { 0xaaaaaaaaaaaaaaab, 0xaaaaaaaaaaaaaaab,
+     0xaaaaaaaaaaaaaaab, 0xaaaaaaaaaaaaaaab,
+     0xaaaaaaaaaaaaaaab, 0xaaaaaaaaaaaaaaab,
+     0xaaaaaaaaaaaaaaab, 0xaaaaaaaaaaaaaaab },
+   { 0xaaaaaaaaaaaaaaac, 0xaaaaaaaaaaaaaaac,
+     0xaaaaaaaaaaaaaaac, 0xaaaaaaaaaaaaaaac,
+     0xaaaaaaaaaaaaaaac, 0xaaaaaaaaaaaaaaac,
+     0xaaaaaaaaaaaaaaac, 0xaaaaaaaaaaaaaaac },
+   { 0xaaaaaaaaaaaaaaad, 0xaaaaaaaaaaaaaaad,
+     0xaaaaaaaaaaaaaaad, 0xaaaaaaaaaaaaaaad,
+     0xaaaaaaaaaaaaaaad, 0xaaaaaaaaaaaaaaad,
+     0xaaaaaaaaaaaaaaad, 0xaaaaaaaaaaaaaaad },
+   { 0xaaaaaaaaaaaaaaae, 0xaaaaaaaaaaaaaaae,
+     0xaaaaaaaaaaaaaaae, 0xaaaaaaaaaaaaaaae,
+     0xaaaaaaaaaaaaaaae, 0xaaaaaaaaaaaaaaae,
+     0xaaaaaaaaaaaaaaae, 0xaaaaaaaaaaaaaaae },
+   { 0xaaaaaaaaaaaaaaaf, 0xaaaaaaaaaaaaaaaf,
+     0xaaaaaaaaaaaaaaaf, 0xaaaaaaaaaaaaaaaf,
+     0xaaaaaaaaaaaaaaaf, 0xaaaaaaaaaaaaaaaf,
+     0xaaaaaaaaaaaaaaaf, 0xaaaaaaaaaaaaaaaf }
+};
+
+
+void bmw512_8way_init( bmw512_8way_context *ctx )
+//bmw64_4way_init( bmw_4way_big_context *sc, const sph_u64 *iv )
+{
+   ctx->H[ 0] = m512_const1_64( 0x8081828384858687 );
+   ctx->H[ 1] = m512_const1_64( 0x88898A8B8C8D8E8F );
+   ctx->H[ 2] = m512_const1_64( 0x9091929394959697 );
+   ctx->H[ 3] = m512_const1_64( 0x98999A9B9C9D9E9F );
+   ctx->H[ 4] = m512_const1_64( 0xA0A1A2A3A4A5A6A7 );
+   ctx->H[ 5] = m512_const1_64( 0xA8A9AAABACADAEAF );
+   ctx->H[ 6] = m512_const1_64( 0xB0B1B2B3B4B5B6B7 );
+   ctx->H[ 7] = m512_const1_64( 0xB8B9BABBBCBDBEBF );
+   ctx->H[ 8] = m512_const1_64( 0xC0C1C2C3C4C5C6C7 );
+   ctx->H[ 9] = m512_const1_64( 0xC8C9CACBCCCDCECF );
+   ctx->H[10] = m512_const1_64( 0xD0D1D2D3D4D5D6D7 );
+   ctx->H[11] = m512_const1_64( 0xD8D9DADBDCDDDEDF );
+   ctx->H[12] = m512_const1_64( 0xE0E1E2E3E4E5E6E7 );
+   ctx->H[13] = m512_const1_64( 0xE8E9EAEBECEDEEEF );
+   ctx->H[14] = m512_const1_64( 0xF0F1F2F3F4F5F6F7 );
+   ctx->H[15] = m512_const1_64( 0xF8F9FAFBFCFDFEFF );
+   ctx->ptr = 0;
+   ctx->bit_count = 0;
+}
+
+void bmw512_8way_update( bmw512_8way_context *ctx, const void *data,
+                                size_t len )
+{
+   __m512i *vdata = (__m512i*)data;
+   __m512i *buf;
+   __m512i htmp[16];
+   __m512i *h1, *h2;
+   size_t ptr;
+   const int buf_size = 128;  // bytes of one lane, compatible with len
+
+   ctx->bit_count += len << 3;
+   buf = ctx->buf;
+   ptr = ctx->ptr;
+   h1 = ctx->H;
+   h2 = htmp;
+   while ( len > 0 )
+   {
+      size_t clen;
+      clen = buf_size - ptr;
+      if ( clen > len )
+         clen = len;
+      memcpy_512( buf + (ptr>>3), vdata, clen >> 3 );
+      vdata = vdata + (clen>>3);
+      len -= clen;
+      ptr += clen;
+      if ( ptr == buf_size )
+      {
+         __m512i *ht;
+         compress_big_8way( buf, h1, h2 );
+         ht = h1;
+         h1 = h2;
+         h2 = ht;
+         ptr = 0;
+      }
+   }
+   ctx->ptr = ptr;
+   if ( h1 != ctx->H )
+        memcpy_512( ctx->H, h1, 16 );
+}
+
+void bmw512_8way_close( bmw512_8way_context *ctx, void *dst )
+{
+   __m512i *buf;
+   __m512i h1[16], h2[16], *h;
+   size_t ptr, u, v;
+   const int buf_size = 128;  // bytes of one lane, compatible with len
+
+   buf = ctx->buf;
+   ptr = ctx->ptr;
+   buf[ ptr>>3 ] = m512_const1_64( 0x80 );
+   ptr += 8;
+   h = ctx->H;
+
+   if (  ptr > (buf_size - 8) )
+   {
+      memset_zero_512( buf + (ptr>>3), (buf_size - ptr) >> 3 );
+      compress_big_8way( buf, h, h1 );
+      ptr = 0;
+      h = h1;
+   }
+   memset_zero_512( buf + (ptr>>3), (buf_size - 8 - ptr) >> 3 );
+   buf[ (buf_size - 8) >> 3 ] = _mm512_set1_epi64( ctx->bit_count );
+   compress_big_8way( buf, h, h2 );
+   for ( u = 0; u < 16; u ++ )
+      buf[ u ] = h2[ u ];
+   compress_big_8way( buf, final_b8, h1 );
+   for (u = 0, v = 8; u < 8; u ++, v ++)
+      casti_m512i( dst, u ) = h1[ v ];
+}
+
+#endif // AVX512
+
 #ifdef __cplusplus
 }
 #endif
--- a/algo/cubehash/cube-hash-2way.c
+++ b/algo/cubehash/cube-hash-2way.c
@@ -7,7 +7,7 @@

 // 2x128

-/*
+
 // The result of hashing 10 rounds of initial data which consists of params
 // zero padded.
 static const uint64_t IV256[] =
@@ -25,7 +25,7 @@ static const uint64_t IV512[] =
 0x148FE485FCD398D9, 0xB64445321B017BEF, 0x2FF5781C6A536159, 0x0DBADEA991FA7934,
 0xA5A70E75D65C8A2B, 0xBC796576B1C62456, 0xE7989AF11921C8F7, 0xD43E3B447795D246
 };
-*/
+

 static void transform_2way( cube_2way_context *sp )
 {
@@ -97,39 +97,30 @@ static void transform_2way( cube_2way_context *sp )
 int cube_2way_init( cube_2way_context *sp, int hashbitlen, int rounds,
                    int blockbytes )
 {
-    __m128i* h = (__m128i*)sp->h;
+    __m256i *h = (__m256i*)sp->h;
+    __m128i *iv = (__m128i*)( hashbitlen == 512 ? (__m128i*)IV512
+                                                : (__m128i*)IV256 );
    sp->hashlen   = hashbitlen/128;
    sp->blocksize = blockbytes/16;
    sp->rounds    = rounds;
    sp->pos       = 0;

-    if ( hashbitlen == 512 )
-    {
-
-       h[ 0] = m128_const_64( 0x4167D83E2D538B8B, 0x50F494D42AEA2A61 );
-       h[ 2] = m128_const_64( 0x50AC5695CC39968E, 0xC701CF8C3FEE2313 );
-       h[ 4] = m128_const_64( 0x825B453797CF0BEF, 0xA647A8B34D42C787 );
-       h[ 6] = m128_const_64( 0xA23911AED0E5CD33, 0xF22090C4EEF864D2 );
-       h[ 8] = m128_const_64( 0xB64445321B017BEF, 0x148FE485FCD398D9 );
-       h[10] = m128_const_64( 0x0DBADEA991FA7934, 0x2FF5781C6A536159 );
-       h[12] = m128_const_64( 0xBC796576B1C62456, 0xA5A70E75D65C8A2B );
-       h[14] = m128_const_64( 0xD43E3B447795D246, 0xE7989AF11921C8F7 );
-       h[1] = h[ 0];  h[ 3] = h[ 2]; h[ 5] = h[ 4]; h[ 7] = h[ 6];
-       h[9] = h[ 8];  h[11] = h[10]; h[13] = h[12]; h[15] = h[14];
-    }
-    else
-    {
-       h[ 0] = m128_const_64( 0x35481EAE63117E71, 0xCCD6F29FEA2BD4B4 );
-       h[ 2] = m128_const_64( 0xF4CC12BE7E624131, 0xE5D94E6322512D5B );
-       h[ 4] = m128_const_64( 0x3361DA8CD0720C35, 0x42AF2070C2D0B696 );
-       h[ 6] = m128_const_64( 0x40E5FBAB4680AC00, 0x8EF8AD8328CCECA4 );
-       h[ 8] = m128_const_64( 0xF0B266796C859D41, 0x6107FBD5D89041C3 );
-       h[10] = m128_const_64( 0x93CB628565C892FD, 0x5FA2560309392549 );
-       h[12] = m128_const_64( 0x85254725774ABFDD, 0x9E4B4E602AF2B5AE );
-       h[14] = m128_const_64( 0xD6032C0A9CDAF8AF, 0x4AB6AAD615815AEB );
-       h[1] = h[ 0];  h[ 3] = h[ 2]; h[ 5] = h[ 4]; h[ 7] = h[ 6];
-       h[9] = h[ 8];  h[11] = h[10]; h[13] = h[12]; h[15] = h[14];
-    }
+    h[ 0] = m256_const1_128( iv[0] );
+    h[ 1] = m256_const1_128( iv[1] );
+    h[ 2] = m256_const1_128( iv[2] );
+    h[ 3] = m256_const1_128( iv[3] );
+    h[ 4] = m256_const1_128( iv[4] );
+    h[ 5] = m256_const1_128( iv[5] );
+    h[ 6] = m256_const1_128( iv[6] );
+    h[ 7] = m256_const1_128( iv[7] );
+    h[ 0] = m256_const1_128( iv[0] );
+    h[ 1] = m256_const1_128( iv[1] );
+    h[ 2] = m256_const1_128( iv[2] );
+    h[ 3] = m256_const1_128( iv[3] );
+    h[ 4] = m256_const1_128( iv[4] );
+    h[ 5] = m256_const1_128( iv[5] );
+    h[ 6] = m256_const1_128( iv[6] );
+    h[ 7] = m256_const1_128( iv[7] );
    
    return 0;
 }
@@ -164,11 +155,11 @@ int cube_2way_close( cube_2way_context *sp, void *output )

    // pos is zero for 64 byte data, 1 for 80 byte data.
    sp->h[ sp->pos ] = _mm256_xor_si256( sp->h[ sp->pos ],
-                                _mm256_set_epi32( 0,0,0,0x80,  0,0,0,0x80 ) );
+                                   m256_const2_64( 0, 0x0000000000000080 ) );
    transform_2way( sp );

    sp->h[7] = _mm256_xor_si256( sp->h[7],
-		                 _mm256_set_epi32( 1,0,0,0,  1,0,0,0 ) );
+                                   m256_const2_64( 0x0000000100000000, 0 ) );

    for ( i = 0; i < 10; ++i )           transform_2way( sp );

@@ -197,13 +188,13 @@ int cube_2way_update_close( cube_2way_context *sp, void *output,

    // pos is zero for 64 byte data, 1 for 80 byte data.
    sp->h[ sp->pos ] = _mm256_xor_si256( sp->h[ sp->pos ],
-                    _mm256_set_epi32( 0,0,0,0x80,  0,0,0,0x80 ) );
+                                    m256_const2_64( 0, 0x0000000000000080 ) );
    transform_2way( sp );

-    sp->h[7] = _mm256_xor_si256( sp->h[7], _mm256_set_epi32( 1,0,0,0,
-                                                             1,0,0,0 ) );
+    sp->h[7] = _mm256_xor_si256( sp->h[7],
+                                    m256_const2_64( 0x0000000100000000, 0 ) );

-    for ( i = 0; i < 10; ++i )            transform_2way( sp );
+    for ( i = 0; i < 10; ++i )    transform_2way( sp );

    memcpy( hash, sp->h, sp->hashlen<<5 );
    return 0;
--- a/algo/hamsi/hamsi-hash-4way.c
+++ b/algo/hamsi/hamsi-hash-4way.c
@@ -32,8 +32,6 @@

 #include <stddef.h>
 #include <string.h>
-
-//#include "miner.h"
 #include "hamsi-hash-4way.h"

 #if defined(__AVX2__)
@@ -100,7 +98,7 @@ extern "C"{
 #endif

 //#include "hamsi-helper-4way.c"
-
+/*
 static const sph_u32 IV512[] = {
 	SPH_C32(0x73746565), SPH_C32(0x6c706172), SPH_C32(0x6b204172),
 	SPH_C32(0x656e6265), SPH_C32(0x72672031), SPH_C32(0x302c2062),
@@ -109,7 +107,7 @@ static const sph_u32 IV512[] = {
 	SPH_C32(0x65766572), SPH_C32(0x6c65652c), SPH_C32(0x2042656c),
 	SPH_C32(0x6769756d)
 };
-
+*/
 static const sph_u32 alpha_n[] = {
 	SPH_C32(0xff00f0f0), SPH_C32(0xccccaaaa), SPH_C32(0xf0f0cccc),
 	SPH_C32(0xff00aaaa), SPH_C32(0xccccaaaa), SPH_C32(0xf0f0ff00),
@@ -138,6 +136,7 @@ static const sph_u32 alpha_f[] = {
 	SPH_C32(0xcaf9f9c0), SPH_C32(0x0ff0639c)
 };

+
 // imported from hamsi helper

 /* Note: this table lists bits within each byte from least
@@ -529,49 +528,34 @@ static const sph_u32 T512[64][16] = {
 	  SPH_C32(0xe7e00a94) }
 };

+
 #define INPUT_BIG \
 do { \
-  const __m256i zero = _mm256_setzero_si256(); \
  __m256i db = *buf; \
-  const sph_u32 *tp = &T512[0][0]; \
-  m0 = zero; \
-  m1 = zero; \
-  m2 = zero; \
-  m3 = zero; \
-  m4 = zero; \
-  m5 = zero; \
-  m6 = zero; \
-  m7 = zero; \
+  const uint64_t *tp = (uint64_t*)&T512[0][0];  \
+  m0 = m1 = m2 = m3 = m4 = m5 = m6 = m7 = m256_zero; \
  for ( int u = 0; u < 64; u++ ) \
  { \
     __m256i dm = _mm256_and_si256( db, m256_one_64 ) ; \
     dm = mm256_negate_32( _mm256_or_si256( dm, \
                         _mm256_slli_epi64( dm, 32 ) ) ); \
     m0 = _mm256_xor_si256( m0, _mm256_and_si256( dm, \
-                  _mm256_set_epi32( tp[0x1], tp[0x0], tp[0x1], tp[0x0], \
-                                    tp[0x1], tp[0x0], tp[0x1], tp[0x0] ) ) ); \
+                                          m256_const1_64( tp[0] ) ) ); \
     m1 = _mm256_xor_si256( m1, _mm256_and_si256( dm, \
-                  _mm256_set_epi32( tp[0x3], tp[0x2], tp[0x3], tp[0x2], \
-                                    tp[0x3], tp[0x2], tp[0x3], tp[0x2] ) ) ); \
+                                          m256_const1_64( tp[1] ) ) ); \
     m2 = _mm256_xor_si256( m2, _mm256_and_si256( dm, \
-                  _mm256_set_epi32( tp[0x5], tp[0x4], tp[0x5], tp[0x4], \
-                                    tp[0x5], tp[0x4], tp[0x5], tp[0x4] ) ) ); \
+                                          m256_const1_64( tp[2] ) ) ); \
     m3 = _mm256_xor_si256( m3, _mm256_and_si256( dm, \
-                  _mm256_set_epi32( tp[0x7], tp[0x6], tp[0x7], tp[0x6], \
-                                    tp[0x7], tp[0x6], tp[0x7], tp[0x6] ) ) ); \
+                                          m256_const1_64( tp[3] ) ) ); \
     m4 = _mm256_xor_si256( m4, _mm256_and_si256( dm, \
-                  _mm256_set_epi32( tp[0x9], tp[0x8], tp[0x9], tp[0x8], \
-                                    tp[0x9], tp[0x8], tp[0x9], tp[0x8] ) ) ); \
+                                          m256_const1_64( tp[4] ) ) ); \
     m5 = _mm256_xor_si256( m5, _mm256_and_si256( dm, \
-                  _mm256_set_epi32( tp[0xB], tp[0xA], tp[0xB], tp[0xA], \
-                                    tp[0xB], tp[0xA], tp[0xB], tp[0xA] ) ) ); \
+                                          m256_const1_64( tp[5] ) ) ); \
     m6 = _mm256_xor_si256( m6, _mm256_and_si256( dm, \
-                  _mm256_set_epi32( tp[0xD], tp[0xC], tp[0xD], tp[0xC], \
-                                    tp[0xD], tp[0xC], tp[0xD], tp[0xC] ) ) ); \
+                                          m256_const1_64( tp[6] ) ) ); \
     m7 = _mm256_xor_si256( m7, _mm256_and_si256( dm, \
-                  _mm256_set_epi32( tp[0xF], tp[0xE], tp[0xF], tp[0xE], \
-                                    tp[0xF], tp[0xE], tp[0xF], tp[0xE] ) ) ); \
-     tp += 0x10; \
+                                          m256_const1_64( tp[7] ) ) ); \
+     tp += 8; \
     db = _mm256_srli_epi64( db, 1 ); \
  } \
 } while (0)
@@ -662,55 +646,39 @@ do { \

 #define ROUND_BIG(rc, alpha) \
 do { \
-  __m256i t0, t1, t2, t3; \
-  s0 = _mm256_xor_si256( s0, _mm256_set_epi32( \
-        alpha[0x01] ^ (rc), alpha[0x00], alpha[0x01] ^ (rc), alpha[0x00], \
-        alpha[0x01] ^ (rc), alpha[0x00], alpha[0x01] ^ (rc), alpha[0x00] ) ); \
-  s1 = _mm256_xor_si256( s1, _mm256_set_epi32( \
-                     alpha[0x03], alpha[0x02], alpha[0x03], alpha[0x02], \
-                     alpha[0x03], alpha[0x02], alpha[0x03], alpha[0x02] ) ); \
-  s2 = _mm256_xor_si256( s2, _mm256_set_epi32( \
-                     alpha[0x05], alpha[0x04], alpha[0x05], alpha[0x04], \
-                     alpha[0x05], alpha[0x04], alpha[0x05], alpha[0x04] ) ); \
-  s3 = _mm256_xor_si256( s3, _mm256_set_epi32( \
-                     alpha[0x07], alpha[0x06], alpha[0x07], alpha[0x06], \
-                     alpha[0x07], alpha[0x06], alpha[0x07], alpha[0x06] ) ); \
-  s4 = _mm256_xor_si256( s4, _mm256_set_epi32( \
-                     alpha[0x09], alpha[0x08], alpha[0x09], alpha[0x08], \
-                     alpha[0x09], alpha[0x08], alpha[0x09], alpha[0x08] ) ); \
-  s5 = _mm256_xor_si256( s5, _mm256_set_epi32( \
-                     alpha[0x0B], alpha[0x0A], alpha[0x0B], alpha[0x0A], \
-                     alpha[0x0B], alpha[0x0A], alpha[0x0B], alpha[0x0A] ) ); \
-  s6 = _mm256_xor_si256( s6, _mm256_set_epi32( \
-                     alpha[0x0D], alpha[0x0C], alpha[0x0D], alpha[0x0C], \
-                     alpha[0x0D], alpha[0x0C], alpha[0x0D], alpha[0x0C] ) ); \
-  s7 = _mm256_xor_si256( s7, _mm256_set_epi32( \
-                     alpha[0x0F], alpha[0x0E], alpha[0x0F], alpha[0x0E], \
-                     alpha[0x0F], alpha[0x0E], alpha[0x0F], alpha[0x0E] ) ); \
-  s8 = _mm256_xor_si256( s8, _mm256_set_epi32( \
-                     alpha[0x11], alpha[0x10], alpha[0x11], alpha[0x10], \
-                     alpha[0x11], alpha[0x10], alpha[0x11], alpha[0x10] ) ); \
-  s9 = _mm256_xor_si256( s9, _mm256_set_epi32( \
-                     alpha[0x13], alpha[0x12], alpha[0x13], alpha[0x12], \
-                     alpha[0x13], alpha[0x12], alpha[0x13], alpha[0x12] ) ); \
-  sA = _mm256_xor_si256( sA, _mm256_set_epi32( \
-                     alpha[0x15], alpha[0x14], alpha[0x15], alpha[0x14], \
-                     alpha[0x15], alpha[0x14], alpha[0x15], alpha[0x14] ) ); \
-  sB = _mm256_xor_si256( sB, _mm256_set_epi32( \
-                     alpha[0x17], alpha[0x16], alpha[0x17], alpha[0x16], \
-                     alpha[0x17], alpha[0x16], alpha[0x17], alpha[0x16] ) ); \
-  sC = _mm256_xor_si256( sC, _mm256_set_epi32( \
-                     alpha[0x19], alpha[0x18], alpha[0x19], alpha[0x18], \
-                     alpha[0x19], alpha[0x18], alpha[0x19], alpha[0x18] ) ); \
-  sD = _mm256_xor_si256( sD, _mm256_set_epi32( \
-                     alpha[0x1B], alpha[0x1A], alpha[0x1B], alpha[0x1A], \
-                     alpha[0x1B], alpha[0x1A], alpha[0x1B], alpha[0x1A] ) ); \
-  sE = _mm256_xor_si256( sE, _mm256_set_epi32( \
-                     alpha[0x1D], alpha[0x1C], alpha[0x1D], alpha[0x1C], \
-                     alpha[0x1D], alpha[0x1C], alpha[0x1D], alpha[0x1C] ) ); \
-  sF = _mm256_xor_si256( sF, _mm256_set_epi32( \
-                     alpha[0x1F], alpha[0x1E], alpha[0x1F], alpha[0x1E], \
-                     alpha[0x1F], alpha[0x1E], alpha[0x1F], alpha[0x1E] ) ); \
+   __m256i t0, t1, t2, t3; \
+   s0 = _mm256_xor_si256( s0, m256_const1_64( \
+        ( ( (uint64_t)( (rc) ^ alpha[1] ) << 32 ) ) | (uint64_t)alpha[0] ) ); \
+   s1 = _mm256_xor_si256( s1, m256_const1_64( \
+        ( (uint64_t)alpha[ 3] << 32 ) | (uint64_t)alpha[ 2] ) ); \
+   s2 = _mm256_xor_si256( s2, m256_const1_64( \
+        ( (uint64_t)alpha[ 5] << 32 ) | (uint64_t)alpha[ 4] ) ); \
+   s3 = _mm256_xor_si256( s3, m256_const1_64( \
+        ( (uint64_t)alpha[ 7] << 32 ) | (uint64_t)alpha[ 6] ) ); \
+   s4 = _mm256_xor_si256( s4, m256_const1_64( \
+        ( (uint64_t)alpha[ 9] << 32 ) | (uint64_t)alpha[ 8] ) ); \
+   s5 = _mm256_xor_si256( s5, m256_const1_64( \
+        ( (uint64_t)alpha[11] << 32 ) | (uint64_t)alpha[10] ) ); \
+   s6 = _mm256_xor_si256( s6, m256_const1_64( \
+        ( (uint64_t)alpha[13] << 32 ) | (uint64_t)alpha[12] ) ); \
+   s7 = _mm256_xor_si256( s7, m256_const1_64( \
+        ( (uint64_t)alpha[15] << 32 ) | (uint64_t)alpha[14] ) ); \
+   s8 = _mm256_xor_si256( s8, m256_const1_64( \
+        ( (uint64_t)alpha[17] << 32 ) | (uint64_t)alpha[16] ) ); \
+   s9 = _mm256_xor_si256( s9, m256_const1_64( \
+        ( (uint64_t)alpha[19] << 32 ) | (uint64_t)alpha[18] ) ); \
+   sA = _mm256_xor_si256( sA, m256_const1_64( \
+        ( (uint64_t)alpha[21] << 32 ) | (uint64_t)alpha[20] ) ); \
+   sB = _mm256_xor_si256( sB, m256_const1_64( \
+        ( (uint64_t)alpha[23] << 32 ) | (uint64_t)alpha[22] ) ); \
+   sC = _mm256_xor_si256( sC, m256_const1_64( \
+        ( (uint64_t)alpha[25] << 32 ) | (uint64_t)alpha[24] ) ); \
+   sD = _mm256_xor_si256( sD, m256_const1_64( \
+        ( (uint64_t)alpha[27] << 32 ) | (uint64_t)alpha[26] ) ); \
+   sE = _mm256_xor_si256( sE, m256_const1_64( \
+        ( (uint64_t)alpha[29] << 32 ) | (uint64_t)alpha[28] ) ); \
+   sF = _mm256_xor_si256( sF, m256_const1_64( \
+        ( (uint64_t)alpha[31] << 32 ) | (uint64_t)alpha[30] ) ); \
 \
  SBOX( s0, s4, s8, sC ); \
  SBOX( s1, s5, s9, sD ); \
@@ -864,47 +832,22 @@ void hamsi_big_final( hamsi_4way_big_context *sc, __m256i *buf )
 void hamsi512_4way_init( hamsi_4way_big_context *sc )
 {
   sc->partial_len = 0;
-   sph_u32 lo, hi;
   sc->count_high = sc->count_low = 0;
-   for ( int i = 0; i < 8; i++ )
-   {
-      lo = 2*i;
-      hi = 2*i + 1;
-      sc->h[i] = _mm256_set_epi32( IV512[hi], IV512[lo], IV512[hi], IV512[lo],
-                                   IV512[hi], IV512[lo], IV512[hi], IV512[lo] );
-   }
+
+   sc->h[0] = m256_const1_64( 0x6c70617273746565 );
+   sc->h[1] = m256_const1_64( 0x656e62656b204172 );
+   sc->h[2] = m256_const1_64( 0x302c206272672031 );
+   sc->h[3] = m256_const1_64( 0x3434362c75732032 );
+   sc->h[4] = m256_const1_64( 0x3030312020422d33 );
+   sc->h[5] = m256_const1_64( 0x656e2d484c657576 );
+   sc->h[6] = m256_const1_64( 0x6c65652c65766572 );
+   sc->h[7] = m256_const1_64( 0x6769756d2042656c );
 }

 void hamsi512_4way( hamsi_4way_big_context *sc, const void *data, size_t len )
 {
   __m256i *vdata = (__m256i*)data;

-// It looks like the only way to get in here is if core was previously called
-// with a very small len
-// That's not likely even with 80 byte input so deprecate partial len
-/*
-   if ( sc->partial_len != 0 )
-   {
-      size_t mlen;
-
-      mlen = 8 - sc->partial_len;
-      if ( len < mlen )
-      {
-         memcpy_256( sc->partial + (sc->partial_len >> 3), data, len>>3 );
-         sc->partial_len += len;
-         return;
-      }
-      else
-      {
-         memcpy_256( sc->partial + (sc->partial_len >> 3), data, mlen>>3 );
-         len -= mlen;
-         vdata += mlen>>3;
-         hamsi_big( sc, sc->partial, 1 );
-         sc->partial_len = 0;
-      }
-   }
-*/
-
   hamsi_big( sc, vdata, len>>3 );
   vdata += ( (len& ~(size_t)7) >> 3 );
   len &= (size_t)7;
@@ -920,8 +863,9 @@ void hamsi512_4way_close( hamsi_4way_big_context *sc, void *dst )
   sph_enc32be( &ch, sc->count_high );
   sph_enc32be( &cl, sc->count_low + ( sc->partial_len << 3 ) );
   pad[0] =  _mm256_set_epi32( cl, ch, cl, ch, cl, ch, cl, ch );
-   sc->buf[0] = _mm256_set_epi32( 0UL, 0x80UL, 0UL, 0x80UL,
-                                  0UL, 0x80UL, 0UL, 0x80UL );
+   sc->buf[0] = m256_const1_64( 0x80 );
+//      sc->buf[0] = _mm256_set_epi32( 0UL, 0x80UL, 0UL, 0x80UL,
+//                                  0UL, 0x80UL, 0UL, 0x80UL );
   hamsi_big( sc, sc->buf, 1 );
   hamsi_big_final( sc, pad );

--- a/algo/jh/jh-hash-4way.c
+++ b/algo/jh/jh-hash-4way.c
@@ -94,7 +94,7 @@ extern "C"{

 #define Sb(x0, x1, x2, x3, c) \
 do { \
-   __m256i cc = _mm256_set_epi64x( c, c, c, c ); \
+   __m256i cc = _mm256_set1_epi64x( c ); \
    x3 = mm256_not( x3 ); \
    x0 = _mm256_xor_si256( x0, _mm256_andnot_si256( x2, cc ) ); \
    tmp = _mm256_xor_si256( cc, _mm256_and_si256( x0, x1 ) ); \
--- a/algo/keccak/keccak-4way.c
+++ b/algo/keccak/keccak-4way.c
@@ -1,18 +1,68 @@
 #include "keccak-gate.h"
-
-#ifdef KECCAK_4WAY
-
 #include <stdlib.h>
 #include <string.h>
 #include <stdint.h>
 #include "sph_keccak.h"
 #include "keccak-hash-4way.h"

+#if defined(KECCAK_8WAY)
+
+void keccakhash_8way(void *state, const void *input)
+{
+    keccak256_8way_context ctx;
+    keccak256_8way_init( &ctx );
+    keccak256_8way_update( &ctx, input, 80 );
+    keccak256_8way_close( &ctx, state );
+}
+
+int scanhash_keccak_8way( struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t vdata[24*8] __attribute__ ((aligned (128)));
+   uint32_t hash[16*8] __attribute__ ((aligned (64)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
+   uint32_t *hash7 = &(hash[49]);   // 3*16+1
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   uint32_t n = pdata[19];
+   const uint32_t first_nonce = pdata[19];
+   __m512i  *noncev = (__m512i*)vdata + 9;   // aligned
+   const uint32_t Htarg = ptarget[7];
+   int thr_id = mythr->id;  
+
+   mm512_bswap32_intrlv80_8x64( vdata, pdata );
+   do {
+       *noncev = mm512_intrlv_blend_32( mm512_bswap_32(
+                _mm512_set_epi32( n+7, 0, n+6, 0, n+5, 0, n+4, 0,
+                                  n+3, 0, n+2, 0, n+1, 0, n  , 0 ) ), *noncev );
+
+      keccakhash_8way( hash, vdata );
+
+      for ( int lane = 0; lane < 8; lane++ )
+      if ( hash7[ lane<<1 ] < Htarg ) 
+      {
+          extr_lane_8x64( lane_hash, hash, lane, 256 );
+          if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
+          {
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
+          }
+      }
+      n += 8;
+
+   } while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+#elif defined(KECCAK_4WAY)
+
 void keccakhash_4way(void *state, const void *input)
 {
    keccak256_4way_context ctx;
    keccak256_4way_init( &ctx );
-    keccak256_4way( &ctx, input, 80 );
+    keccak256_4way_update( &ctx, input, 80 );
    keccak256_4way_close( &ctx, state );
 }

@@ -28,8 +78,8 @@ int scanhash_keccak_4way( struct work *work, uint32_t max_nonce,
   uint32_t n = pdata[19];
   const uint32_t first_nonce = pdata[19];
   __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
-//   const uint32_t Htarg = ptarget[7];
-    int thr_id = mythr->id;  // thr_id arg is deprecated
+   const uint32_t Htarg = ptarget[7];
+   int thr_id = mythr->id;

   mm256_bswap32_intrlv80_4x64( vdata, pdata );
   do {
@@ -39,7 +89,7 @@ int scanhash_keccak_4way( struct work *work, uint32_t max_nonce,
      keccakhash_4way( hash, vdata );

      for ( int lane = 0; lane < 4; lane++ )
-      if ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 )
+      if ( hash7[ lane<<1 ] < Htarg )
      {
          extr_lane_4x64( lane_hash, hash, lane, 256 );
          if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
--- a/algo/keccak/keccak-gate.c
+++ b/algo/keccak/keccak-gate.c
@@ -3,30 +3,36 @@

 bool register_keccak_algo( algo_gate_t* gate )
 {
-  gate->optimizations = AVX2_OPT;
+  gate->optimizations = AVX2_OPT | AVX512_OPT;
  gate->gen_merkle_root = (void*)&SHA256_gen_merkle_root;
  opt_target_factor = 128.0;
-#if defined (KECCAK_4WAY)
+#if defined (KECCAK_8WAY)
+  gate->scanhash  = (void*)&scanhash_keccak_8way;
+  gate->hash      = (void*)&keccakhash_8way;
+#elif defined (KECCAK_4WAY)
  gate->scanhash  = (void*)&scanhash_keccak_4way;
  gate->hash      = (void*)&keccakhash_4way;
 #else
-  gate->scanhash        = (void*)&scanhash_keccak;
-  gate->hash            = (void*)&keccakhash;
+  gate->scanhash  = (void*)&scanhash_keccak;
+  gate->hash      = (void*)&keccakhash;
 #endif
  return true;
 };

 bool register_keccakc_algo( algo_gate_t* gate )
 {
-  gate->optimizations = AVX2_OPT;
+  gate->optimizations = AVX2_OPT | AVX512_OPT;
  gate->gen_merkle_root = (void*)&sha256d_gen_merkle_root;
  opt_target_factor = 256.0;
-#if defined (KECCAK_4WAY)
+#if defined (KECCAK_8WAY)
+  gate->scanhash  = (void*)&scanhash_keccak_8way;
+  gate->hash      = (void*)&keccakhash_8way;
+#elif defined (KECCAK_4WAY)
  gate->scanhash  = (void*)&scanhash_keccak_4way;
  gate->hash      = (void*)&keccakhash_4way;
 #else
-  gate->scanhash        = (void*)&scanhash_keccak;
-  gate->hash            = (void*)&keccakhash;
+  gate->scanhash  = (void*)&scanhash_keccak;
+  gate->hash      = (void*)&keccakhash;
 #endif
  return true;
 };
--- a/algo/keccak/keccak-gate.h
+++ b/algo/keccak/keccak-gate.h
@@ -1,23 +1,33 @@
 #ifndef KECCAK_GATE_H__
-#define KECCAK_GATE_H__
+#define KECCAK_GATE_H__ 1

 #include "algo-gate-api.h"
 #include <stdint.h>

-#if defined(__AVX2__)
-  #define KECCAK_4WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define KECCAK_8WAY 1
+#elif defined(__AVX2__)
+  #define KECCAK_4WAY 1
 #endif

-#if defined(KECCAK_4WAY)
+#if defined(KECCAK_8WAY)
+
+void keccakhash_8way( void *state, const void *input );
+int scanhash_keccak_8way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+
+#elif defined(KECCAK_4WAY)

 void keccakhash_4way( void *state, const void *input );
 int scanhash_keccak_4way( struct work *work, uint32_t max_nonce,
                         uint64_t *hashes_done, struct thr_info *mythr );

-#endif
+#else

 void keccakhash( void *state, const void *input );
 int scanhash_keccak( struct work *work, uint32_t max_nonce,
                    uint64_t *hashes_done, struct thr_info *mythr );

 #endif
+
+#endif
--- a/algo/keccak/keccak-hash-4way.c
+++ b/algo/keccak/keccak-hash-4way.c
@@ -1,23 +1,24 @@
 #include <stddef.h>
+#include <stdint.h>
 #include "keccak-hash-4way.h"

-#if defined(__AVX2__)
-
-static const sph_u64 RC[] = {
-        SPH_C64(0x0000000000000001), SPH_C64(0x0000000000008082),
-        SPH_C64(0x800000000000808A), SPH_C64(0x8000000080008000),
-        SPH_C64(0x000000000000808B), SPH_C64(0x0000000080000001),
-        SPH_C64(0x8000000080008081), SPH_C64(0x8000000000008009),
-        SPH_C64(0x000000000000008A), SPH_C64(0x0000000000000088),
-        SPH_C64(0x0000000080008009), SPH_C64(0x000000008000000A),
-        SPH_C64(0x000000008000808B), SPH_C64(0x800000000000008B),
-        SPH_C64(0x8000000000008089), SPH_C64(0x8000000000008003),
-        SPH_C64(0x8000000000008002), SPH_C64(0x8000000000000080),
-        SPH_C64(0x000000000000800A), SPH_C64(0x800000008000000A),
-        SPH_C64(0x8000000080008081), SPH_C64(0x8000000000008080),
-        SPH_C64(0x0000000080000001), SPH_C64(0x8000000080008008)
+static const uint64_t RC[] = {
+        0x0000000000000001, 0x0000000000008082,
+        0x800000000000808A, 0x8000000080008000,
+        0x000000000000808B, 0x0000000080000001,
+        0x8000000080008081, 0x8000000000008009,
+        0x000000000000008A, 0x0000000000000088,
+        0x0000000080008009, 0x000000008000000A,
+        0x000000008000808B, 0x800000000000008B,
+        0x8000000000008089, 0x8000000000008003,
+        0x8000000000008002, 0x8000000000000080,
+        0x000000000000800A, 0x800000008000000A,
+        0x8000000080008081, 0x8000000000008080,
+        0x0000000080000001, 0x8000000080008008
 };

+// generic macros
+
 #define a00   (kc->w[ 0])
 #define a10   (kc->w[ 1])
 #define a20   (kc->w[ 2])
@@ -48,6 +49,197 @@ static const sph_u64 RC[] = {
 #define READ_STATE(sc)
 #define WRITE_STATE(sc)

+#define MOV64(d, s)      (d = s)
+#define XOR64_IOTA       XOR64
+
+#define LPAR   (
+#define RPAR   )
+
+#define DO(x)   x
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define INPUT_BUF(size)   do { \
+    size_t j; \
+    for (j = 0; j < (size>>3); j++ ) \
+        kc->w[j ] = _mm512_xor_si512( kc->w[j], buf[j] ); \
+} while (0)
+
+// Targetted macros, keccak-macros.h is included for each target.
+
+#define DECL64(x)        __m512i x
+#define XOR64(d, a, b)   (d = _mm512_xor_si512(a,b))
+#define AND64(d, a, b)   (d = _mm512_and_si512(a,b))
+#define OR64(d, a, b)    (d = _mm512_or_si512(a,b))
+#define NOT64(d, s)      (d = _mm512_xor_si512(s,m512_neg1))
+#define ROL64(d, v, n)   (d = mm512_rol_64(v, n))
+
+#include "keccak-macros.c"
+
+#define KECCAK_F_1600   DO(KECCAK_F_1600_512)
+
+#define KECCAK_F_1600_512   do { \
+    int j; \
+    for (j = 0; j < 24; j += 8) \
+    { \
+       KF_ELT( 0,  1, _mm512_set1_epi64( RC[j + 0] ) ); \
+       KF_ELT( 1,  2, _mm512_set1_epi64( RC[j + 1] ) ); \
+       KF_ELT( 2,  3, _mm512_set1_epi64( RC[j + 2] ) ); \
+       KF_ELT( 3,  4, _mm512_set1_epi64( RC[j + 3] ) ); \
+       KF_ELT( 4,  5, _mm512_set1_epi64( RC[j + 4] ) ); \
+       KF_ELT( 5,  6, _mm512_set1_epi64( RC[j + 5] ) ); \
+       KF_ELT( 6,  7, _mm512_set1_epi64( RC[j + 6] ) ); \
+       KF_ELT( 7,  8, _mm512_set1_epi64( RC[j + 7] ) ); \
+       P8_TO_P0; \
+    } \
+} while (0)
+
+static void keccak64_8way_init( keccak64_ctx_m512i *kc, unsigned out_size )
+{
+   __m512i zero = m512_zero;
+   __m512i neg1 = m512_neg1;
+
+   // Initialization for the "lane complement".
+   kc->w[ 0] = zero;   kc->w[ 1] = neg1;
+   kc->w[ 2] = neg1;   kc->w[ 3] = zero;
+   kc->w[ 4] = zero;   kc->w[ 5] = zero;
+   kc->w[ 6] = zero;   kc->w[ 7] = zero;
+   kc->w[ 8] = neg1;   kc->w[ 9] = zero;
+   kc->w[10] = zero;   kc->w[11] = zero;
+   kc->w[12] = neg1;   kc->w[13] = zero;
+   kc->w[14] = zero;   kc->w[15] = zero;
+   kc->w[16] = zero;   kc->w[17] = neg1;
+   kc->w[18] = zero;   kc->w[19] = zero;
+   kc->w[20] = neg1;   kc->w[21] = zero;
+   kc->w[22] = zero;   kc->w[23] = zero;
+   kc->w[24] = zero;   kc->ptr = 0;
+   kc->lim = 200 - (out_size >> 2);
+}
+
+static void
+keccak64_8way_core( keccak64_ctx_m512i *kc, const void *data, size_t len,
+               size_t lim )
+{
+    __m512i *buf;
+    __m512i *vdata = (__m512i*)data;
+    size_t ptr;
+    DECL_STATE
+
+    buf = kc->buf;
+    ptr = kc->ptr;
+
+    if ( len < (lim - ptr) )
+    {
+        memcpy_512( buf + (ptr>>3), vdata, len>>3 );
+        kc->ptr = ptr + len;
+        return;
+    }
+    READ_STATE( kc );
+    while ( len > 0 )
+    {
+        size_t clen;
+
+        clen = (lim - ptr);
+        if ( clen > len )
+             clen = len;
+        memcpy_512( buf + (ptr>>3), vdata, clen>>3 );
+        ptr += clen;
+        vdata = vdata + (clen>>3);
+        len -= clen;
+        if ( ptr == lim )
+        {
+            INPUT_BUF( lim );
+            KECCAK_F_1600;
+            ptr = 0;
+        }
+    }
+    WRITE_STATE( kc );
+    kc->ptr = ptr;
+}
+
+static void keccak64_8way_close( keccak64_ctx_m512i *kc, void *dst,
+                                 size_t byte_len, size_t lim )
+{
+    unsigned eb;
+    union {
+       __m512i tmp[lim + 1];
+       sph_u64 dummy;   /* for alignment */
+    } u;
+    size_t j;
+    size_t m512_len = byte_len >> 3;
+
+    eb = 0x100  >> 8;
+    if ( kc->ptr == (lim - 8) )
+    {
+        const uint64_t t = eb | 0x8000000000000000;
+        u.tmp[0] = m512_const1_64( t );
+        j = 8;
+    }
+    else
+    {
+        j = lim - kc->ptr;
+        u.tmp[0] = m512_const1_64( eb );
+        memset_zero_512( u.tmp + 1, (j>>3) - 2 );
+        u.tmp[ (j>>3) - 1] = m512_const1_64( 0x8000000000000000 );
+    }
+    keccak64_8way_core( kc, u.tmp, j, lim );
+    /* Finalize the "lane complement" */
+    NOT64( kc->w[ 1], kc->w[ 1] );
+    NOT64( kc->w[ 2], kc->w[ 2] );
+    NOT64( kc->w[ 8], kc->w[ 8] );
+    NOT64( kc->w[12], kc->w[12] );
+    NOT64( kc->w[17], kc->w[17] );
+    NOT64( kc->w[20], kc->w[20] );
+    memcpy_512( dst, kc->w, m512_len );
+}
+
+void keccak256_8way_init( void *kc )
+{
+   keccak64_8way_init( kc, 256 );
+}
+
+void
+keccak256_8way_update(void *cc, const void *data, size_t len)
+{
+    keccak64_8way_core(cc, data, len, 136);
+}
+
+void
+keccak256_8way_close(void *cc, void *dst)
+{
+    keccak64_8way_close(cc, dst, 32, 136);
+}
+
+void keccak512_8way_init( void *kc )
+{
+   keccak64_8way_init( kc, 512 );
+}
+
+void
+keccak512_8way_update(void *cc, const void *data, size_t len)
+{
+        keccak64_8way_core(cc, data, len, 72);
+}
+
+void
+keccak512_8way_close(void *cc, void *dst)
+{
+        keccak64_8way_close(cc, dst, 64, 72);
+}
+
+#undef INPUT_BUF
+#undef DECL64
+#undef XOR64
+#undef AND64
+#undef OR64
+#undef NOT64
+#undef ROL64
+#undef KECCAK_F_1600
+
+#endif  // AVX512
+
+#if defined(__AVX2__)
+
 #define INPUT_BUF(size)   do { \
    size_t j; \
    for (j = 0; j < (size>>3); j++ ) \
@@ -55,314 +247,28 @@ static const sph_u64 RC[] = {
 } while (0)

 #define DECL64(x)        __m256i x
-#define MOV64(d, s)      (d = s)
 #define XOR64(d, a, b)   (d = _mm256_xor_si256(a,b))
 #define AND64(d, a, b)   (d = _mm256_and_si256(a,b))
 #define OR64(d, a, b)    (d = _mm256_or_si256(a,b))
 #define NOT64(d, s)      (d = _mm256_xor_si256(s,m256_neg1))
 #define ROL64(d, v, n)   (d = mm256_rol_64(v, n))
-#define XOR64_IOTA       XOR64

-#define TH_ELT(t, c0, c1, c2, c3, c4, d0, d1, d2, d3, d4)   do { \
-                DECL64(tt0); \
-                DECL64(tt1); \
-                DECL64(tt2); \
-                DECL64(tt3); \
-                XOR64(tt0, d0, d1); \
-                XOR64(tt1, d2, d3); \
-                XOR64(tt0, tt0, d4); \
-                XOR64(tt0, tt0, tt1); \
-                ROL64(tt0, tt0, 1); \
-                XOR64(tt2, c0, c1); \
-                XOR64(tt3, c2, c3); \
-                XOR64(tt0, tt0, c4); \
-                XOR64(tt2, tt2, tt3); \
-                XOR64(t, tt0, tt2); \
-        } while (0)
+#include "keccak-macros.c"

-#define THETA(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
-        b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
-        b40, b41, b42, b43, b44) \
-        do { \
-                DECL64(t0); \
-                DECL64(t1); \
-                DECL64(t2); \
-                DECL64(t3); \
-                DECL64(t4); \
-                TH_ELT(t0, b40, b41, b42, b43, b44, b10, b11, b12, b13, b14); \
-                TH_ELT(t1, b00, b01, b02, b03, b04, b20, b21, b22, b23, b24); \
-                TH_ELT(t2, b10, b11, b12, b13, b14, b30, b31, b32, b33, b34); \
-                TH_ELT(t3, b20, b21, b22, b23, b24, b40, b41, b42, b43, b44); \
-                TH_ELT(t4, b30, b31, b32, b33, b34, b00, b01, b02, b03, b04); \
-                XOR64(b00, b00, t0); \
-                XOR64(b01, b01, t0); \
-                XOR64(b02, b02, t0); \
-                XOR64(b03, b03, t0); \
-                XOR64(b04, b04, t0); \
-                XOR64(b10, b10, t1); \
-                XOR64(b11, b11, t1); \
-                XOR64(b12, b12, t1); \
-                XOR64(b13, b13, t1); \
-                XOR64(b14, b14, t1); \
-                XOR64(b20, b20, t2); \
-                XOR64(b21, b21, t2); \
-                XOR64(b22, b22, t2); \
-                XOR64(b23, b23, t2); \
-                XOR64(b24, b24, t2); \
-                XOR64(b30, b30, t3); \
-                XOR64(b31, b31, t3); \
-                XOR64(b32, b32, t3); \
-                XOR64(b33, b33, t3); \
-                XOR64(b34, b34, t3); \
-                XOR64(b40, b40, t4); \
-                XOR64(b41, b41, t4); \
-                XOR64(b42, b42, t4); \
-                XOR64(b43, b43, t4); \
-                XOR64(b44, b44, t4); \
-        } while (0)
+#define KECCAK_F_1600   DO(KECCAK_F_1600_256)

-#define RHO(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
-        b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
-        b40, b41, b42, b43, b44) \
-        do { \
-                /* ROL64(b00, b00,  0); */ \
-                ROL64(b01, b01, 36); \
-                ROL64(b02, b02,  3); \
-                ROL64(b03, b03, 41); \
-                ROL64(b04, b04, 18); \
-                ROL64(b10, b10,  1); \
-                ROL64(b11, b11, 44); \
-                ROL64(b12, b12, 10); \
-                ROL64(b13, b13, 45); \
-                ROL64(b14, b14,  2); \
-                ROL64(b20, b20, 62); \
-                ROL64(b21, b21,  6); \
-                ROL64(b22, b22, 43); \
-                ROL64(b23, b23, 15); \
-                ROL64(b24, b24, 61); \
-                ROL64(b30, b30, 28); \
-                ROL64(b31, b31, 55); \
-                ROL64(b32, b32, 25); \
-                ROL64(b33, b33, 21); \
-                ROL64(b34, b34, 56); \
-                ROL64(b40, b40, 27); \
-                ROL64(b41, b41, 20); \
-                ROL64(b42, b42, 39); \
-                ROL64(b43, b43,  8); \
-                ROL64(b44, b44, 14); \
-        } while (0)
-
-/*
- * The KHI macro integrates the "lane complement" optimization. On input,
- * some words are complemented:
- *    a00 a01 a02 a04 a13 a20 a21 a22 a30 a33 a34 a43
- * On output, the following words are complemented:
- *    a04 a10 a20 a22 a23 a31
- *
- * The (implicit) permutation and the theta expansion will bring back
- * the input mask for the next round.
- */
-
-#define KHI_XO(d, a, b, c)   do { \
-                DECL64(kt); \
-                OR64(kt, b, c); \
-                XOR64(d, a, kt); \
-        } while (0)
-
-#define KHI_XA(d, a, b, c)   do { \
-                DECL64(kt); \
-                AND64(kt, b, c); \
-                XOR64(d, a, kt); \
-        } while (0)
-
-#define KHI(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
-        b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
-        b40, b41, b42, b43, b44) \
-        do { \
-                DECL64(c0); \
-                DECL64(c1); \
-                DECL64(c2); \
-                DECL64(c3); \
-                DECL64(c4); \
-                DECL64(bnn); \
-                NOT64(bnn, b20); \
-                KHI_XO(c0, b00, b10, b20); \
-                KHI_XO(c1, b10, bnn, b30); \
-                KHI_XA(c2, b20, b30, b40); \
-                KHI_XO(c3, b30, b40, b00); \
-                KHI_XA(c4, b40, b00, b10); \
-                MOV64(b00, c0); \
-                MOV64(b10, c1); \
-                MOV64(b20, c2); \
-                MOV64(b30, c3); \
-                MOV64(b40, c4); \
-                NOT64(bnn, b41); \
-                KHI_XO(c0, b01, b11, b21); \
-                KHI_XA(c1, b11, b21, b31); \
-                KHI_XO(c2, b21, b31, bnn); \
-                KHI_XO(c3, b31, b41, b01); \
-                KHI_XA(c4, b41, b01, b11); \
-                MOV64(b01, c0); \
-                MOV64(b11, c1); \
-                MOV64(b21, c2); \
-                MOV64(b31, c3); \
-                MOV64(b41, c4); \
-                NOT64(bnn, b32); \
-                KHI_XO(c0, b02, b12, b22); \
-                KHI_XA(c1, b12, b22, b32); \
-                KHI_XA(c2, b22, bnn, b42); \
-                KHI_XO(c3, bnn, b42, b02); \
-                KHI_XA(c4, b42, b02, b12); \
-                MOV64(b02, c0); \
-                MOV64(b12, c1); \
-                MOV64(b22, c2); \
-                MOV64(b32, c3); \
-                MOV64(b42, c4); \
-                NOT64(bnn, b33); \
-                KHI_XA(c0, b03, b13, b23); \
-                KHI_XO(c1, b13, b23, b33); \
-                KHI_XO(c2, b23, bnn, b43); \
-                KHI_XA(c3, bnn, b43, b03); \
-                KHI_XO(c4, b43, b03, b13); \
-                MOV64(b03, c0); \
-                MOV64(b13, c1); \
-                MOV64(b23, c2); \
-                MOV64(b33, c3); \
-                MOV64(b43, c4); \
-                NOT64(bnn, b14); \
-                KHI_XA(c0, b04, bnn, b24); \
-                KHI_XO(c1, bnn, b24, b34); \
-                KHI_XA(c2, b24, b34, b44); \
-                KHI_XO(c3, b34, b44, b04); \
-                KHI_XA(c4, b44, b04, b14); \
-                MOV64(b04, c0); \
-                MOV64(b14, c1); \
-                MOV64(b24, c2); \
-                MOV64(b34, c3); \
-                MOV64(b44, c4); \
-        } while (0)
-
-#define IOTA(r)   XOR64_IOTA(a00, a00, r)
-
-#define P0    a00, a01, a02, a03, a04, a10, a11, a12, a13, a14, a20, a21, \
-              a22, a23, a24, a30, a31, a32, a33, a34, a40, a41, a42, a43, a44
-#define P1    a00, a30, a10, a40, a20, a11, a41, a21, a01, a31, a22, a02, \
-              a32, a12, a42, a33, a13, a43, a23, a03, a44, a24, a04, a34, a14
-#define P2    a00, a33, a11, a44, a22, a41, a24, a02, a30, a13, a32, a10, \
-              a43, a21, a04, a23, a01, a34, a12, a40, a14, a42, a20, a03, a31
-#define P3    a00, a23, a41, a14, a32, a24, a42, a10, a33, a01, a43, a11, \
-              a34, a02, a20, a12, a30, a03, a21, a44, a31, a04, a22, a40, a13
-#define P4    a00, a12, a24, a31, a43, a42, a04, a11, a23, a30, a34, a41, \
-              a03, a10, a22, a21, a33, a40, a02, a14, a13, a20, a32, a44, a01
-#define P5    a00, a21, a42, a13, a34, a04, a20, a41, a12, a33, a03, a24, \
-              a40, a11, a32, a02, a23, a44, a10, a31, a01, a22, a43, a14, a30
-#define P6    a00, a02, a04, a01, a03, a20, a22, a24, a21, a23, a40, a42, \
-              a44, a41, a43, a10, a12, a14, a11, a13, a30, a32, a34, a31, a33
-#define P7    a00, a10, a20, a30, a40, a22, a32, a42, a02, a12, a44, a04, \
-              a14, a24, a34, a11, a21, a31, a41, a01, a33, a43, a03, a13, a23
-#define P8    a00, a11, a22, a33, a44, a32, a43, a04, a10, a21, a14, a20, \
-              a31, a42, a03, a41, a02, a13, a24, a30, a23, a34, a40, a01, a12
-#define P9    a00, a41, a32, a23, a14, a43, a34, a20, a11, a02, a31, a22, \
-              a13, a04, a40, a24, a10, a01, a42, a33, a12, a03, a44, a30, a21
-#define P10   a00, a24, a43, a12, a31, a34, a03, a22, a41, a10, a13, a32, \
-              a01, a20, a44, a42, a11, a30, a04, a23, a21, a40, a14, a33, a02
-#define P11   a00, a42, a34, a21, a13, a03, a40, a32, a24, a11, a01, a43, \
-              a30, a22, a14, a04, a41, a33, a20, a12, a02, a44, a31, a23, a10
-#define P12   a00, a04, a03, a02, a01, a40, a44, a43, a42, a41, a30, a34, \
-              a33, a32, a31, a20, a24, a23, a22, a21, a10, a14, a13, a12, a11
-#define P13   a00, a20, a40, a10, a30, a44, a14, a34, a04, a24, a33, a03, \
-              a23, a43, a13, a22, a42, a12, a32, a02, a11, a31, a01, a21, a41
-#define P14   a00, a22, a44, a11, a33, a14, a31, a03, a20, a42, a23, a40, \
-              a12, a34, a01, a32, a04, a21, a43, a10, a41, a13, a30, a02, a24
-#define P15   a00, a32, a14, a41, a23, a31, a13, a40, a22, a04, a12, a44, \
-              a21, a03, a30, a43, a20, a02, a34, a11, a24, a01, a33, a10, a42
-#define P16   a00, a43, a31, a24, a12, a13, a01, a44, a32, a20, a21, a14, \
-              a02, a40, a33, a34, a22, a10, a03, a41, a42, a30, a23, a11, a04
-#define P17   a00, a34, a13, a42, a21, a01, a30, a14, a43, a22, a02, a31, \
-              a10, a44, a23, a03, a32, a11, a40, a24, a04, a33, a12, a41, a20
-#define P18   a00, a03, a01, a04, a02, a30, a33, a31, a34, a32, a10, a13, \
-              a11, a14, a12, a40, a43, a41, a44, a42, a20, a23, a21, a24, a22
-#define P19   a00, a40, a30, a20, a10, a33, a23, a13, a03, a43, a11, a01, \
-              a41, a31, a21, a44, a34, a24, a14, a04, a22, a12, a02, a42, a32
-#define P20   a00, a44, a33, a22, a11, a23, a12, a01, a40, a34, a41, a30, \
-              a24, a13, a02, a14, a03, a42, a31, a20, a32, a21, a10, a04, a43
-#define P21   a00, a14, a23, a32, a41, a12, a21, a30, a44, a03, a24, a33, \
-              a42, a01, a10, a31, a40, a04, a13, a22, a43, a02, a11, a20, a34
-#define P22   a00, a31, a12, a43, a24, a21, a02, a33, a14, a40, a42, a23, \
-              a04, a30, a11, a13, a44, a20, a01, a32, a34, a10, a41, a22, a03
-#define P23   a00, a13, a21, a34, a42, a02, a10, a23, a31, a44, a04, a12, \
-              a20, a33, a41, a01, a14, a22, a30, a43, a03, a11, a24, a32, a40
-
-#define P8_TO_P0   do { \
-                DECL64(t); \
-                MOV64(t, a01); \
-                MOV64(a01, a11); \
-                MOV64(a11, a43); \
-                MOV64(a43, t); \
-                MOV64(t, a02); \
-                MOV64(a02, a22); \
-                MOV64(a22, a31); \
-                MOV64(a31, t); \
-                MOV64(t, a03); \
-                MOV64(a03, a33); \
-                MOV64(a33, a24); \
-                MOV64(a24, t); \
-                MOV64(t, a04); \
-                MOV64(a04, a44); \
-                MOV64(a44, a12); \
-                MOV64(a12, t); \
-                MOV64(t, a10); \
-                MOV64(a10, a32); \
-                MOV64(a32, a13); \
-                MOV64(a13, t); \
-                MOV64(t, a14); \
-                MOV64(a14, a21); \
-                MOV64(a21, a20); \
-                MOV64(a20, t); \
-                MOV64(t, a23); \
-                MOV64(a23, a42); \
-                MOV64(a42, a40); \
-                MOV64(a40, t); \
-                MOV64(t, a30); \
-                MOV64(a30, a41); \
-                MOV64(a41, a34); \
-                MOV64(a34, t); \
-        } while (0)
-
-#define LPAR   (
-#define RPAR   )
-
-#define KF_ELT(r, s, k)   do { \
-                THETA LPAR P ## r RPAR; \
-                RHO LPAR P ## r RPAR; \
-                KHI LPAR P ## s RPAR; \
-                IOTA(k); \
-        } while (0)
-
-#define DO(x)   x
-
-#define KECCAK_F_1600   DO(KECCAK_F_1600_)
-
-#define KECCAK_F_1600_   do { \
+#define KECCAK_F_1600_256   do { \
    int j; \
    for (j = 0; j < 24; j += 8) \
    { \
-       KF_ELT( 0,  1, (_mm256_set_epi64x( RC[j + 0], RC[j + 0], \
-                                       RC[j + 0], RC[j + 0])) ); \
-       KF_ELT( 1,  2, (_mm256_set_epi64x( RC[j + 1], RC[j + 1], \
-                                       RC[j + 1], RC[j + 1])) ); \
-       KF_ELT( 2,  3, (_mm256_set_epi64x( RC[j + 2], RC[j + 2], \
-                                       RC[j + 2], RC[j + 2])) ); \
-       KF_ELT( 3,  4, (_mm256_set_epi64x( RC[j + 3], RC[j + 3], \
-                                       RC[j + 3], RC[j + 3])) ); \
-       KF_ELT( 4,  5, (_mm256_set_epi64x( RC[j + 4], RC[j + 4], \
-                                       RC[j + 4], RC[j + 4])) ); \
-       KF_ELT( 5,  6, (_mm256_set_epi64x( RC[j + 5], RC[j + 5], \
-                                       RC[j + 5], RC[j + 5])) ); \
-       KF_ELT( 6,  7, (_mm256_set_epi64x( RC[j + 6], RC[j + 6], \
-                                       RC[j + 6], RC[j + 6])) ); \
-       KF_ELT( 7,  8, (_mm256_set_epi64x( RC[j + 7], RC[j + 7], \
-                                       RC[j + 7], RC[j + 7])) ); \
+       KF_ELT( 0,  1, _mm256_set1_epi64x( RC[j + 0] ) ); \
+       KF_ELT( 1,  2, _mm256_set1_epi64x( RC[j + 1] ) ); \
+       KF_ELT( 2,  3, _mm256_set1_epi64x( RC[j + 2] ) ); \
+       KF_ELT( 3,  4, _mm256_set1_epi64x( RC[j + 3] ) ); \
+       KF_ELT( 4,  5, _mm256_set1_epi64x( RC[j + 4] ) ); \
+       KF_ELT( 5,  6, _mm256_set1_epi64x( RC[j + 5] ) ); \
+       KF_ELT( 6,  7, _mm256_set1_epi64x( RC[j + 6] ) ); \
+       KF_ELT( 7,  8, _mm256_set1_epi64x( RC[j + 7] ) ); \
       P8_TO_P0; \
    } \
 } while (0)
@@ -453,7 +359,7 @@ static void keccak64_close( keccak64_ctx_m256i *kc, void *dst, size_t byte_len,
    else
    {
        j = lim - kc->ptr;
-        u.tmp[0] = _mm256_set_epi64x( eb, eb, eb, eb );
+        u.tmp[0] = m256_const1_64( eb );
        memset_zero_256( u.tmp + 1, (j>>3) - 2 );
        u.tmp[ (j>>3) - 1] = m256_const1_64( 0x8000000000000000 );
    }
@@ -474,7 +380,7 @@ void keccak256_4way_init( void *kc )
 }

 void
-keccak256_4way(void *cc, const void *data, size_t len)
+keccak256_4way_update(void *cc, const void *data, size_t len)
 {
    keccak64_core(cc, data, len, 136);
 }
@@ -491,15 +397,24 @@ void keccak512_4way_init( void *kc )
 }

 void
-keccak512_4way(void *cc, const void *data, size_t len)
+keccak512_4way_update(void *cc, const void *data, size_t len)
 {
-        keccak64_core(cc, data, len, 72);
+   keccak64_core(cc, data, len, 72);
 }

 void
 keccak512_4way_close(void *cc, void *dst)
 {
-        keccak64_close(cc, dst, 64, 72);
+   keccak64_close(cc, dst, 64, 72);
 }

-#endif
+#undef INPUT_BUF
+#undef DECL64
+#undef XOR64
+#undef AND64
+#undef OR64
+#undef NOT64
+#undef ROL64
+#undef KECCAK_F_1600
+
+#endif  // AVX2
--- a/algo/keccak/keccak-hash-4way.h
+++ b/algo/keccak/keccak-hash-4way.h
@@ -64,26 +64,49 @@ extern "C"{
 * <code>memcpy()</code>).
 */

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 typedef struct {
-        __m256i buf[144*8];    /* first field, for alignment */
+        __m512i buf[144*8];
+        __m512i w[25];
+        size_t ptr, lim;
+} keccak64_ctx_m512i __attribute__((aligned(128)));
+
+typedef keccak64_ctx_m512i keccak256_8way_context;
+typedef keccak64_ctx_m512i keccak512_8way_context;
+
+void keccak256_8way_init(void *cc);
+void keccak256_8way_update(void *cc, const void *data, size_t len);
+void keccak256_8way_close(void *cc, void *dst);
+
+void keccak512_8way_init(void *cc);
+void keccak512_8way_update(void *cc, const void *data, size_t len);
+void keccak512_8way_close(void *cc, void *dst);
+void keccak512_8way_addbits_and_close(
+        void *cc, unsigned ub, unsigned n, void *dst);
+
+#endif   
+
+typedef struct {
+        __m256i buf[144*8];  
        __m256i w[25];
        size_t ptr, lim;
-//        sph_u64 wide[25];
-} keccak64_ctx_m256i;
+} keccak64_ctx_m256i __attribute__((aligned(128)));

 typedef keccak64_ctx_m256i keccak256_4way_context;
 typedef keccak64_ctx_m256i keccak512_4way_context;

 void keccak256_4way_init(void *cc);
-void keccak256_4way(void *cc, const void *data, size_t len);
+void keccak256_4way_update(void *cc, const void *data, size_t len);
 void keccak256_4way_close(void *cc, void *dst);
-
+#define keccak256_4way keccak256_4way_update

 void keccak512_4way_init(void *cc);
-void keccak512_4way(void *cc, const void *data, size_t len);
+void keccak512_4way_update(void *cc, const void *data, size_t len);
 void keccak512_4way_close(void *cc, void *dst);
 void keccak512_4way_addbits_and_close(
        void *cc, unsigned ub, unsigned n, void *dst);
+#define keccak512_4way keccak512_4way_update

 #endif

--- a/algo/keccak/keccak-macros.c
+++ b/algo/keccak/keccak-macros.c
@@ -0,0 +1,324 @@
+#ifdef TH_ELT
+#undef TH_ELT
+#endif
+#define TH_ELT(t, c0, c1, c2, c3, c4, d0, d1, d2, d3, d4)   do { \
+                DECL64(tt0); \
+                DECL64(tt1); \
+                DECL64(tt2); \
+                DECL64(tt3); \
+                XOR64(tt0, d0, d1); \
+                XOR64(tt1, d2, d3); \
+                XOR64(tt0, tt0, d4); \
+                XOR64(tt0, tt0, tt1); \
+                ROL64(tt0, tt0, 1); \
+                XOR64(tt2, c0, c1); \
+                XOR64(tt3, c2, c3); \
+                XOR64(tt0, tt0, c4); \
+                XOR64(tt2, tt2, tt3); \
+                XOR64(t, tt0, tt2); \
+        } while (0)
+
+#ifdef THETA
+#undef THETA
+#endif
+#define THETA(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
+        b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
+        b40, b41, b42, b43, b44) \
+        do { \
+                DECL64(t0); \
+                DECL64(t1); \
+                DECL64(t2); \
+                DECL64(t3); \
+                DECL64(t4); \
+                TH_ELT(t0, b40, b41, b42, b43, b44, b10, b11, b12, b13, b14); \
+                TH_ELT(t1, b00, b01, b02, b03, b04, b20, b21, b22, b23, b24); \
+                TH_ELT(t2, b10, b11, b12, b13, b14, b30, b31, b32, b33, b34); \
+                TH_ELT(t3, b20, b21, b22, b23, b24, b40, b41, b42, b43, b44); \
+                TH_ELT(t4, b30, b31, b32, b33, b34, b00, b01, b02, b03, b04); \
+                XOR64(b00, b00, t0); \
+                XOR64(b01, b01, t0); \
+                XOR64(b02, b02, t0); \
+                XOR64(b03, b03, t0); \
+                XOR64(b04, b04, t0); \
+                XOR64(b10, b10, t1); \
+                XOR64(b11, b11, t1); \
+                XOR64(b12, b12, t1); \
+                XOR64(b13, b13, t1); \
+                XOR64(b14, b14, t1); \
+                XOR64(b20, b20, t2); \
+                XOR64(b21, b21, t2); \
+                XOR64(b22, b22, t2); \
+                XOR64(b23, b23, t2); \
+                XOR64(b24, b24, t2); \
+                XOR64(b30, b30, t3); \
+                XOR64(b31, b31, t3); \
+                XOR64(b32, b32, t3); \
+                XOR64(b33, b33, t3); \
+                XOR64(b34, b34, t3); \
+                XOR64(b40, b40, t4); \
+                XOR64(b41, b41, t4); \
+                XOR64(b42, b42, t4); \
+                XOR64(b43, b43, t4); \
+                XOR64(b44, b44, t4); \
+        } while (0)
+
+#ifdef RHO
+#undef RHO
+#endif
+#define RHO(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
+        b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
+        b40, b41, b42, b43, b44) \
+        do { \
+                /* ROL64(b00, b00,  0); */ \
+                ROL64(b01, b01, 36); \
+                ROL64(b02, b02,  3); \
+                ROL64(b03, b03, 41); \
+                ROL64(b04, b04, 18); \
+                ROL64(b10, b10,  1); \
+                ROL64(b11, b11, 44); \
+                ROL64(b12, b12, 10); \
+                ROL64(b13, b13, 45); \
+                ROL64(b14, b14,  2); \
+                ROL64(b20, b20, 62); \
+                ROL64(b21, b21,  6); \
+                ROL64(b22, b22, 43); \
+                ROL64(b23, b23, 15); \
+                ROL64(b24, b24, 61); \
+                ROL64(b30, b30, 28); \
+                ROL64(b31, b31, 55); \
+                ROL64(b32, b32, 25); \
+                ROL64(b33, b33, 21); \
+                ROL64(b34, b34, 56); \
+                ROL64(b40, b40, 27); \
+                ROL64(b41, b41, 20); \
+                ROL64(b42, b42, 39); \
+                ROL64(b43, b43,  8); \
+                ROL64(b44, b44, 14); \
+        } while (0)
+
+/*
+ * The KHI macro integrates the "lane complement" optimization. On input,
+ * some words are complemented:
+ *    a00 a01 a02 a04 a13 a20 a21 a22 a30 a33 a34 a43
+ * On output, the following words are complemented:
+ *    a04 a10 a20 a22 a23 a31
+ *
+ * The (implicit) permutation and the theta expansion will bring back
+ * the input mask for the next round.
+ */
+
+#ifdef KHI_XO
+#undef KHI_XO
+#endif
+#define KHI_XO(d, a, b, c)   do { \
+                DECL64(kt); \
+                OR64(kt, b, c); \
+                XOR64(d, a, kt); \
+        } while (0)
+
+#ifdef KHI_XA
+#undef KHI_XA
+#endif
+#define KHI_XA(d, a, b, c)   do { \
+                DECL64(kt); \
+                AND64(kt, b, c); \
+                XOR64(d, a, kt); \
+        } while (0)
+
+#ifdef KHI
+#undef KHI
+#endif
+#define KHI(b00, b01, b02, b03, b04, b10, b11, b12, b13, b14, \
+        b20, b21, b22, b23, b24, b30, b31, b32, b33, b34, \
+        b40, b41, b42, b43, b44) \
+        do { \
+                DECL64(c0); \
+                DECL64(c1); \
+                DECL64(c2); \
+                DECL64(c3); \
+                DECL64(c4); \
+                DECL64(bnn); \
+                NOT64(bnn, b20); \
+                KHI_XO(c0, b00, b10, b20); \
+                KHI_XO(c1, b10, bnn, b30); \
+                KHI_XA(c2, b20, b30, b40); \
+                KHI_XO(c3, b30, b40, b00); \
+                KHI_XA(c4, b40, b00, b10); \
+                MOV64(b00, c0); \
+                MOV64(b10, c1); \
+                MOV64(b20, c2); \
+                MOV64(b30, c3); \
+                MOV64(b40, c4); \
+                NOT64(bnn, b41); \
+                KHI_XO(c0, b01, b11, b21); \
+                KHI_XA(c1, b11, b21, b31); \
+                KHI_XO(c2, b21, b31, bnn); \
+                KHI_XO(c3, b31, b41, b01); \
+                KHI_XA(c4, b41, b01, b11); \
+                MOV64(b01, c0); \
+                MOV64(b11, c1); \
+                MOV64(b21, c2); \
+                MOV64(b31, c3); \
+                MOV64(b41, c4); \
+                NOT64(bnn, b32); \
+                KHI_XO(c0, b02, b12, b22); \
+                KHI_XA(c1, b12, b22, b32); \
+                KHI_XA(c2, b22, bnn, b42); \
+                KHI_XO(c3, bnn, b42, b02); \
+                KHI_XA(c4, b42, b02, b12); \
+                MOV64(b02, c0); \
+                MOV64(b12, c1); \
+                MOV64(b22, c2); \
+                MOV64(b32, c3); \
+                MOV64(b42, c4); \
+                NOT64(bnn, b33); \
+                KHI_XA(c0, b03, b13, b23); \
+                KHI_XO(c1, b13, b23, b33); \
+                KHI_XO(c2, b23, bnn, b43); \
+                KHI_XA(c3, bnn, b43, b03); \
+                KHI_XO(c4, b43, b03, b13); \
+                MOV64(b03, c0); \
+                MOV64(b13, c1); \
+                MOV64(b23, c2); \
+                MOV64(b33, c3); \
+                MOV64(b43, c4); \
+                NOT64(bnn, b14); \
+                KHI_XA(c0, b04, bnn, b24); \
+                KHI_XO(c1, bnn, b24, b34); \
+                KHI_XA(c2, b24, b34, b44); \
+                KHI_XO(c3, b34, b44, b04); \
+                KHI_XA(c4, b44, b04, b14); \
+                MOV64(b04, c0); \
+                MOV64(b14, c1); \
+                MOV64(b24, c2); \
+                MOV64(b34, c3); \
+                MOV64(b44, c4); \
+        } while (0)
+
+#ifdef IOTA
+#undef IOTA
+#endif
+#define IOTA(r)   XOR64_IOTA(a00, a00, r)
+
+#ifdef P0
+#undef P1
+#undef P2
+#undef P3
+#undef P4
+#undef P5
+#undef P6
+#undef P7
+#undef P8
+#undef P9
+#undef P10
+#undef p11
+#undef P12
+#undef P13
+#undef P14
+#undef P15
+#undef P16
+#undef P17
+#undef P18
+#undef P19
+#undef P20
+#undef P21
+#undef P22
+#undef P23
+#endif
+
+#define P0    a00, a01, a02, a03, a04, a10, a11, a12, a13, a14, a20, a21, \
+              a22, a23, a24, a30, a31, a32, a33, a34, a40, a41, a42, a43, a44
+#define P1    a00, a30, a10, a40, a20, a11, a41, a21, a01, a31, a22, a02, \
+              a32, a12, a42, a33, a13, a43, a23, a03, a44, a24, a04, a34, a14
+#define P2    a00, a33, a11, a44, a22, a41, a24, a02, a30, a13, a32, a10, \
+              a43, a21, a04, a23, a01, a34, a12, a40, a14, a42, a20, a03, a31
+#define P3    a00, a23, a41, a14, a32, a24, a42, a10, a33, a01, a43, a11, \
+              a34, a02, a20, a12, a30, a03, a21, a44, a31, a04, a22, a40, a13
+#define P4    a00, a12, a24, a31, a43, a42, a04, a11, a23, a30, a34, a41, \
+              a03, a10, a22, a21, a33, a40, a02, a14, a13, a20, a32, a44, a01
+#define P5    a00, a21, a42, a13, a34, a04, a20, a41, a12, a33, a03, a24, \
+              a40, a11, a32, a02, a23, a44, a10, a31, a01, a22, a43, a14, a30
+#define P6    a00, a02, a04, a01, a03, a20, a22, a24, a21, a23, a40, a42, \
+              a44, a41, a43, a10, a12, a14, a11, a13, a30, a32, a34, a31, a33
+#define P7    a00, a10, a20, a30, a40, a22, a32, a42, a02, a12, a44, a04, \
+              a14, a24, a34, a11, a21, a31, a41, a01, a33, a43, a03, a13, a23
+#define P8    a00, a11, a22, a33, a44, a32, a43, a04, a10, a21, a14, a20, \
+              a31, a42, a03, a41, a02, a13, a24, a30, a23, a34, a40, a01, a12
+#define P9    a00, a41, a32, a23, a14, a43, a34, a20, a11, a02, a31, a22, \
+              a13, a04, a40, a24, a10, a01, a42, a33, a12, a03, a44, a30, a21
+#define P10   a00, a24, a43, a12, a31, a34, a03, a22, a41, a10, a13, a32, \
+              a01, a20, a44, a42, a11, a30, a04, a23, a21, a40, a14, a33, a02
+#define P11   a00, a42, a34, a21, a13, a03, a40, a32, a24, a11, a01, a43, \
+              a30, a22, a14, a04, a41, a33, a20, a12, a02, a44, a31, a23, a10
+#define P12   a00, a04, a03, a02, a01, a40, a44, a43, a42, a41, a30, a34, \
+              a33, a32, a31, a20, a24, a23, a22, a21, a10, a14, a13, a12, a11
+#define P13   a00, a20, a40, a10, a30, a44, a14, a34, a04, a24, a33, a03, \
+              a23, a43, a13, a22, a42, a12, a32, a02, a11, a31, a01, a21, a41
+#define P14   a00, a22, a44, a11, a33, a14, a31, a03, a20, a42, a23, a40, \
+              a12, a34, a01, a32, a04, a21, a43, a10, a41, a13, a30, a02, a24
+#define P15   a00, a32, a14, a41, a23, a31, a13, a40, a22, a04, a12, a44, \
+              a21, a03, a30, a43, a20, a02, a34, a11, a24, a01, a33, a10, a42
+#define P16   a00, a43, a31, a24, a12, a13, a01, a44, a32, a20, a21, a14, \
+              a02, a40, a33, a34, a22, a10, a03, a41, a42, a30, a23, a11, a04
+#define P17   a00, a34, a13, a42, a21, a01, a30, a14, a43, a22, a02, a31, \
+              a10, a44, a23, a03, a32, a11, a40, a24, a04, a33, a12, a41, a20
+#define P18   a00, a03, a01, a04, a02, a30, a33, a31, a34, a32, a10, a13, \
+              a11, a14, a12, a40, a43, a41, a44, a42, a20, a23, a21, a24, a22
+#define P19   a00, a40, a30, a20, a10, a33, a23, a13, a03, a43, a11, a01, \
+              a41, a31, a21, a44, a34, a24, a14, a04, a22, a12, a02, a42, a32
+#define P20   a00, a44, a33, a22, a11, a23, a12, a01, a40, a34, a41, a30, \
+              a24, a13, a02, a14, a03, a42, a31, a20, a32, a21, a10, a04, a43
+#define P21   a00, a14, a23, a32, a41, a12, a21, a30, a44, a03, a24, a33, \
+              a42, a01, a10, a31, a40, a04, a13, a22, a43, a02, a11, a20, a34
+#define P22   a00, a31, a12, a43, a24, a21, a02, a33, a14, a40, a42, a23, \
+              a04, a30, a11, a13, a44, a20, a01, a32, a34, a10, a41, a22, a03
+#define P23   a00, a13, a21, a34, a42, a02, a10, a23, a31, a44, a04, a12, \
+              a20, a33, a41, a01, a14, a22, a30, a43, a03, a11, a24, a32, a40
+
+#ifdef P8_TO_P0
+#undef P8_TO_P0
+#endif
+#define P8_TO_P0   do { \
+                DECL64(t); \
+                MOV64(t, a01); \
+                MOV64(a01, a11); \
+                MOV64(a11, a43); \
+                MOV64(a43, t); \
+                MOV64(t, a02); \
+                MOV64(a02, a22); \
+                MOV64(a22, a31); \
+                MOV64(a31, t); \
+                MOV64(t, a03); \
+                MOV64(a03, a33); \
+                MOV64(a33, a24); \
+                MOV64(a24, t); \
+                MOV64(t, a04); \
+                MOV64(a04, a44); \
+                MOV64(a44, a12); \
+                MOV64(a12, t); \
+                MOV64(t, a10); \
+                MOV64(a10, a32); \
+                MOV64(a32, a13); \
+                MOV64(a13, t); \
+                MOV64(t, a14); \
+                MOV64(a14, a21); \
+                MOV64(a21, a20); \
+                MOV64(a20, t); \
+                MOV64(t, a23); \
+                MOV64(a23, a42); \
+                MOV64(a42, a40); \
+                MOV64(a40, t); \
+                MOV64(t, a30); \
+                MOV64(a30, a41); \
+                MOV64(a41, a34); \
+                MOV64(a34, t); \
+        } while (0)
+
+#define KF_ELT(r, s, k)   do { \
+                THETA LPAR P ## r RPAR; \
+                RHO LPAR P ## r RPAR; \
+                KHI LPAR P ## s RPAR; \
+                IOTA(k); \
+        } while (0)
+
+
--- a/algo/lanehash/lane.c
+++ b/algo/lanehash/lane.c
--- a/algo/lanehash/lane.h
+++ b/algo/lanehash/lane.h
@@ -0,0 +1,50 @@
+/*
+ * Copyright (c) 2008 Sebastiaan Indesteege
+ *                              <sebastiaan.indesteege@esat.kuleuven.be>
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+/*
+ * Optimised ANSI-C implementation of LANE
+ */
+
+#ifndef LANE_H
+#define LANE_H
+
+#include <string.h>
+//#include "algo/sha/sha3-defs.h"
+#include <stdint.h>
+
+typedef unsigned char BitSequence;
+typedef unsigned long long DataLength;
+
+//typedef enum { SUCCESS = 0, FAIL = 1, BAD_HASHBITLEN = 2, BAD_DATABITLEN = 3 } HashReturn;
+
+//typedef unsigned char u8;
+//typedef unsigned int u32;
+//typedef unsigned long long u64;
+
+typedef struct {
+  int hashbitlen;
+  uint64_t ctr;
+  uint32_t h[16];
+  uint8_t buffer[128];
+} hashState;
+
+void laneInit (hashState *state, int hashbitlen);
+void laneUpdate (hashState *state, const BitSequence *data, DataLength databitlen);
+void laneFinal (hashState *state, BitSequence *hashval);
+void laneHash (int hashbitlen, const BitSequence *data, DataLength databitlen, BitSequence *hashval);
+
+#endif /* LANE_H */
--- a/algo/luffa/luffa-hash-2way.c
+++ b/algo/luffa/luffa-hash-2way.c
@@ -1,23 +1,3 @@
-/*
- * luffa_for_sse2.c
- * Version 2.0 (Sep 15th 2009)
- *
- * Copyright (C) 2008-2009 Hitachi, Ltd. All rights reserved.
- *
- * Hitachi, Ltd. is the owner of this software and hereby grant
- * the U.S. Government and any interested party the right to use
- * this software for the purposes of the SHA-3 evaluation process,
- * notwithstanding that this software is copyrighted.
- *
- * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
- * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
- * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
- * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
- * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
- * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
- */
-
 #include <string.h>
 #include <immintrin.h>
 #include "luffa-hash-2way.h"
@@ -26,31 +6,30 @@

 #include "simd-utils.h"

-#define MASK _mm256_set_epi32( 0UL, 0UL, 0UL, 0xffffffffUL, \
-                               0UL, 0UL, 0UL, 0xffffffffUL )
+#define cns(i)  m256_const1_128( ( (__m128i*)CNS_INIT)[i] )

 #define ADD_CONSTANT(a,b,c0,c1)\
    a = _mm256_xor_si256(a,c0);\
    b = _mm256_xor_si256(b,c1);\

-#define MULT2(a0,a1) \
+#define MULT2( a0, a1, mask ) \
 do { \
-  register __m256i b = _mm256_xor_si256( a0, \
-                   _mm256_shuffle_epi32( _mm256_and_si256(a1,MASK), 16 ) ); \
+  __m256i b = _mm256_xor_si256( a0, \
+                   _mm256_shuffle_epi32( _mm256_and_si256(a1,mask), 16 ) ); \
  a0 = _mm256_or_si256( _mm256_srli_si256(b,4), _mm256_slli_si256(a1,12) ); \
  a1 = _mm256_or_si256( _mm256_srli_si256(a1,4), _mm256_slli_si256(b,12) );  \
 } while(0)

 // confirm pointer arithmetic
 // ok but use array indexes
-#define STEP_PART(x,c,t)\
+#define STEP_PART(x,c0,c1,t)\
    SUBCRUMB(*x,*(x+1),*(x+2),*(x+3),*t);\
    SUBCRUMB(*(x+5),*(x+6),*(x+7),*(x+4),*t);\
    MIXWORD(*x,*(x+4),*t,*(t+1));\
    MIXWORD(*(x+1),*(x+5),*t,*(t+1));\
    MIXWORD(*(x+2),*(x+6),*t,*(t+1));\
    MIXWORD(*(x+3),*(x+7),*t,*(t+1));\
-    ADD_CONSTANT(*x, *(x+4), *c, *(c+1));
+    ADD_CONSTANT(*x, *(x+4), c0, c1);

 #define SUBCRUMB(a0,a1,a2,a3,t)\
    t  = _mm256_load_si256(&a0);\
@@ -245,7 +224,7 @@ static const uint32 CNS_INIT[128] __attribute((aligned(32))) = {
    0x00000000,0x00000000,0x00000000,0xfc053c31
 };

-__m256i CNS[32];
+

 /***************************************************/
 /* Round function         */
@@ -258,6 +237,7 @@ void rnd512_2way( luffa_2way_context *state, __m256i *msg )
    __m256i msg0, msg1;
    __m256i tmp[2];
    __m256i x[8];
+    const __m256i MASK = m256_const2_64( 0, 0x00000000ffffffff );

    t0 = chainv[0];
    t1 = chainv[1];
@@ -271,7 +251,7 @@ void rnd512_2way( luffa_2way_context *state, __m256i *msg )
    t0 = _mm256_xor_si256( t0, chainv[8] );
    t1 = _mm256_xor_si256( t1, chainv[9] );

-    MULT2( t0, t1 );
+    MULT2( t0, t1, MASK );

    msg0 = _mm256_shuffle_epi32( msg[0], 27 );
    msg1 = _mm256_shuffle_epi32( msg[1], 27 );
@@ -290,66 +270,66 @@ void rnd512_2way( luffa_2way_context *state, __m256i *msg )
    t0 = chainv[0];
    t1 = chainv[1];

-    MULT2( chainv[0], chainv[1]);
+    MULT2( chainv[0], chainv[1], MASK );
    chainv[0] = _mm256_xor_si256( chainv[0], chainv[2] );
    chainv[1] = _mm256_xor_si256( chainv[1], chainv[3] );

-    MULT2( chainv[2], chainv[3]);
+    MULT2( chainv[2], chainv[3], MASK );
    chainv[2] = _mm256_xor_si256(chainv[2], chainv[4]);
    chainv[3] = _mm256_xor_si256(chainv[3], chainv[5]);

-    MULT2( chainv[4], chainv[5]);
+    MULT2( chainv[4], chainv[5], MASK );
    chainv[4] = _mm256_xor_si256(chainv[4], chainv[6]);
    chainv[5] = _mm256_xor_si256(chainv[5], chainv[7]);

-    MULT2( chainv[6], chainv[7]);
+    MULT2( chainv[6], chainv[7], MASK );
    chainv[6] = _mm256_xor_si256(chainv[6], chainv[8]);
    chainv[7] = _mm256_xor_si256(chainv[7], chainv[9]);

-    MULT2( chainv[8], chainv[9]);
+    MULT2( chainv[8], chainv[9], MASK );
    chainv[8] = _mm256_xor_si256( chainv[8], t0 );
    chainv[9] = _mm256_xor_si256( chainv[9], t1 );

    t0 = chainv[8];
    t1 = chainv[9];

-    MULT2( chainv[8], chainv[9]);
+    MULT2( chainv[8], chainv[9], MASK );
    chainv[8] = _mm256_xor_si256( chainv[8], chainv[6] );
    chainv[9] = _mm256_xor_si256( chainv[9], chainv[7] );

-    MULT2( chainv[6], chainv[7]);
+    MULT2( chainv[6], chainv[7], MASK );
    chainv[6] = _mm256_xor_si256( chainv[6], chainv[4] );
    chainv[7] = _mm256_xor_si256( chainv[7], chainv[5] );

-    MULT2( chainv[4], chainv[5]);
+    MULT2( chainv[4], chainv[5], MASK );
    chainv[4] = _mm256_xor_si256( chainv[4], chainv[2] );
    chainv[5] = _mm256_xor_si256( chainv[5], chainv[3] );

-    MULT2( chainv[2], chainv[3] );
+    MULT2( chainv[2], chainv[3], MASK );
    chainv[2] = _mm256_xor_si256( chainv[2], chainv[0] );
    chainv[3] = _mm256_xor_si256( chainv[3], chainv[1] );

-    MULT2( chainv[0], chainv[1] );
+    MULT2( chainv[0], chainv[1], MASK );
    chainv[0] = _mm256_xor_si256( _mm256_xor_si256( chainv[0], t0 ), msg0 );
    chainv[1] = _mm256_xor_si256( _mm256_xor_si256( chainv[1], t1 ), msg1 );

-    MULT2( msg0, msg1);
+    MULT2( msg0, msg1, MASK );
    chainv[2] = _mm256_xor_si256( chainv[2], msg0 );
    chainv[3] = _mm256_xor_si256( chainv[3], msg1 );

-    MULT2( msg0, msg1);
+    MULT2( msg0, msg1, MASK );
    chainv[4] = _mm256_xor_si256( chainv[4], msg0 );
    chainv[5] = _mm256_xor_si256( chainv[5], msg1 );

-    MULT2( msg0, msg1);
+    MULT2( msg0, msg1, MASK );
    chainv[6] = _mm256_xor_si256( chainv[6], msg0 );
    chainv[7] = _mm256_xor_si256( chainv[7], msg1 );

-    MULT2( msg0, msg1);
+    MULT2( msg0, msg1, MASK );
    chainv[8] = _mm256_xor_si256( chainv[8], msg0 );
    chainv[9] = _mm256_xor_si256( chainv[9], msg1 );

-    MULT2( msg0, msg1);
+    MULT2( msg0, msg1, MASK );

    chainv[3] = _mm256_or_si256( _mm256_slli_epi32( chainv[3],  1 ),
                                 _mm256_srli_epi32( chainv[3], 31 ) );
@@ -365,14 +345,14 @@ void rnd512_2way( luffa_2way_context *state, __m256i *msg )
                chainv[1],chainv[3],chainv[5],chainv[7],
                x[4], x[5], x[6], x[7] );

-    STEP_PART( &x[0], &CNS[ 0], &tmp[0] );
-    STEP_PART( &x[0], &CNS[ 2], &tmp[0] );
-    STEP_PART( &x[0], &CNS[ 4], &tmp[0] );
-    STEP_PART( &x[0], &CNS[ 6], &tmp[0] );
-    STEP_PART( &x[0], &CNS[ 8], &tmp[0] );
-    STEP_PART( &x[0], &CNS[10], &tmp[0] );
-    STEP_PART( &x[0], &CNS[12], &tmp[0] );
-    STEP_PART( &x[0], &CNS[14], &tmp[0] );
+    STEP_PART( &x[0], cns( 0), cns( 1), &tmp[0] );
+    STEP_PART( &x[0], cns( 2), cns( 3), &tmp[0] );
+    STEP_PART( &x[0], cns( 4), cns( 5), &tmp[0] );
+    STEP_PART( &x[0], cns( 6), cns( 7), &tmp[0] );
+    STEP_PART( &x[0], cns( 8), cns( 9), &tmp[0] );
+    STEP_PART( &x[0], cns(10), cns(11), &tmp[0] );
+    STEP_PART( &x[0], cns(12), cns(13), &tmp[0] );
+    STEP_PART( &x[0], cns(14), cns(15), &tmp[0] );

    MIXTON1024( x[0], x[1], x[2], x[3],
                chainv[0], chainv[2], chainv[4],chainv[6],
@@ -380,25 +360,24 @@ void rnd512_2way( luffa_2way_context *state, __m256i *msg )
                chainv[1],chainv[3],chainv[5],chainv[7]);

    /* Process last 256-bit block */
-    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[16], CNS[17],
+    STEP_PART2( chainv[8], chainv[9], t0, t1, cns(16), cns(17),
                tmp[0], tmp[1] );
-    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[18], CNS[19],
+    STEP_PART2( chainv[8], chainv[9], t0, t1, cns(18), cns(19),
                tmp[0], tmp[1] );
-    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[20], CNS[21],
+    STEP_PART2( chainv[8], chainv[9], t0, t1, cns(20), cns(21),
                tmp[0], tmp[1] );
-    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[22], CNS[23],
+    STEP_PART2( chainv[8], chainv[9], t0, t1, cns(22), cns(23),
                tmp[0], tmp[1] );
-    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[24], CNS[25],
+    STEP_PART2( chainv[8], chainv[9], t0, t1, cns(24), cns(25),
                tmp[0], tmp[1] );
-    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[26], CNS[27],
+    STEP_PART2( chainv[8], chainv[9], t0, t1, cns(26), cns(27),
                tmp[0], tmp[1] );
-    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[28], CNS[29],
+    STEP_PART2( chainv[8], chainv[9], t0, t1, cns(28), cns(29),
                tmp[0], tmp[1] );
-    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[30], CNS[31],
+    STEP_PART2( chainv[8], chainv[9], t0, t1, cns(30), cns(31),
                tmp[0], tmp[1] );
 }

-
 /***************************************************/
 /* Finalization function  */
 /* state: hash context    */
@@ -410,8 +389,9 @@ void finalization512_2way( luffa_2way_context *state, uint32 *b )
    __m256i* chainv = state->chainv;
    __m256i t[2];
    __m256i zero[2];
-    zero[0] = zero[1] = _mm256_setzero_si256();
-
+    zero[0] = zero[1] = m256_zero;
+    const __m256i shuff_bswap32 = m256_const2_64( 0x0c0d0e0f08090a0b,
+                                                  0x0405060700010203 );
    /*---- blank round with m=0 ----*/
    rnd512_2way( state, zero );

@@ -433,8 +413,10 @@ void finalization512_2way( luffa_2way_context *state, uint32 *b )
    _mm256_store_si256( (__m256i*)&hash[0], t[0] );
    _mm256_store_si256( (__m256i*)&hash[8], t[1] );

-    casti_m256i( b, 0 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );
-    casti_m256i( b, 1 ) = mm256_bswap_32( casti_m256i( hash, 1 ) );
+    casti_m256i( b, 0 ) = _mm256_shuffle_epi8(
+                                  casti_m256i( hash, 0 ), shuff_bswap32 );
+    casti_m256i( b, 1 ) = _mm256_shuffle_epi8( 
+                                  casti_m256i( hash, 1 ), shuff_bswap32 );

    rnd512_2way( state, zero );

@@ -455,26 +437,27 @@ void finalization512_2way( luffa_2way_context *state, uint32 *b )
    _mm256_store_si256( (__m256i*)&hash[0], t[0] );
    _mm256_store_si256( (__m256i*)&hash[8], t[1] );

-    casti_m256i( b, 2 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );
-    casti_m256i( b, 3 ) = mm256_bswap_32( casti_m256i( hash, 1 ) );
+    casti_m256i( b, 2 ) = _mm256_shuffle_epi8( 
+                                  casti_m256i( hash, 0 ), shuff_bswap32 );
+    casti_m256i( b, 3 ) = _mm256_shuffle_epi8( 
+                                  casti_m256i( hash, 1 ), shuff_bswap32 );
 }

 int luffa_2way_init( luffa_2way_context *state, int hashbitlen )
 {
-    int i;
    state->hashbitlen = hashbitlen;
-
-    for ( i=0; i<32; i++ ) CNS[i] =
-          _mm256_set_epi32( CNS_INIT[ (i<<2) + 3 ], CNS_INIT[ (i<<2) +2 ],
-                            CNS_INIT[ (i<<2) + 1 ], CNS_INIT[ (i<<2)    ],
-                            CNS_INIT[ (i<<2) + 3 ], CNS_INIT[ (i<<2) +2 ],
-                            CNS_INIT[ (i<<2) + 1 ], CNS_INIT[ (i<<2)    ] );
-
-    for ( i=0; i<10; i++ ) state->chainv[i] =
-          _mm256_set_epi32( IV[ (i<<2) +3 ], IV[ (i<<2) +2 ],
-                            IV[ (i<<2) +1 ], IV[ (i<<2)    ],
-                            IV[ (i<<2) +3 ], IV[ (i<<2) +2 ],
-                            IV[ (i<<2) +1 ], IV[ (i<<2)    ] );
+    __m128i *iv = (__m128i*)IV;
+    
+    state->chainv[0] = m256_const1_128( iv[0] );
+    state->chainv[1] = m256_const1_128( iv[1] );
+    state->chainv[2] = m256_const1_128( iv[2] );
+    state->chainv[3] = m256_const1_128( iv[3] );
+    state->chainv[4] = m256_const1_128( iv[4] );
+    state->chainv[5] = m256_const1_128( iv[5] );
+    state->chainv[6] = m256_const1_128( iv[6] );
+    state->chainv[7] = m256_const1_128( iv[7] );
+    state->chainv[8] = m256_const1_128( iv[8] );
+    state->chainv[9] = m256_const1_128( iv[9] );

    ((__m256i*)state->buffer)[0] = m256_zero;
    ((__m256i*)state->buffer)[1] = m256_zero;
@@ -492,13 +475,15 @@ int luffa_2way_update( luffa_2way_context *state, const void *data,
    __m256i msg[2];
    int i;
    int blocks = (int)len >> 5;
+    const __m256i shuff_bswap32 = m256_const2_64( 0x0c0d0e0f08090a0b,
+                                                  0x0405060700010203 );
    state-> rembytes = (int)len & 0x1F;

    // full blocks
    for ( i = 0; i < blocks; i++, vdata+=2 )
    {
-       msg[0] = mm256_bswap_32( vdata[ 0] );
-       msg[1] = mm256_bswap_32( vdata[ 1 ] );
+       msg[0] = _mm256_shuffle_epi8( vdata[ 0 ], shuff_bswap32 );
+       msg[1] = _mm256_shuffle_epi8( vdata[ 1 ], shuff_bswap32 );
       rnd512_2way( state, msg );
    }

@@ -507,9 +492,8 @@ int luffa_2way_update( luffa_2way_context *state, const void *data,
    if ( state->rembytes  )
    {
      // remaining data bytes
-      buffer[0] = mm256_bswap_32( vdata[0] );
-      buffer[1] = _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0,
-                                   0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
+      buffer[0] = _mm256_shuffle_epi8( vdata[0], shuff_bswap32 );
+      buffer[1] = m256_const2_64( 0, 0x0000000080000000 );
    }
    return 0;
 }
@@ -525,8 +509,7 @@ int luffa_2way_close( luffa_2way_context *state, void *hashval )
      rnd512_2way( state, buffer );
    else
    {     // empty pad block, constant data
-      msg[0] = _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0,
-                                0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
+      msg[0] = m256_const2_64( 0, 0x0000000080000000 );
      msg[1] = m256_zero;
      rnd512_2way( state, msg );
    }
@@ -545,13 +528,16 @@ int luffa_2way_update_close( luffa_2way_context *state,
    __m256i msg[2];
    int i;
    const int blocks = (int)( inlen >> 5 );
+    const __m256i shuff_bswap32 = m256_const2_64( 0x0c0d0e0f08090a0b,
+                                                  0x0405060700010203 );
+
    state->rembytes = inlen & 0x1F;

    // full blocks
    for ( i = 0; i < blocks; i++, vdata+=2 )
    {
-       msg[0] = mm256_bswap_32( vdata[ 0 ] );
-       msg[1] = mm256_bswap_32( vdata[ 1 ] );
+       msg[0] = _mm256_shuffle_epi8( vdata[ 0 ], shuff_bswap32 );
+       msg[1] = _mm256_shuffle_epi8( vdata[ 1 ], shuff_bswap32 );
       rnd512_2way( state, msg );
    }

@@ -559,16 +545,14 @@ int luffa_2way_update_close( luffa_2way_context *state,
    if ( state->rembytes  )
    {
       // padding of partial block
-       msg[0] = mm256_bswap_32( vdata[0] );
-       msg[1] = _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0,
-                                 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
+       msg[0] = _mm256_shuffle_epi8( vdata[ 0 ], shuff_bswap32 );
+       msg[1] = m256_const2_64( 0, 0x0000000080000000 );
       rnd512_2way( state, msg );
    }
    else
    {
       // empty pad block
-       msg[0] = _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0,
-                                 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
+       msg[0] = m256_const2_64( 0, 0x0000000080000000 );
       msg[1] = m256_zero;
       rnd512_2way( state, msg );
    }
--- a/algo/luffa/luffa_for_sse2.c
+++ b/algo/luffa/luffa_for_sse2.c
@@ -541,7 +541,9 @@ static void finalization512( hashState_luffa *state, uint32 *b )
    uint32   hash[8] __attribute((aligned(64)));
    __m256i* chainv = (__m256i*)state->chainv;
    __m256i  t;
-    const __m128i zero = _mm_setzero_si128();
+    const __m128i zero = m128_zero;
+    const __m256i shuff_bswap32 = m256_const2_64( 0x0c0d0e0f08090a0b,
+                                                  0x0405060700010203 );

    rnd512( state, zero, zero );

@@ -555,7 +557,9 @@ static void finalization512( hashState_luffa *state, uint32 *b )

    _mm256_store_si256( (__m256i*)hash, t );

-    casti_m256i( b, 0 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );
+    casti_m256i( b, 0 ) = _mm256_shuffle_epi8(
+                                 casti_m256i( hash, 0 ), shuff_bswap32 );
+//    casti_m256i( b, 0 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );

    rnd512( state, zero, zero );

@@ -568,7 +572,9 @@ static void finalization512( hashState_luffa *state, uint32 *b )

    _mm256_store_si256( (__m256i*)hash, t );

-    casti_m256i( b, 1 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );
+    casti_m256i( b, 1 ) = _mm256_shuffle_epi8( 
+                                 casti_m256i( hash, 0 ), shuff_bswap32 );
+//    casti_m256i( b, 1 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );
 }

 #else
--- a/algo/lyra2/lyra2rev3-4way.c
+++ b/algo/lyra2/lyra2rev3-4way.c
@@ -5,7 +5,6 @@
 #include "algo/bmw/bmw-hash-4way.h"
 #include "algo/cubehash/cubehash_sse2.h" 

-
 #if defined (LYRA2REV3_8WAY)

 typedef struct {
@@ -14,7 +13,7 @@ typedef struct {
   bmw256_8way_context       bmw;
 } lyra2v3_8way_ctx_holder;

-static lyra2v3_8way_ctx_holder l2v3_8way_ctx;
+static __thread lyra2v3_8way_ctx_holder l2v3_8way_ctx;

 bool init_lyra2rev3_8way_ctx()
 {
@@ -38,7 +37,7 @@ void lyra2rev3_8way_hash( void *state, const void *input )
   lyra2v3_8way_ctx_holder ctx __attribute__ ((aligned (64)));
   memcpy( &ctx, &l2v3_8way_ctx, sizeof(l2v3_8way_ctx) );

-   blake256_8way( &ctx.blake, input, 80 );
+   blake256_8way( &ctx.blake, input + (64*8), 16 );
   blake256_8way_close( &ctx.blake, vhash );

   dintrlv_8x32( hash0, hash1, hash2, hash3,
@@ -91,7 +90,7 @@ int scanhash_lyra2rev3_8way( struct work *work, const uint32_t max_nonce,
 {
   uint32_t hash[8*8] __attribute__ ((aligned (64)));
   uint32_t vdata[20*8] __attribute__ ((aligned (64)));
-   uint32_t *hash7 = &(hash[7<<3]);
+   uint32_t *hash7 = &hash[7<<3];
   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
   uint32_t *pdata = work->data;
   const uint32_t *ptarget = work->target;
@@ -99,12 +98,15 @@ int scanhash_lyra2rev3_8way( struct work *work, const uint32_t max_nonce,
   uint32_t n = first_nonce;
   const uint32_t Htarg = ptarget[7];
   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
-   const int thr_id = mythr->id;  // thr_id arg is deprecated
+   const int thr_id = mythr->id;

-   if ( opt_benchmark )
-      ( (uint32_t*)ptarget )[7] = 0x0000ff;
+   if ( opt_benchmark )  ( (uint32_t*)ptarget )[7] = 0x0000ff;

   mm256_bswap32_intrlv80_8x32( vdata, pdata );
+
+   blake256_8way_init( &l2v3_8way_ctx.blake );
+   blake256_8way( &l2v3_8way_ctx.blake, vdata, 64 );
+
   do
   {
      *noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
@@ -113,17 +115,18 @@ int scanhash_lyra2rev3_8way( struct work *work, const uint32_t max_nonce,
      lyra2rev3_8way_hash( hash, vdata );
      pdata[19] = n;

-      for ( int lane = 0; lane < 8; lane++ ) if ( hash7[lane] <= Htarg )
+      for ( int lane = 0; lane < 8; lane++ )
+      if ( unlikely( hash7[lane] <= Htarg ) )
      {
         extr_lane_8x32( lane_hash, hash, lane, 256 );
-         if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
+         if ( likely( fulltest( lane_hash, ptarget ) && !opt_benchmark ) )
         {
-              pdata[19] = n + lane;
-              submit_lane_solution( work, lane_hash, mythr, lane );
+             pdata[19] = n + lane;
+             submit_lane_solution( work, lane_hash, mythr, lane );
         }
      }
      n += 8;
-   } while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
+   } while ( likely( (n < max_nonce-8) && !work_restart[thr_id].restart ) );
   *hashes_done = n - first_nonce + 1;
   return 0;
 }
@@ -132,14 +135,14 @@ int scanhash_lyra2rev3_8way( struct work *work, const uint32_t max_nonce,

 #if defined (LYRA2REV3_4WAY)  

-
 typedef struct {
   blake256_4way_context     blake;
   cubehashParam             cube;
   bmw256_4way_context       bmw;
 } lyra2v3_4way_ctx_holder;

-static lyra2v3_4way_ctx_holder l2v3_4way_ctx;
+//static lyra2v3_4way_ctx_holder l2v3_4way_ctx;
+static __thread lyra2v3_4way_ctx_holder l2v3_4way_ctx;

 bool init_lyra2rev3_4way_ctx()
 {
@@ -159,7 +162,8 @@ void lyra2rev3_4way_hash( void *state, const void *input )
   lyra2v3_4way_ctx_holder ctx __attribute__ ((aligned (64))); 
   memcpy( &ctx, &l2v3_4way_ctx, sizeof(l2v3_4way_ctx) );

-   blake256_4way( &ctx.blake, input, 80 );
+//   blake256_4way( &ctx.blake, input, 80 );
+   blake256_4way( &ctx.blake, input + (64*4), 16 );
   blake256_4way_close( &ctx.blake, vhash );
   dintrlv_4x32( hash0, hash1, hash2, hash3, vhash, 256 );

@@ -205,6 +209,10 @@ int scanhash_lyra2rev3_4way( struct work *work, const uint32_t max_nonce,
      ( (uint32_t*)ptarget )[7] = 0x0000ff;

   mm128_bswap32_intrlv80_4x32( vdata, pdata );
+
+   blake256_4way_init( &l2v3_4way_ctx.blake );
+   blake256_4way( &l2v3_4way_ctx.blake, vdata, 64 );
+
   do
   {
      *noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
--- a/algo/sha/sha256-hash-4way.c
+++ b/algo/sha/sha256-hash-4way.c
@@ -305,9 +305,11 @@ void sha256_4way_close( sha256_4way_context *sc, void *dst )
    low = low << 3;

    sc->buf[ pad >> 2 ] =
-                 mm128_bswap_32( _mm_set1_epi32( high ) );
+                 mm128_bswap_32( m128_const1_32( high ) );
+//                 mm128_bswap_32( _mm_set1_epi32( high ) );
    sc->buf[ ( pad+4 ) >> 2 ] =
-                 mm128_bswap_32( _mm_set1_epi32( low ) );
+                 mm128_bswap_32( m128_const1_32( low ) );
+//                 mm128_bswap_32( _mm_set1_epi32( low ) );
    sha256_4way_round( sc, sc->buf, sc->val );

    mm128_block_bswap_32( dst, sc->val );
@@ -538,9 +540,9 @@ void sha256_8way_close( sha256_8way_context *sc, void *dst )
    low = low << 3;

    sc->buf[ pad >> 2 ] =
-                 mm256_bswap_32( _mm256_set1_epi32( high ) );
+                 mm256_bswap_32( m256_const1_32( high ) );
    sc->buf[ ( pad+4 ) >> 2 ] =
-                 mm256_bswap_32( _mm256_set1_epi32( low ) );
+                 mm256_bswap_32( m256_const1_32( low ) );

    sha256_8way_round( sc, sc->buf, sc->val );

--- a/algo/sha/sha256_hash_11way.c
+++ b/algo/sha/sha256_hash_11way.c
@@ -1,538 +0,0 @@
-#if 0
-
-#include <stddef.h>
-#include <string.h>
-
-#include "sha2-hash-4way.h"
-
-#if defined(__AVX2__)
-
-// naming convention for variables and macros
-// VARx: AVX2 8 way 32 bit
-// VARy: MMX 2 way 32 bit
-// VARz: scalar integer 32 bit
-
-
-static const uint32_t H256[8] =
-{
-        0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
-        0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
-};
-
-static const uint32_t K256[64] = 
-{
-        0x428A2F98, 0x71374491, 0xB5C0FBCF, 0xE9B5DBA5,
-        0x3956C25B, 0x59F111F1, 0x923F82A4, 0xAB1C5ED5,
-        0xD807AA98, 0x12835B01, 0x243185BE, 0x550C7DC3,
-        0x72BE5D74, 0x80DEB1FE, 0x9BDC06A7, 0xC19BF174,
-        0xE49B69C1, 0xEFBE4786, 0x0FC19DC6, 0x240CA1CC,
-        0x2DE92C6F, 0x4A7484AA, 0x5CB0A9DC, 0x76F988DA,
-        0x983E5152, 0xA831C66D, 0xB00327C8, 0xBF597FC7,
-        0xC6E00BF3, 0xD5A79147, 0x06CA6351, 0x14292967,
-        0x27B70A85, 0x2E1B2138, 0x4D2C6DFC, 0x53380D13,
-        0x650A7354, 0x766A0ABB, 0x81C2C92E, 0x92722C85,
-        0xA2BFE8A1, 0xA81A664B, 0xC24B8B70, 0xC76C51A3,
-        0xD192E819, 0xD6990624, 0xF40E3585, 0x106AA070,
-        0x19A4C116, 0x1E376C08, 0x2748774C, 0x34B0BCB5,
-        0x391C0CB3, 0x4ED8AA4A, 0x5B9CCA4F, 0x682E6FF3,
-        0x748F82EE, 0x78A5636F, 0x84C87814, 0x8CC70208,
-        0x90BEFFFA, 0xA4506CEB, 0xBEF9A3F7, 0xC67178F2
-};
-
-#define CHx(X, Y, Z) \
-   _mm256_xor_si256( _mm256_and_si256( _mm256_xor_si256( Y, Z ), X ), Z ) 
-
-#define CHy(X, Y, Z) \
-   _mm_xor_si64( _mm_and_si64( _mm_xor_si64( Y, Z ), X ), Z )
-
-#define CHz(X, Y, Z) ((( (Y) ^ (Z) ) & (X) ) ^ (Z) )
-
-
-#define MAJx(X, Y, Z) \
-   _mm256_or_si256( _mm256_and_si256( X, Y ), \
-                    _mm256_and_si256( _mm256_or_si256( X, Y ), Z ) )
-
-#define MAJy(X, Y, Z) \
-   _mm_or_si64( _mm_and_si64( X, Y ), \
-                    _mm_and_si64( _mm_or_si64( X, Y ), Z ) )
-
-#define MAJz(X, Y, Z)  ( ( (X) & (Y) ) | ( ( (X) | (Y) ) & (Z) ) )
-
-#define BSG2_0x(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,2), mm256_ror_32(x,13) ), _mm256_srli_epi32(x,22) )
-
-#define BSG2_0y(x) \
-   _mm_xor_si64( _mm_xor_si64( \
-       mm64_ror_32(x,2), mm64_ror_32(x,13) ), _mm_srli_pi32(x,22) )
-
-#define BSG2_0z(x)  ( u32_ror_32(x,2) ^ u32_ror_32(x,13)  ^ ((x)>>22) )
-
-#define BSG2_1x(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,6), mm256_ror_32(x,11) ), _mm256_srli_epi32(x,25) )
-
-#define BSG2_1y(x) \
-   _mm_xor_si64( _mm_xor_si64( \
-       mm64_ror_32(x,6), mm64_ror_32(x,11) ), _mm_srli_pi32(x,25) )
-
-#define BSG2_1z(x)   ( u32_ror_32(x,6) ^ u32_ror_32(x,11) ^ ((x)>>25) )
-
-#define SSG2_0x(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,7), mm256_ror_32(x,18) ), _mm256_srli_epi32(x,3) ) 
-
-#define SSG2_0y(x) \
-   _mm_xor_si64( _mm_xor_si64( \
-       mm64_ror_32(x,7), mm64_ror_32(x,18) ), _mm_srli_pi32(x,3) )
-
-#define SSG2_0z(x)  (( u32_ror_32(x,7) ^ u32_ror_32(x,18) ) ^ ((x)>>3) )
-
-#define SSG2_1x(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,17), mm256_ror_32(x,19) ), _mm256_srli_epi32(x,10) )
-
-#define SSG2_1y(x) \
-   _mm_xor_si64( _mm_xor_si64( \
-       mm64_ror_32(x,17), mm64_ror_32(x,19) ), _mm_srli_pi32(x,10) )
-
-#define SSG2_1z(x)   ( u32_ror_32(x,17) ^ u32_ror_32(x,19)  ^ ((x)>>10) )
-
-#define SHA2x_MEXP( a, b, c, d ) \
-     _mm256_add_epi32( _mm256_add_epi32( _mm256_add_epi32( \
-                 SSG2_1x( Wx[a] ), Wx[b] ), SSG2_0x( Wx[c] ) ), Wx[d] )
-
-#define SHA2y_MEXP( a, b, c, d ) \
-     _mm_add_pi32( _mm_add_pi32( _mm_add_pi32( \
-                 SSG2_1y( Wy[a] ), Wy[b] ), SSG2_0y( Wy[c] ) ), Wy[d] )
-
-#define SHA2z_MEXP( a, b, c, d ) \
-               ( SSG2_1z( Wz[a] ) + Wz[b] + SSG2_0z( Wz[c] ) + Wz[d] )
-
-
-#define SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx, \
-	                  Ay, By, Cy, Dy, Ey, Fy, Gy, Hy, \
-		          Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz, i, j) \
-do { \
-  __m256i T1x, T2x; \
-  __m64 T1y, T2y; \
-  uint32_t T1z, T2z; \
-  T1x = _mm256_add_epi32( _mm256_add_epi32( _mm256_add_epi32( \
-        _mm256_add_epi32( Hx, BSG2_1x(Ex) ), CHx(Ex, Fx, Gx) ), \
-                          _mm256_set1_epi32( K256[( (j)+(i) )] ) ), Wx[i] ); \
-  T1y = _mm_add_pi32( _mm_add_pi32( _mm_add_pi32( \
-        _mm_add_pi32( Hy, BSG2_1y(Ey) ), CHy(Ey, Fy, Gy) ), \
-                          _mm_set1_pi32( K256[( (j)+(i) )] ) ), Wy[i] ); \
-  T1z = Hz + BSG2_1z( Ez ) + CHz( Ez, Fz, Gz ) + K256[ ((j)+(i)) ] + Wz[i]; \
-  T2x = _mm256_add_epi32( BSG2_0x(Ax), MAJx(Ax, Bx, Cx) ); \
-  T2y = _mm_add_pi32( BSG2_0y(Ay), MAJy(Ay, By, Cy) ); \
-  T2z = BSG2_0z( Az ) + MAJz( Az, Bz, Cz ); \
-  Dx  = _mm256_add_epi32( Dx,  T1x ); \
-  Dy  = _mm_add_pi32( Dy, T1y ); \
-  Dz  = Dz + T1z; \
-  Hx  = _mm256_add_epi32( T1x, T2x ); \
-  Hy  = _mm_add_pi32( T1y, T2y ); \
-  Hz  = T1z + T2z; \
-} while (0)
-	
-void sha256_11way_round( __m256i *inx, __m256i rx[8], __m64 *iny, __m64 ry[8],
-                         uint32_t *inz, uint32_t rz[8] )
-{
-   __m256i Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx;
-   __m256i Wx[16];
-   __m64 Ay, By, Cy, Dy, Ey, Fy, Gy, Hy;
-   __m64 Wy[16];
-   uint32_t Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz;
-   uint32_t Wz[16];
-
-   Wx[ 0] = mm256_bswap_32( inx[ 0] );
-   Wy[ 0] =  mm64_bswap_32( iny[ 0] );
-   Wz[ 0] =       bswap_32( inz[ 0] );
-
-   Wx[ 1] = mm256_bswap_32( inx[ 1] );
-   Wy[ 1] =  mm64_bswap_32( iny[ 1] );
-   Wz[ 1] =       bswap_32( inz[ 1] );
-
-   Wx[ 2] = mm256_bswap_32( inx[ 2] );
-   Wy[ 2] =  mm64_bswap_32( iny[ 2] );
-   Wz[ 2] =       bswap_32( inz[ 2] );
-
-   Wx[ 3] = mm256_bswap_32( inx[ 3] );
-   Wy[ 3] =  mm64_bswap_32( iny[ 3] );
-   Wz[ 3] =       bswap_32( inz[ 3] );
-
-   Wx[ 4] = mm256_bswap_32( inx[ 4] );
-   Wy[ 4] =  mm64_bswap_32( iny[ 4] );
-   Wz[ 4] =       bswap_32( inz[ 4] );
-
-   Wx[ 5] = mm256_bswap_32( inx[ 5] );
-   Wy[ 5] =  mm64_bswap_32( iny[ 5] );
-   Wz[ 5] =       bswap_32( inz[ 5] );
-
-   Wx[ 6] = mm256_bswap_32( inx[ 6] );
-   Wy[ 6] =  mm64_bswap_32( iny[ 6] );
-   Wz[ 6] =       bswap_32( inz[ 6] );
-
-   Wx[ 7] = mm256_bswap_32( inx[ 7] );
-   Wy[ 7] =  mm64_bswap_32( iny[ 7] );
-   Wz[ 7] =       bswap_32( inz[ 7] );
-
-   Wx[ 8] = mm256_bswap_32( inx[ 8] );
-   Wy[ 8] =  mm64_bswap_32( iny[ 8] );
-   Wz[ 8] =       bswap_32( inz[ 8] );
-
-   Wx[ 9] = mm256_bswap_32( inx[ 9] );
-   Wy[ 9] =  mm64_bswap_32( iny[ 9] );
-   Wz[ 9] =       bswap_32( inz[ 9] );
-
-   Wx[10] = mm256_bswap_32( inx[10] );
-   Wy[10] =  mm64_bswap_32( iny[10] );
-   Wz[10] =       bswap_32( inz[10] );
-
-   Wx[11] = mm256_bswap_32( inx[11] );
-   Wy[11] =  mm64_bswap_32( iny[11] );
-   Wz[11] =       bswap_32( inz[11] );
-
-   Wx[12] = mm256_bswap_32( inx[12] );
-   Wy[12] =  mm64_bswap_32( iny[12] );
-   Wz[12] =       bswap_32( inz[12] );
-
-   Wx[13] = mm256_bswap_32( inx[13] );
-   Wy[13] =  mm64_bswap_32( iny[13] );
-   Wz[13] =       bswap_32( inz[13] );
-
-   Wx[14] = mm256_bswap_32( inx[14] );
-   Wy[14] =  mm64_bswap_32( iny[14] );
-   Wz[14] =       bswap_32( inz[14] );
-
-   Wx[15] = mm256_bswap_32( inx[15] );
-   Wy[15] =  mm64_bswap_32( iny[15] );
-   Wz[15] =       bswap_32( inz[15] );
-
-   Ax = rx[0];     Ay = ry[0];     Az = rz[0];
-   Bx = rx[1];     By = ry[1];     Bz = rz[1];
-   Cx = rx[2];     Cy = ry[2];     Cz = rz[2];
-   Dx = rx[3];     Dy = ry[3];     Dz = rz[3];
-   Ex = rx[4];     Ey = ry[4];     Ez = rz[4];
-   Fx = rx[5];     Fy = ry[5];     Fz = rz[5];
-   Gx = rx[6];     Gy = ry[6];     Gz = rz[6];
-   Hx = rx[7];     Hy = ry[7];     Hz = rz[7];
-
-   SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx,
-                     Ay, By, Cy, Dy, Ey, Fy, Gy, Hy,
-                     Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz,  0, 0 );
-   SHA2s_11WAY_STEP( Hx, Ax, Bx, Cx, Dx, Ex, Fx, Gx,
-		     Hy, Ay, By, Cy, Dy, Ey, Fy, Gy,
-		     Hz, Az, Bz, Cz, Dz, Ez, Fz, Gz,  1, 0 );
-   SHA2s_11WAY_STEP( Gx, Hx, Ax, Bx, Cx, Dx, Ex, Fx,
-		     Gy, Hy, Ay, By, Cy, Dy, Ey, Fy,
-		     Gz, Hz, Az, Bz, Cz, Dz, Ez, Fz,  2, 0 );
-   SHA2s_11WAY_STEP( Fx, Gx, Hx, Ax, Bx, Cx, Dx, Ex,
-		     Fy, Gy, Hy, Ay, By, Cy, Dy, Ey,
-		     Fz, Gz, Hz, Az, Bz, Cz, Dz, Ez,  3, 0 );
-   SHA2s_11WAY_STEP( Ex, Fx, Gx, Hx, Ax, Bx, Cx, Dx,
-		     Ey, Fy, Gy, Hy, Ay, By, Cy, Dy,
-		     Ez, Fz, Gz, Hz, Az, Bz, Cz, Dz,  4, 0 );
-   SHA2s_11WAY_STEP( Dx, Ex, Fx, Gx, Hx, Ax, Bx, Cx,
-		     Dy, Ey, Fy, Gy, Hy, Ay, By, Cy,
-		     Dz, Ez, Fz, Gz, Hz, Az, Bz, Cz,  5, 0 );
-   SHA2s_11WAY_STEP( Cx, Dx, Ex, Fx, Gx, Hx, Ax, Bx,
-		     Cy, Dy, Ey, Fy, Gy, Hy, Ay, By,
-		     Cz, Dz, Ez, Fz, Gz, Hz, Az, Bz,  6, 0 );
-   SHA2s_11WAY_STEP( Bx, Cx, Dx, Ex, Fx, Gx, Hx, Ax,
-		     By, Cy, Dy, Ey, Fy, Gy, Hy, Ay,
-		     Bz, Cz, Dz, Ez, Fz, Gz, Hz, Az,  7, 0 );
-   SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx,
-		     Ay, By, Cy, Dy, Ey, Fy, Gy, Hy,
-		     Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz,  8, 0 );
-   SHA2s_11WAY_STEP( Hx, Ax, Bx, Cx, Dx, Ex, Fx, Gx,
-		     Hy, Ay, By, Cy, Dy, Ey, Fy, Gy,
-		     Hz, Az, Bz, Cz, Dz, Ez, Fz, Gz,  9, 0 );
-   SHA2s_11WAY_STEP( Gx, Hx, Ax, Bx, Cx, Dx, Ex, Fx,
-		     Gy, Hy, Ay, By, Cy, Dy, Ey, Fy,
-		     Gz, Hz, Az, Bz, Cz, Dz, Ez, Fz, 10, 0 );
-   SHA2s_11WAY_STEP( Fx, Gx, Hx, Ax, Bx, Cx, Dx, Ex,
-		     Fy, Gy, Hy, Ay, By, Cy, Dy, Ey,
-		     Fz, Gz, Hz, Az, Bz, Cz, Dz, Ez, 11, 0 );
-   SHA2s_11WAY_STEP( Ex, Fx, Gx, Hx, Ax, Bx, Cx, Dx,
-		     Ey, Fy, Gy, Hy, Ay, By, Cy, Dy,
-		     Ez, Fz, Gz, Hz, Az, Bz, Cz, Dz, 12, 0 );
-   SHA2s_11WAY_STEP( Dx, Ex, Fx, Gx, Hx, Ax, Bx, Cx,
-		     Dy, Ey, Fy, Gy, Hy, Ay, By, Cy,
-		     Dz, Ez, Fz, Gz, Hz, Az, Bz, Cz, 13, 0 );
-   SHA2s_11WAY_STEP( Cx, Dx, Ex, Fx, Gx, Hx, Ax, Bx,
-		     Cy, Dy, Ey, Fy, Gy, Hy, Ay, By,
-		     Cz, Dz, Ez, Fz, Gz, Hz, Az, Bz, 14, 0 );
-   SHA2s_11WAY_STEP( Bx, Cx, Dx, Ex, Fx, Gx, Hx, Ax,
-		     By, Cy, Dy, Ey, Fy, Gy, Hy, Ay,
-		     Bz, Cz, Dz, Ez, Fz, Gz, Hz, Az, 15, 0 );
-
-   for ( int j = 16; j < 64; j += 16 )
-   {
-      Wx[ 0] = SHA2x_MEXP( 14,  9,  1,  0 );
-      Wy[ 0] = SHA2y_MEXP( 14,  9,  1,  0 );
-      Wz[ 0] = SHA2z_MEXP( 14,  9,  1,  0 );
-
-      Wx[ 1] = SHA2x_MEXP( 15, 10,  2,  1 );
-      Wy[ 1] = SHA2y_MEXP( 15, 10,  2,  1 );
-      Wz[ 1] = SHA2z_MEXP( 15, 10,  2,  1 );
-
-      Wx[ 2] = SHA2x_MEXP(  0, 11,  3,  2 );
-      Wy[ 2] = SHA2y_MEXP(  0, 11,  3,  2 );
-      Wz[ 2] = SHA2z_MEXP(  0, 11,  3,  2 );
-
-      Wx[ 3] = SHA2x_MEXP(  1, 12,  4,  3 );
-      Wy[ 3] = SHA2y_MEXP(  1, 12,  4,  3 );
-      Wz[ 3] = SHA2z_MEXP(  1, 12,  4,  3 );
-
-      Wx[ 4] = SHA2x_MEXP(  2, 13,  5,  4 );
-      Wy[ 4] = SHA2y_MEXP(  2, 13,  5,  4 );
-      Wz[ 4] = SHA2z_MEXP(  2, 13,  5,  4 );
-
-      Wx[ 5] = SHA2x_MEXP(  3, 14,  6,  5 );
-      Wy[ 5] = SHA2y_MEXP(  3, 14,  6,  5 );
-      Wz[ 5] = SHA2z_MEXP(  3, 14,  6,  5 );
-
-      Wx[ 6] = SHA2x_MEXP(  4, 15,  7,  6 );
-      Wy[ 6] = SHA2y_MEXP(  4, 15,  7,  6 );
-      Wz[ 6] = SHA2z_MEXP(  4, 15,  7,  6 );
-
-      Wx[ 7] = SHA2x_MEXP(  5,  0,  8,  7);
-      Wy[ 7] = SHA2y_MEXP(  5,  0,  8,  7);
-      Wz[ 7] = SHA2z_MEXP(  5,  0,  8,  7);
-
-      Wx[ 8] = SHA2x_MEXP(  6,  1,  9,  8);
-      Wy[ 8] = SHA2y_MEXP(  6,  1,  9,  8);
-      Wz[ 8] = SHA2z_MEXP(  6,  1,  9,  8);
-
-      Wx[ 9] = SHA2x_MEXP(  7,  2, 10,  9 );
-      Wy[ 9] = SHA2y_MEXP(  7,  2, 10,  9);
-      Wz[ 9] = SHA2z_MEXP(  7,  2, 10,  9);
-
-      Wx[10] = SHA2x_MEXP(  8,  3, 11, 10 );
-      Wy[10] = SHA2y_MEXP(  8,  3, 11, 10);
-      Wz[10] = SHA2z_MEXP(  8,  3, 11, 10);
-
-      Wx[11] = SHA2x_MEXP(  9,  4, 12, 11);
-      Wy[11] = SHA2y_MEXP(  9,  4, 12, 11);
-      Wz[11] = SHA2z_MEXP(  9,  4, 12, 11 );
-
-      Wx[12] = SHA2x_MEXP( 10,  5, 13, 12 );
-      Wy[12] = SHA2y_MEXP( 10,  5, 13, 12 );
-      Wz[12] = SHA2z_MEXP( 10,  5, 13, 12 );
-
-      Wx[13] = SHA2x_MEXP( 11,  6, 14, 13 );
-      Wy[13] = SHA2y_MEXP( 11,  6, 14, 13 );
-      Wz[13] = SHA2z_MEXP( 11,  6, 14, 13 );
-
-      Wx[14] = SHA2x_MEXP( 12,  7, 15, 14 );
-      Wy[14] = SHA2y_MEXP( 12,  7, 15, 14 );
-      Wz[14] = SHA2z_MEXP( 12,  7, 15, 14 );
-
-      Wx[15] = SHA2x_MEXP( 13,  8,  0, 15 );
-      Wy[15] = SHA2y_MEXP( 13,  8,  0, 15 );
-      Wz[15] = SHA2z_MEXP( 13,  8,  0, 15 );
-
-
-      SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx,
-                        Ay, By, Cy, Dy, Ey, Fy, Gy, Hy,
-			Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz,	 0, j );
-      SHA2s_11WAY_STEP( Hx, Ax, Bx, Cx, Dx, Ex, Fx, Gx,
-		        Hy, Ay, By, Cy, Dy, Ey, Fy, Gy,
-		       	Hz, Az, Bz, Cz, Dz, Ez, Fz, Gz,  1, j );
-      SHA2s_11WAY_STEP( Gx, Hx, Ax, Bx, Cx, Dx, Ex, Fx,
-		        Gy, Hy, Ay, By, Cy, Dy, Ey, Fy,
-		       	Gz, Hz, Az, Bz, Cz, Dz, Ez, Fz,  2, j );
-      SHA2s_11WAY_STEP( Fx, Gx, Hx, Ax, Bx, Cx, Dx, Ex,
-		        Fy, Gy, Hy, Ay, By, Cy, Dy, Ey,
-		       	Fz, Gz, Hz, Az, Bz, Cz, Dz, Ez,  3, j );
-      SHA2s_11WAY_STEP( Ex, Fx, Gx, Hx, Ax, Bx, Cx, Dx,
-		        Ey, Fy, Gy, Hy, Ay, By, Cy, Dy,
-		       	Ez, Fz, Gz, Hz, Az, Bz, Cz, Dz,  4, j );
-      SHA2s_11WAY_STEP( Dx, Ex, Fx, Gx, Hx, Ax, Bx, Cx,
-		        Dy, Ey, Fy, Gy, Hy, Ay, By, Cy,
-		       	Dz, Ez, Fz, Gz, Hz, Az, Bz, Cz,  5, j );
-      SHA2s_11WAY_STEP( Cx, Dx, Ex, Fx, Gx, Hx, Ax, Bx,
-		        Cy, Dy, Ey, Fy, Gy, Hy, Ay, By,
-		       	Cz, Dz, Ez, Fz, Gz, Hz, Az, Bz,  6, j );
-      SHA2s_11WAY_STEP( Bx, Cx, Dx, Ex, Fx, Gx, Hx, Ax,
-		        By, Cy, Dy, Ey, Fy, Gy, Hy, Ay,
-		       	Bz, Cz, Dz, Ez, Fz, Gz, Hz, Az,  7, j );
-      SHA2s_11WAY_STEP( Ax, Bx, Cx, Dx, Ex, Fx, Gx, Hx,
-                        Ay, By, Cy, Dy, Ey, Fy, Gy, Hy,
-                        Az, Bz, Cz, Dz, Ez, Fz, Gz, Hz,  8, j );
-      SHA2s_11WAY_STEP( Hx, Ax, Bx, Cx, Dx, Ex, Fx, Gx, 
-                        Hy, Ay, By, Cy, Dy, Ey, Fy, Gy, 
-                        Hz, Az, Bz, Cz, Dz, Ez, Fz, Gz,  9, j );
-      SHA2s_11WAY_STEP( Gx, Hx, Ax, Bx, Cx, Dx, Ex, Fx, 
-                        Gy, Hy, Ay, By, Cy, Dy, Ey, Fy, 
-                        Gz, Hz, Az, Bz, Cz, Dz, Ez, Fz, 10, j );
-      SHA2s_11WAY_STEP( Fx, Gx, Hx, Ax, Bx, Cx, Dx, Ex, 
-                        Fy, Gy, Hy, Ay, By, Cy, Dy, Ey, 
-                        Fz, Gz, Hz, Az, Bz, Cz, Dz, Ez, 11, j );
-      SHA2s_11WAY_STEP( Ex, Fx, Gx, Hx, Ax, Bx, Cx, Dx, 
-                        Ey, Fy, Gy, Hy, Ay, By, Cy, Dy, 
-                        Ez, Fz, Gz, Hz, Az, Bz, Cz, Dz, 12, j );
-      SHA2s_11WAY_STEP( Dx, Ex, Fx, Gx, Hx, Ax, Bx, Cx, 
-                        Dy, Ey, Fy, Gy, Hy, Ay, By, Cy, 
-                        Dz, Ez, Fz, Gz, Hz, Az, Bz, Cz, 13, j );
-      SHA2s_11WAY_STEP( Cx, Dx, Ex, Fx, Gx, Hx, Ax, Bx, 
-                        Cy, Dy, Ey, Fy, Gy, Hy, Ay, By, 
-                        Cz, Dz, Ez, Fz, Gz, Hz, Az, Bz, 14, j );
-      SHA2s_11WAY_STEP( Bx, Cx, Dx, Ex, Fx, Gx, Hx, Ax, 
-                        By, Cy, Dy, Ey, Fy, Gy, Hy, Ay, 
-                        Bz, Cz, Dz, Ez, Fz, Gz, Hz, Az, 15, j );
-   }
-
-   rx[0] = _mm256_add_epi32( rx[0], Ax );
-   ry[0] =     _mm_add_pi32( ry[0], Ay );
-   rz[0] =                   rz[0]+ Az;
-   rx[1] = _mm256_add_epi32( rx[1], Bx );
-   ry[1] =     _mm_add_pi32( ry[1], By );
-   rz[1] =                   rz[1]+ Bz;
-   rx[2] = _mm256_add_epi32( rx[2], Cx );
-   ry[2] =     _mm_add_pi32( ry[2], Cy );
-   rz[3] =                   rz[3]+ Dz;
-   rx[4] = _mm256_add_epi32( rx[4], Ex );
-   ry[4] =     _mm_add_pi32( ry[4], Ey );
-   rz[4] =                   rz[4]+ Ez;
-   rx[5] = _mm256_add_epi32( rx[5], Fx );
-   ry[5] =     _mm_add_pi32( ry[5], Fy );
-   rz[5] =                   rz[5]+ Fz;
-   rx[6] = _mm256_add_epi32( rx[6], Gx );
-   ry[6] =     _mm_add_pi32( ry[6], Gy );
-   rz[6] =                   rz[6]+ Gz;
-   rx[7] = _mm256_add_epi32( rx[7], Hx );
-   ry[7] =     _mm_add_pi32( ry[7], Hy );
-   rz[7] =                   rz[7]+ Hz;
-
-}
-
-void sha256_11way_init( sha256_11way_context *ctx )
-{
-   ctx->count_high = ctx->count_low = 0;
-   ctx->valx[0] = _mm256_set1_epi32( H256[0] );
-   ctx->valy[0] =     _mm_set1_pi32( H256[0] );
-   ctx->valx[1] = _mm256_set1_epi32( H256[0] );
-   ctx->valy[1] =     _mm_set1_pi32( H256[0] );
-   ctx->valx[2] = _mm256_set1_epi32( H256[0] );
-   ctx->valy[2] =     _mm_set1_pi32( H256[0] );
-   ctx->valx[3] = _mm256_set1_epi32( H256[0] );
-   ctx->valy[3] =     _mm_set1_pi32( H256[0] );
-   ctx->valx[4] = _mm256_set1_epi32( H256[0] );
-   ctx->valy[4] =     _mm_set1_pi32( H256[0] );
-   ctx->valx[5] = _mm256_set1_epi32( H256[0] );
-   ctx->valy[5] =     _mm_set1_pi32( H256[0] );
-   ctx->valx[6] = _mm256_set1_epi32( H256[0] );
-   ctx->valy[6] =     _mm_set1_pi32( H256[0] );
-   ctx->valx[7] = _mm256_set1_epi32( H256[0] );
-   ctx->valy[7] =     _mm_set1_pi32( H256[0] );
-   memcpy( ctx->valz, H256, 32 );
-}
-
-
-void sha256_11way_update( sha256_11way_context *ctx, const void *datax,
-	                  const void *datay, const void *dataz, size_t len )
-{
-   __m256i  *vdatax = (__m256i*) datax;
-    __m64   *vdatay = (__m64*)   datay;
-   uint32_t *idataz = (uint32_t*)dataz;
-   size_t ptr;
-   const int buf_size = 64;
-
-   ptr = (unsigned)ctx->count_low & (buf_size - 1U);
-   while ( len > 0 )
-   {
-      size_t clen;
-      uint32_t clow, clow2;
-
-      clen = buf_size - ptr;
-      if ( clen > len )
-         clen = len;
-      memcpy_256( ctx->bufx + (ptr>>2), vdatax + (ptr>>2), clen>>2 );
-      memcpy_m64( ctx->bufy + (ptr>>2), vdatay + (ptr>>2), clen>>2 );
-      memcpy    ( ctx->bufz +  ptr,     idataz +  ptr,     clen    );
-      ptr += clen;
-      len -= clen;
-      if ( ptr == buf_size )
-      {
-         sha256_11way_round( ctx->bufx, ctx->valx,
-			     ctx->bufy, ctx->valy,
-			     ctx->bufz, ctx->valz );
-         ptr = 0;
-      }
-      clow = ctx->count_low;
-      clow2 = clow + clen;
-      ctx->count_low = clow2;
-      if ( clow2 < clow )
-         ctx->count_high++;
-   }
-}
-
-
-void sha256_11way_close( sha256_11way_context *ctx, void *dstx, void *dsty,
-	                                            void *dstz)
-{
-    unsigned ptr, u;
-    uint32_t low, high;
-    const int buf_size = 64;
-    const int pad = buf_size - 8;
-
-    ptr = (unsigned)ctx->count_low & (buf_size - 1U);
-    ctx->bufx[ ptr>>2 ] = _mm256_set1_epi32( 0x80 );
-    ctx->bufy[ ptr>>2 ] = _mm_set1_pi32( 0x80 );
-    ctx->bufz[ ptr>>2 ] = 0x80;
-    ptr += 4;
-
-    if ( ptr > pad )
-    {
-         memset_zero_256( ctx->bufx + (ptr>>2), (buf_size - ptr) >> 2 );
-         memset_zero_m64( ctx->bufy + (ptr>>2), (buf_size - ptr) >> 2 );
-         memset(      ctx->bufz + (ptr>>2), 0,  (buf_size - ptr) >> 2 );
-         sha256_11way_round( ctx->bufx, ctx->valx,
-			     ctx->bufy, ctx->valy,
-			     ctx->bufz, ctx->valz );
-         memset_zero_256( ctx->bufx, pad >> 2 );
-         memset_zero_m64(  ctx->bufy, pad >> 2 );
-         memset(      ctx->bufz, 0,  pad >> 2 );
-    }
-    else
-    {
-        memset_zero_256( ctx->bufx + (ptr>>2),    (pad - ptr) >> 2 );
-        memset_zero_m64(  ctx->bufy + (ptr>>2),    (pad - ptr) >> 2 );
-        memset(          ctx->bufz + (ptr>>2), 0, (pad - ptr) >> 2 );
-    }
-
-    low = ctx->count_low;
-    high = (ctx->count_high << 3) | (low >> 29);
-    low = low << 3;
-
-    ctx->bufx[ pad >> 2 ] =
-                 mm256_bswap_32( _mm256_set1_epi32( high ) );
-    ctx->bufy[ pad >> 2 ] =
-                 mm64_bswap_32( _mm_set1_pi32( high ) );
-    ctx->bufz[ pad >> 2 ] =
-                 bswap_32( high );
-
-
-    ctx->bufx[ ( pad+4 ) >> 2 ] =
-                 mm256_bswap_32( _mm256_set1_epi32( low ) );
-    ctx->bufy[ ( pad+4 ) >> 2 ] =
-                 mm64_bswap_32( _mm_set1_pi32( low ) );
-    ctx->bufz[ ( pad+4 ) >> 2 ] =
-                 bswap_32( low );
-
-    sha256_11way_round( ctx->bufx, ctx->valx,
-		       ctx->bufy, ctx->valy,
-		       ctx->bufz, ctx->valz  );
-
-    for ( u = 0; u < 8; u ++ )
-    {
-       casti_m256i( dstx, u ) = mm256_bswap_32( ctx->valx[u] );
-       casti_m64  ( dsty, u ) =  mm64_bswap_32( ctx->valy[u] );
-       ((uint32_t*)dstz)[u] = bswap_32( ctx->valz[u] );
-   }
-}
-
-#endif
-#endif   // 0
--- a/algo/sha/sha256t-4way.c
+++ b/algo/sha/sha256t-4way.c
@@ -5,137 +5,6 @@
 #include <stdio.h>
 #include "sha-hash-4way.h"

-#if defined(SHA256T_11WAY)
-
-static __thread sha256_11way_context sha256_ctx11 __attribute__ ((aligned (64)));
-
-void sha256t_11way_hash( void *outx, void *outy, void *outz, const void *inpx,
-	                 const void *inpy, const void*inpz )
-{
-   uint32_t hashx[8*8] __attribute__ ((aligned (64)));
-   uint32_t hashy[8*2] __attribute__ ((aligned (64)));
-   uint32_t hashz[8]   __attribute__ ((aligned (64)));
-   sha256_11way_context ctx;
-   const void *inpx64 = inpx+(64<<3);
-   const void *inpy64 = inpy+(64<<1);
-   const void *inpz64 = inpz+ 64;
-
-   memcpy( &ctx, &sha256_ctx11, sizeof ctx );
-   sha256_11way_update( &ctx, inpx64, inpy64, inpz64,  16 );
-   sha256_11way_close( &ctx, hashx, hashy, hashz );
-
-   sha256_11way_init( &ctx );
-   sha256_11way_update( &ctx, hashx, hashy, hashz, 32 );
-   sha256_11way_close( &ctx, hashx, hashy, hashz );
-
-   sha256_11way_init( &ctx );
-   sha256_11way_update( &ctx, hashx, hashy, hashz, 32 );
-   sha256_11way_close( &ctx, outx, outy, outz );
-}
-
-int scanhash_sha256t_11way( struct work *work, uint32_t max_nonce,
-	                    uint64_t *hashes_done, struct thr_info *mythr )
-{
-   uint32_t datax[20*8]  __attribute__ ((aligned (64)));
-   uint32_t datay[20*2]  __attribute__ ((aligned (32)));
-   uint32_t dataz[20]    __attribute__ ((aligned (32)));
-   uint32_t hashx[8*8]   __attribute__ ((aligned (32)));
-   uint32_t hashy[8*2]   __attribute__ ((aligned (32)));
-   uint32_t hashz[8]     __attribute__ ((aligned (32)));
-   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
-   uint32_t *hash7;
-   uint32_t *pdata = work->data;
-   uint32_t *ptarget = work->target;
-   const uint32_t Htarg = ptarget[7];
-   const uint32_t first_nonce = pdata[19];
-   uint32_t n = first_nonce;
-   __m256i  *noncex = (__m256i*) datax + 19;
-   __m64    *noncey = (__m64*)   datay + 19;
-   uint32_t *noncez = (uint32_t*)dataz + 19;
-   int thr_id = mythr->id;  // thr_id arg is deprecated
-   int i;
-   const uint64_t htmax[] = {           0,
-                                      0xF,
-                                     0xFF,
-                                    0xFFF,
-                                   0xFFFF,
-                               0x10000000 };
-   const uint32_t masks[] = {  0xFFFFFFFF,
-                               0xFFFFFFF0,
-                               0xFFFFFF00,
-                               0xFFFFF000,
-                               0xFFFF0000,
-                                        0 };
-
-   // Use dataz (scalar) to stage bswapped data for the vectors.
-   casti_m256i( dataz, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
-   casti_m256i( dataz, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
-   casti_m128i( dataz, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
-
-   intrlv_8x32( datax, dataz, dataz, dataz, dataz,
-                                 dataz, dataz, dataz, dataz, 640 );
-   mm64_interleave_2x32( datay, dataz, dataz, 640 );
-
-   sha256_11way_init( &sha256_ctx11 );
-   sha256_11way_update( &sha256_ctx11, datax, datay, dataz, 64 );
-
-   for ( int m = 0; m < 6; m++ ) if ( Htarg <= htmax[m] )
-   {
-      uint32_t mask = masks[m];
-      do
-      {
-        *noncex = mm256_bswap_32(
-         _mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n ) );
-        *noncey = mm64_bswap_32( _mm_set_pi32( n+9, n+8 ) );
-        *noncez = bswap_32( n+10 );
-
-        pdata[19] = n;
-
-        sha256t_11way_hash( hashx, hashy, hashz, datax, datay, dataz );
-
-        if ( opt_benchmark ) { n += 11; continue; }
-
-        hash7 = &(hashx[7<<3]); 
-        for ( i = 0; i < 8; i++ ) if ( !( hash7[ i ] & mask ) )
-        { 
-            // deinterleave hash for lane
-            extr_lane_8x32( lane_hash, hashx, i, 256 );
-            if ( fulltest( lane_hash, ptarget ) )
-            {
-	            pdata[19] = n + i;
-               submit_lane_solution( work, lane_hash, mythr, i );
-            }
-        }
-
-        hash7 = &(hashy[7<<1]);
-        for( i = 0; i < 2; i++ ) if ( !(hash7[ 0] & mask ) )
- 
-        {
-            mm64_extr_lane_2x32( lane_hash, hashy, i, 256 );
-           if ( fulltest( lane_hash, ptarget ) )
-           {
-               pdata[19] = n + 8 + i;
-               submit_lane_solution( work, lane_hash, mythr, i+8 );
-           }
-	     }
-
-        if ( !(hashz[7] & mask ) && fulltest( hashz, ptarget ) )
-        {
-            pdata[19] = n+10;
-            submit_lane_solution( work, hashz, mythr, 10 );
-        }
-        n += 11;
-
-      } while ( (n < max_nonce-12) && !work_restart[thr_id].restart );
-      break;
-   }
-    
-   *hashes_done = n - first_nonce + 1;
-   return 0;
-}
-
-#endif
-
 #if defined(SHA256T_8WAY)

 static __thread sha256_8way_context sha256_ctx8 __attribute__ ((aligned (64)));
--- a/algo/sha/sha512-hash-4way.c
+++ b/algo/sha/sha512-hash-4way.c
@@ -252,16 +252,6 @@ void sha512_4way_init( sha512_4way_context *sc )
 {
   sc->initialized = false;
   sc->count = 0;
-/*
-   sc->val[0] = _mm256_set1_epi64x( H512[0] );
-   sc->val[1] = _mm256_set1_epi64x( H512[1] );
-   sc->val[2] = _mm256_set1_epi64x( H512[2] );
-   sc->val[3] = _mm256_set1_epi64x( H512[3] );
-   sc->val[4] = _mm256_set1_epi64x( H512[4] );
-   sc->val[5] = _mm256_set1_epi64x( H512[5] );
-   sc->val[6] = _mm256_set1_epi64x( H512[6] );
-   sc->val[7] = _mm256_set1_epi64x( H512[7] );
-*/
 }

 void sha512_4way( sha512_4way_context *sc, const void *data, size_t len )
@@ -295,6 +285,8 @@ void sha512_4way_close( sha512_4way_context *sc, void *dst )
    unsigned ptr;
    const int buf_size = 128;
    const int pad = buf_size - 16;
+    const __m256i shuff_bswap64 = m256_const2_64( 0x08090a0b0c0d0e0f,
+                                                  0x0001020304050607 );

    ptr = (unsigned)sc->count & (buf_size - 1U);
    sc->buf[ ptr>>3 ] = m256_const1_64( 0x80 );
@@ -308,10 +300,10 @@ void sha512_4way_close( sha512_4way_context *sc, void *dst )
    else
         memset_zero_256( sc->buf + (ptr>>3), (pad - ptr) >> 3 );

-    sc->buf[ pad >> 3 ] =
-                 mm256_bswap_64( _mm256_set1_epi64x( sc->count >> 61 ) );
-    sc->buf[ ( pad+8 ) >> 3 ] = 
-                 mm256_bswap_64( _mm256_set1_epi64x( sc->count << 3 ) );
+    sc->buf[ pad >> 3 ] = _mm256_shuffle_epi8(
+                       _mm256_set1_epi64x( sc->count >> 61 ), shuff_bswap64 );
+    sc->buf[ ( pad+8 ) >> 3 ] = _mm256_shuffle_epi8( 
+                       _mm256_set1_epi64x( sc->count <<  3 ), shuff_bswap64 );
    sha512_4way_round( sc, sc->buf, sc->val );

    mm256_block_bswap_64( dst, sc->val );
--- a/algo/shavite/shavite-hash-2way.c
+++ b/algo/shavite/shavite-hash-2way.c
@@ -5,6 +5,7 @@

 #if defined(__AVX2__)

+
 static const uint32_t IV512[] =
 {
        0x72FCCDD8, 0x79CA4727, 0x128A077B, 0x40D55AEC,
@@ -13,6 +14,7 @@ static const uint32_t IV512[] =
        0xE275EADE, 0x502D9FCD, 0xB9357178, 0x022A4B9A
 };

+
 #define mm256_ror2x256hi_1x32( a, b ) \
   _mm256_blend_epi32( mm256_ror1x32_128( a ), \
                       mm256_ror1x32_128( b ), 0x88 )
@@ -232,18 +234,14 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )

 void shavite512_2way_init( shavite512_2way_context *ctx )
 {
-   casti_m256i( ctx->h, 0 ) =
-            _mm256_set_epi32( IV512[ 3], IV512[ 2], IV512[ 1], IV512[ 0],
-                              IV512[ 3], IV512[ 2], IV512[ 1], IV512[ 0] );  
-   casti_m256i( ctx->h, 1 ) =
-            _mm256_set_epi32( IV512[ 7], IV512[ 6], IV512[ 5], IV512[ 4],
-                              IV512[ 7], IV512[ 6], IV512[ 5], IV512[ 4] );
-   casti_m256i( ctx->h, 2 ) =
-            _mm256_set_epi32( IV512[11], IV512[10], IV512[ 9], IV512[ 8],
-                              IV512[11], IV512[10], IV512[ 9], IV512[ 8] );
-   casti_m256i( ctx->h, 3 ) =
-            _mm256_set_epi32( IV512[15], IV512[14], IV512[13], IV512[12],
-                              IV512[15], IV512[14], IV512[13], IV512[12] );
+    __m256i *h = (__m256i*)ctx->h;
+    __m128i *iv = (__m128i*)IV512;
+   
+   h[0] = m256_const1_128( iv[0] );
+   h[1] = m256_const1_128( iv[1] );
+   h[2] = m256_const1_128( iv[2] );
+   h[3] = m256_const1_128( iv[3] );
+
   ctx->ptr    = 0;
   ctx->count0 = 0;
   ctx->count1 = 0;
@@ -251,6 +249,7 @@ void shavite512_2way_init( shavite512_2way_context *ctx )
   ctx->count3 = 0;
 }

+// not tested, use update_close
 void shavite512_2way_update( shavite512_2way_context *ctx, const void *data,
                             size_t len )
 {
@@ -287,6 +286,7 @@ void shavite512_2way_update( shavite512_2way_context *ctx, const void *data,
   ctx->ptr = ptr;
 }

+// not tested
 void shavite512_2way_close( shavite512_2way_context *ctx, void *dst )
 {
    unsigned char *buf;
@@ -300,7 +300,7 @@ void shavite512_2way_close( shavite512_2way_context *ctx, void *dst )
    uint32_t vp = ctx->ptr>>5;

    // Terminating byte then zero pad
-    casti_m256i( buf, vp++ ) = _mm256_set_epi32( 0,0,0,0x80, 0,0,0,0x80 );
+    casti_m256i( buf, vp++ ) = m256_const2_64( 0, 0x0000000000000080 );

    // Zero pad full vectors up to count
    for ( ; vp < 6; vp++ )      
@@ -314,14 +314,12 @@ void shavite512_2way_close( shavite512_2way_context *ctx, void *dst )
    count.u32[2] = ctx->count2;
    count.u32[3] = ctx->count3;

-    casti_m256i( buf, 6 ) = _mm256_set_epi16( count.u16[0], 0,0,0,0,0,0,0,
-		                              count.u16[0], 0,0,0,0,0,0,0 );
-    casti_m256i( buf, 7 ) = _mm256_set_epi16(
-		    0x0200      , count.u16[7], count.u16[6], count.u16[5],
-		    count.u16[4], count.u16[3], count.u16[2], count.u16[1],
-                    0x0200      , count.u16[7], count.u16[6], count.u16[5],
-                    count.u16[4], count.u16[3], count.u16[2], count.u16[1] );
-
+    casti_m256i( buf, 6 ) = m256_const1_128(
+                  _mm_insert_epi16( m128_zero, count.u16[0], 7 ) ); 
+    casti_m256i( buf, 7 ) = m256_const1_128( _mm_set_epi16(
+                  0x0200,       count.u16[7], count.u16[6], count.u16[5],
+                  count.u16[4], count.u16[3], count.u16[2], count.u16[1] ) );
+                
    c512_2way( ctx, buf);

    casti_m256i( dst, 0 ) = casti_m256i( ctx->h, 0 );
@@ -382,23 +380,21 @@ void shavite512_2way_update_close( shavite512_2way_context *ctx, void *dst,

   if ( vp == 0 )    // empty buf, xevan.
   { 
-      casti_m256i( buf, 0 ) = _mm256_set_epi32( 0,0,0,0x80, 0,0,0,0x80 );
+      casti_m256i( buf, 0 ) = m256_const2_64( 0, 0x0000000000000080 );
      memset_zero_256( (__m256i*)buf + 1, 5 );
      ctx->count0 = ctx->count1 = ctx->count2 = ctx->count3 = 0;
   }
   else     // half full buf, everyone else.
   {
-      casti_m256i( buf, vp++ ) = _mm256_set_epi32( 0,0,0,0x80, 0,0,0,0x80 );
+    casti_m256i( buf, vp++ ) = m256_const2_64( 0, 0x0000000000000080 );
      memset_zero_256( (__m256i*)buf + vp, 6 - vp );
   }

-   casti_m256i( buf, 6 ) = _mm256_set_epi16( count.u16[0], 0,0,0,0,0,0,0,
-                                             count.u16[0], 0,0,0,0,0,0,0 );
-   casti_m256i( buf, 7 ) = _mm256_set_epi16(
-                   0x0200      , count.u16[7], count.u16[6], count.u16[5],
-                   count.u16[4], count.u16[3], count.u16[2], count.u16[1],
-                   0x0200      , count.u16[7], count.u16[6], count.u16[5],
-                   count.u16[4], count.u16[3], count.u16[2], count.u16[1] );
+    casti_m256i( buf, 6 ) = m256_const1_128(
+                  _mm_insert_epi16( m128_zero, count.u16[0], 7 ) ); 
+    casti_m256i( buf, 7 ) = m256_const1_128( _mm_set_epi16(
+                  0x0200,       count.u16[7], count.u16[6], count.u16[5],
+                  count.u16[4], count.u16[3], count.u16[2], count.u16[1] ) );

   c512_2way( ctx, buf);

--- a/algo/simd/nist.c
+++ b/algo/simd/nist.c
@@ -83,13 +83,14 @@ HashReturn init_sd(hashState_sd *state, int hashbitlen) {
  char *init;

 #ifndef NO_PRECOMPUTED_IV
-  if (hashbitlen == 224)
-    r=InitIV(state, hashbitlen, IV_224);
-  else if (hashbitlen == 256)
-    r=InitIV(state, hashbitlen, IV_256);
-  else if (hashbitlen == 384)
-    r=InitIV(state, hashbitlen, IV_384);
-  else if (hashbitlen == 512)
+//  if (hashbitlen == 224)
+//    r=InitIV(state, hashbitlen, IV_224);
+//  else if (hashbitlen == 256)
+//    r=InitIV(state, hashbitlen, IV_256);
+//  else if (hashbitlen == 384)
+//    r=InitIV(state, hashbitlen, IV_384);
+//  else
+  if (hashbitlen == 512)
    r=InitIV(state, hashbitlen, IV_512);
  else
 #endif
--- a/algo/simd/simd-hash-2way.c
+++ b/algo/simd/simd-hash-2way.c
@@ -110,14 +110,26 @@ static const m256_v16 FFT256_Twiddle[] =

 // imported from vector.c

+#define REDUCE(x) \
+  _mm256_sub_epi16( _mm256_and_si256( x, m256_const1_64( \
+                         0x00ff00ff00ff00ff ) ), _mm256_srai_epi16( x, 8 ) )
+/*
 #define REDUCE(x) \
  _mm256_sub_epi16( _mm256_and_si256( x, _mm256_set1_epi16( 255 ) ), \
                                         _mm256_srai_epi16( x, 8 ) )
+*/

+#define EXTRA_REDUCE_S(x)\
+  _mm256_sub_epi16( x, _mm256_and_si256( \
+             m256_const1_64( 0x0101010101010101 ), \
+             _mm256_cmpgt_epi16( x, m256_const1_64( 0x0080008000800080 ) ) ) )
+
+/*
 #define EXTRA_REDUCE_S(x)\
  _mm256_sub_epi16( x, \
         _mm256_and_si256( _mm256_set1_epi16( 257 ), \
                           _mm256_cmpgt_epi16( x, _mm256_set1_epi16( 128 ) ) ) )
+*/

 #define REDUCE_FULL_S( x )  EXTRA_REDUCE_S( REDUCE (x ) )

--- a/algo/skein/skein-4way.c
+++ b/algo/skein/skein-4way.c
@@ -2,13 +2,136 @@
 #include <string.h>
 #include <stdint.h>
 #include "skein-hash-4way.h"
+
+// 8 way is faster than SHA on Icelake
+// SHA is faster than 4 way on Ryzen
+//
 #if defined(__SHA__)
  #include <openssl/sha.h>
-#else
-  #include "algo/sha/sha-hash-4way.h"
 #endif
+#include "algo/sha/sha-hash-4way.h"

-#if defined (SKEIN_4WAY)
+#if defined (SKEIN_8WAY)
+
+void skeinhash_8way( void *state, const void *input )
+{
+     uint64_t vhash64[16*8] __attribute__ ((aligned (128)));
+     skein512_8way_context ctx_skein;
+
+//#if defined(__SHA__)
+//     uint32_t hash0[16] __attribute__ ((aligned (64)));
+//     uint32_t hash1[16] __attribute__ ((aligned (64)));
+//     uint32_t hash2[16] __attribute__ ((aligned (64)));
+//     uint32_t hash3[16] __attribute__ ((aligned (64)));
+//     uint32_t hash4[16] __attribute__ ((aligned (64)));
+//     uint32_t hash5[16] __attribute__ ((aligned (64)));
+//     uint32_t hash6[16] __attribute__ ((aligned (64)));
+//     uint32_t hash7[16] __attribute__ ((aligned (64)));
+//     SHA256_CTX           ctx_sha256;
+//#else
+     uint32_t vhash32[32*8] __attribute__ ((aligned (128)));
+     sha256_8way_context ctx_sha256;
+//#endif
+
+     skein512_8way_init( &ctx_skein );
+     skein512_8way_update( &ctx_skein, input, 80 );
+     skein512_8way_close( &ctx_skein, vhash64 );
+/*
+#if defined(__SHA__)      
+     dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+                   vhash64, 512 );
+
+     SHA256_Init( &ctx_sha256 );
+     SHA256_Update( &ctx_sha256, (unsigned char*)hash0, 64 );
+     SHA256_Final( (unsigned char*)hash0, &ctx_sha256 );
+
+     SHA256_Init( &ctx_sha256 );
+     SHA256_Update( &ctx_sha256, (unsigned char*)hash1, 64 );
+     SHA256_Final( (unsigned char*)hash1, &ctx_sha256 );
+
+     SHA256_Init( &ctx_sha256 );
+     SHA256_Update( &ctx_sha256, (unsigned char*)hash2, 64 );
+     SHA256_Final( (unsigned char*)hash2, &ctx_sha256 );
+
+     SHA256_Init( &ctx_sha256 );
+     SHA256_Update( &ctx_sha256, (unsigned char*)hash3, 64 );
+     SHA256_Final( (unsigned char*)hash3, &ctx_sha256 );
+
+     SHA256_Init( &ctx_sha256 );
+     SHA256_Update( &ctx_sha256, (unsigned char*)hash4, 64 );
+     SHA256_Final( (unsigned char*)hash4, &ctx_sha256 );
+
+     SHA256_Init( &ctx_sha256 );
+     SHA256_Update( &ctx_sha256, (unsigned char*)hash5, 64 );
+     SHA256_Final( (unsigned char*)hash5, &ctx_sha256 );
+
+     SHA256_Init( &ctx_sha256 );
+     SHA256_Update( &ctx_sha256, (unsigned char*)hash6, 64 );
+     SHA256_Final( (unsigned char*)hash6, &ctx_sha256 );
+
+     SHA256_Init( &ctx_sha256 );
+     SHA256_Update( &ctx_sha256, (unsigned char*)hash7, 64 );
+     SHA256_Final( (unsigned char*)hash7, &ctx_sha256 );
+     
+     intrlv_8x32( state, hash0, hash1, hash2, hash3, hash4, hash5, hash6,
+                  hash7, 256 );
+#else
+*/
+
+     rintrlv_8x64_8x32( vhash32, vhash64, 512 );
+//     dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+//                   vhash64, 512 );
+//     intrlv_8x32( vhash32, hash0, hash1, hash2, hash3, hash4, hash5, hash6,
+//                   hash7, 512 );
+
+     sha256_8way_init( &ctx_sha256 );
+     sha256_8way( &ctx_sha256, vhash32, 64 );
+     sha256_8way_close( &ctx_sha256, state );
+//#endif
+}
+
+int scanhash_skein_8way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr )
+{
+    uint32_t vdata[20*8] __attribute__ ((aligned (128)));
+    uint32_t hash[16*8] __attribute__ ((aligned (64)));
+    uint32_t lane_hash[8] __attribute__ ((aligned (64)));
+    uint32_t *hash7 = &(hash[7<<3]);
+    uint32_t *pdata = work->data;
+    uint32_t *ptarget = work->target;
+    const uint32_t Htarg = ptarget[7];
+    const uint32_t first_nonce = pdata[19];
+    uint32_t n = first_nonce;
+    __m512i  *noncev = (__m512i*)vdata + 9;   // aligned
+    int thr_id = mythr->id; 
+
+   mm512_bswap32_intrlv80_8x64( vdata, pdata );
+   do
+   {
+       *noncev = mm512_intrlv_blend_32( mm512_bswap_32(
+                _mm512_set_epi32( n+7, 0, n+6, 0, n+5, 0, n+4, 0,
+                                  n+3, 0, n+2, 0, n+1, 0, n  , 0 ) ), *noncev );
+
+       skeinhash_8way( hash, vdata );
+
+       for ( int lane = 0; lane < 8; lane++ )
+       if (  hash7[ lane ] <= Htarg )
+       {
+          extr_lane_8x32( lane_hash, hash, lane, 256 );
+          if ( fulltest( lane_hash, ptarget ) )
+          {
+             pdata[19] = n + lane;
+             submit_lane_solution( work, lane_hash, mythr, lane );
+          }
+       }
+       n += 8;
+    } while ( (n < max_nonce-8) && !work_restart[thr_id].restart );
+
+    *hashes_done = n - first_nonce;
+    return 0;
+}
+
+#elif defined (SKEIN_4WAY)

 void skeinhash_4way( void *state, const void *input )
 {
@@ -26,7 +149,7 @@ void skeinhash_4way( void *state, const void *input )
 #endif

     skein512_4way_init( &ctx_skein );
-     skein512_4way( &ctx_skein, input, 80 );
+     skein512_4way_update( &ctx_skein, input, 80 );
     skein512_4way_close( &ctx_skein, vhash64 );

 #if defined(__SHA__)      
@@ -71,7 +194,7 @@ int scanhash_skein_4way( struct work *work, uint32_t max_nonce,
    const uint32_t first_nonce = pdata[19];
    uint32_t n = first_nonce;
    __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
-    int thr_id = mythr->id;  // thr_id arg is deprecated
+    int thr_id = mythr->id; 

   mm256_bswap32_intrlv80_4x64( vdata, pdata );
   do
@@ -92,9 +215,9 @@ int scanhash_skein_4way( struct work *work, uint32_t max_nonce,
          }
       }
       n += 4;
-    } while ( (n < max_nonce) && !work_restart[thr_id].restart );
+    } while ( (n < max_nonce-4) && !work_restart[thr_id].restart );

-    *hashes_done = n - first_nonce + 1;
+    *hashes_done = n - first_nonce;
    return 0;
 }

--- a/algo/skein/skein-gate.c
+++ b/algo/skein/skein-gate.c
@@ -4,8 +4,11 @@

 bool register_skein_algo( algo_gate_t* gate )
 {
-    gate->optimizations = AVX2_OPT | SHA_OPT;
-#if defined (SKEIN_4WAY)
+    gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
+#if defined (SKEIN_8WAY)
+    gate->scanhash  = (void*)&scanhash_skein_8way;
+    gate->hash      = (void*)&skeinhash_8way;
+#elif defined (SKEIN_4WAY)
    gate->scanhash  = (void*)&scanhash_skein_4way;
    gate->hash      = (void*)&skeinhash_4way;
 #else
@@ -15,3 +18,20 @@ bool register_skein_algo( algo_gate_t* gate )
    return true;
 };

+bool register_skein2_algo( algo_gate_t* gate )
+{
+  gate->optimizations = AVX2_OPT | AVX512_OPT;
+#if defined (SKEIN_8WAY)
+  gate->scanhash  = (void*)&scanhash_skein2_8way;
+  gate->hash      = (void*)&skein2hash_8way;
+#elif defined (SKEIN_4WAY)
+  gate->scanhash  = (void*)&scanhash_skein2_4way;
+  gate->hash      = (void*)&skein2hash_4way;
+#else
+  gate->scanhash  = (void*)&scanhash_skein2;
+  gate->hash      = (void*)&skein2hash;
+#endif
+  return true;
+};
+
+
--- a/algo/skein/skein-gate.h
+++ b/algo/skein/skein-gate.h
@@ -1,23 +1,44 @@
 #ifndef __SKEIN_GATE_H__
-#define __SKEIN_GATE_H__
+#define __SKEIN_GATE_H__ 1
 #include <stdint.h>
 #include "algo-gate-api.h"

-#if defined(__AVX2__)
-  #define SKEIN_4WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define SKEIN_8WAY 1
+#elif defined(__AVX2__)
+  #define SKEIN_4WAY 1
 #endif

-#if defined(SKEIN_4WAY)
+#if defined(SKEIN_8WAY)
+
+void skeinhash_8way( void *output, const void *input );
+int scanhash_skein_8way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+
+void skein2hash_8way( void *output, const void *input );
+int scanhash_skein2_8way( struct work *work, uint32_t max_nonce,
+                          uint64_t* hashes_done, struct thr_info *mythr );
+
+#elif defined(SKEIN_4WAY)

 void skeinhash_4way( void *output, const void *input );
-
 int scanhash_skein_4way( struct work *work, uint32_t max_nonce,
                         uint64_t *hashes_done, struct thr_info *mythr );
-#endif
+
+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 );
+
+#else

 void skeinhash( void *output, const void *input );
-
 int scanhash_skein( struct work *work, uint32_t max_nonce,
                    uint64_t *hashes_done, struct thr_info *mythr );

+void skein2hash( void *output, const void *input );
+int scanhash_skein2( struct work *work, uint32_t max_nonce,
+                     uint64_t *hashes_done, struct thr_info *mythr );
+
+#endif
+
 #endif
--- a/algo/skein/skein-hash-4way.c
+++ b/algo/skein/skein-hash-4way.c
@@ -36,7 +36,6 @@
 #include <string.h>
 #include "skein-hash-4way.h"

-
 #ifdef __cplusplus
 extern "C"{
 #endif
@@ -45,6 +44,22 @@ extern "C"{
 #pragma warning (disable: 4146)
 #endif

+/*
+static const sph_u64 IV256[] = {
+   SPH_C64(0xCCD044A12FDB3E13), SPH_C64(0xE83590301A79A9EB),
+   SPH_C64(0x55AEA0614F816E6F), SPH_C64(0x2A2767A4AE9B94DB),
+   SPH_C64(0xEC06025E74DD7683), SPH_C64(0xE7A436CDC4746251),
+   SPH_C64(0xC36FBAF9393AD185), SPH_C64(0x3EEDBA1833EDFC13)
+};
+
+static const sph_u64 IV512[] = {
+   SPH_C64(0x4903ADFF749C51CE), SPH_C64(0x0D95DE399746DF03),
+   SPH_C64(0x8FD1934127C79BCE), SPH_C64(0x9A255629FF352CB1),
+   SPH_C64(0x5DB62599DF6CA7B0), SPH_C64(0xEABE394CA9D5C3F4),
+   SPH_C64(0x991112C71A75B523), SPH_C64(0xAE18A40B660FCC33)
+};
+*/
+   
 /*
 * M9_ ## s ## _ ## i  evaluates to s+i mod 9 (0 <= s <= 18, 0 <= i <= 7).
 */
@@ -270,8 +285,151 @@ extern "C"{
 #define SKBI(k, s, i)   XCAT(k, XCAT(XCAT(XCAT(M9_, s), _), i))
 #define SKBT(t, s, v)   XCAT(t, XCAT(XCAT(XCAT(M3_, s), _), v))

+#define READ_STATE_BIG(sc)   do { \
+      h0 = (sc)->h0; \
+      h1 = (sc)->h1; \
+      h2 = (sc)->h2; \
+      h3 = (sc)->h3; \
+      h4 = (sc)->h4; \
+      h5 = (sc)->h5; \
+      h6 = (sc)->h6; \
+      h7 = (sc)->h7; \
+      bcount = sc->bcount; \
+   } while (0)
+
+#define WRITE_STATE_BIG(sc)   do { \
+      (sc)->h0 = h0; \
+      (sc)->h1 = h1; \
+      (sc)->h2 = h2; \
+      (sc)->h3 = h3; \
+      (sc)->h4 = h4; \
+      (sc)->h5 = h5; \
+      (sc)->h6 = h6; \
+      (sc)->h7 = h7; \
+      sc->bcount = bcount; \
+   } while (0)
+   
 // AVX2 all scalar vars are now vectors representing 4 nonces in parallel

+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define TFBIG_KINIT_8WAY( k0, k1, k2, k3, k4, k5, k6, k7, k8, t0, t1, t2 ) \
+do { \
+  k8 = _mm512_xor_si512( _mm512_xor_si512( \
+                            _mm512_xor_si512( _mm512_xor_si512( k0, k1 ), \
+                                              _mm512_xor_si512( k2, k3 ) ), \
+                            _mm512_xor_si512( _mm512_xor_si512( k4, k5 ), \
+                                              _mm512_xor_si512( k6, k7 ) ) ), \
+                         m512_const1_64( 0x1BD11BDAA9FC1A22) ); \
+  t2 = t0 ^ t1; \
+} while (0)
+   
+#define TFBIG_ADDKEY_8WAY(w0, w1, w2, w3, w4, w5, w6, w7, k, t, s) \
+do { \
+  w0 = _mm512_add_epi64( w0, SKBI(k,s,0) ); \
+  w1 = _mm512_add_epi64( w1, SKBI(k,s,1) ); \
+  w2 = _mm512_add_epi64( w2, SKBI(k,s,2) ); \
+  w3 = _mm512_add_epi64( w3, SKBI(k,s,3) ); \
+  w4 = _mm512_add_epi64( w4, SKBI(k,s,4) ); \
+  w5 = _mm512_add_epi64( w5, _mm512_add_epi64( SKBI(k,s,5), \
+                                         m512_const1_64( SKBT(t,s,0) ) ) ); \
+  w6 = _mm512_add_epi64( w6, _mm512_add_epi64( SKBI(k,s,6), \
+                                         m512_const1_64( SKBT(t,s,1) ) ) ); \
+  w7 = _mm512_add_epi64( w7, _mm512_add_epi64( SKBI(k,s,7), \
+                                         m512_const1_64( s ) ) ); \
+} while (0)
+
+
+#define TFBIG_MIX_8WAY(x0, x1, rc) \
+do { \
+     x0 = _mm512_add_epi64( x0, x1 ); \
+     x1 = _mm512_xor_si512( mm512_rol_64( x1, rc ), x0 ); \
+} while (0)
+
+#define TFBIG_MIX8_8WAY(w0, w1, w2, w3, w4, w5, w6, w7, rc0, rc1, rc2, rc3)  do { \
+      TFBIG_MIX_8WAY(w0, w1, rc0); \
+      TFBIG_MIX_8WAY(w2, w3, rc1); \
+      TFBIG_MIX_8WAY(w4, w5, rc2); \
+      TFBIG_MIX_8WAY(w6, w7, rc3); \
+   } while (0)
+
+#define TFBIG_8WAY_4e(s)   do { \
+      TFBIG_ADDKEY_8WAY(p0, p1, p2, p3, p4, p5, p6, p7, h, t, s); \
+      TFBIG_MIX8_8WAY(p0, p1, p2, p3, p4, p5, p6, p7, 46, 36, 19, 37); \
+      TFBIG_MIX8_8WAY(p2, p1, p4, p7, p6, p5, p0, p3, 33, 27, 14, 42); \
+      TFBIG_MIX8_8WAY(p4, p1, p6, p3, p0, p5, p2, p7, 17, 49, 36, 39); \
+      TFBIG_MIX8_8WAY(p6, p1, p0, p7, p2, p5, p4, p3, 44,  9, 54, 56); \
+   } while (0)
+
+#define TFBIG_8WAY_4o(s)   do { \
+      TFBIG_ADDKEY_8WAY(p0, p1, p2, p3, p4, p5, p6, p7, h, t, s); \
+      TFBIG_MIX8_8WAY(p0, p1, p2, p3, p4, p5, p6, p7, 39, 30, 34, 24); \
+      TFBIG_MIX8_8WAY(p2, p1, p4, p7, p6, p5, p0, p3, 13, 50, 10, 17); \
+      TFBIG_MIX8_8WAY(p4, p1, p6, p3, p0, p5, p2, p7, 25, 29, 39, 43); \
+      TFBIG_MIX8_8WAY(p6, p1, p0, p7, p2, p5, p4, p3,  8, 35, 56, 22); \
+   } while (0)
+
+#define UBI_BIG_8WAY(etype, extra) \
+do { \
+  sph_u64 t0, t1, t2; \
+  __m512i h8; \
+  __m512i m0 =  buf[0]; \
+  __m512i m1 =  buf[1]; \
+  __m512i m2 =  buf[2]; \
+  __m512i m3 =  buf[3]; \
+  __m512i m4 =  buf[4]; \
+  __m512i m5 =  buf[5]; \
+  __m512i m6 =  buf[6]; \
+  __m512i m7 =  buf[7]; \
+\
+  __m512i p0 = m0; \
+  __m512i p1 = m1; \
+  __m512i p2 = m2; \
+  __m512i p3 = m3; \
+  __m512i p4 = m4; \
+  __m512i p5 = m5; \
+  __m512i p6 = m6; \
+  __m512i p7 = m7; \
+  t0 = SPH_T64(bcount << 6) + (sph_u64)(extra); \
+  t1 = (bcount >> 58) + ((sph_u64)(etype) << 55); \
+  TFBIG_KINIT_8WAY(h0, h1, h2, h3, h4, h5, h6, h7, h8, t0, t1, t2); \
+  TFBIG_8WAY_4e(0); \
+  TFBIG_8WAY_4o(1); \
+  TFBIG_8WAY_4e(2); \
+  TFBIG_8WAY_4o(3); \
+  TFBIG_8WAY_4e(4); \
+  TFBIG_8WAY_4o(5); \
+  TFBIG_8WAY_4e(6); \
+  TFBIG_8WAY_4o(7); \
+  TFBIG_8WAY_4e(8); \
+  TFBIG_8WAY_4o(9); \
+  TFBIG_8WAY_4e(10); \
+  TFBIG_8WAY_4o(11); \
+  TFBIG_8WAY_4e(12); \
+  TFBIG_8WAY_4o(13); \
+  TFBIG_8WAY_4e(14); \
+  TFBIG_8WAY_4o(15); \
+  TFBIG_8WAY_4e(16); \
+  TFBIG_8WAY_4o(17); \
+  TFBIG_ADDKEY_8WAY(p0, p1, p2, p3, p4, p5, p6, p7, h, t, 18); \
+  h0 = _mm512_xor_si512( m0, p0 );\
+  h1 = _mm512_xor_si512( m1, p1 );\
+  h2 = _mm512_xor_si512( m2, p2 );\
+  h3 = _mm512_xor_si512( m3, p3 );\
+  h4 = _mm512_xor_si512( m4, p4 );\
+  h5 = _mm512_xor_si512( m5, p5 );\
+  h6 = _mm512_xor_si512( m6, p6 );\
+  h7 = _mm512_xor_si512( m7, p7 );\
+} while (0)
+
+#define DECL_STATE_BIG_8WAY \
+  __m512i h0, h1, h2, h3, h4, h5, h6, h7; \
+  sph_u64 bcount;
+
+
+#endif // AVX512
+
 #define TFBIG_KINIT_4WAY( k0, k1, k2, k3, k4, k5, k6, k7, k8, t0, t1, t2 ) \
 do { \
  k8 = _mm256_xor_si256( _mm256_xor_si256( \
@@ -298,39 +456,34 @@ do { \
                                         m256_const1_64( s ) ) ); \
 } while (0)

-
 #define TFBIG_MIX_4WAY(x0, x1, rc) \
 do { \
     x0 = _mm256_add_epi64( x0, x1 ); \
     x1 = _mm256_xor_si256( mm256_rol_64( x1, rc ), x0 ); \
 } while (0)
- 

-// typeless
-#define TFBIG_MIX8(w0, w1, w2, w3, w4, w5, w6, w7, rc0, rc1, rc2, rc3)  do { \
-		TFBIG_MIX_4WAY(w0, w1, rc0); \
-		TFBIG_MIX_4WAY(w2, w3, rc1); \
-		TFBIG_MIX_4WAY(w4, w5, rc2); \
-		TFBIG_MIX_4WAY(w6, w7, rc3); \
-	} while (0)
+#define TFBIG_MIX8_4WAY(w0, w1, w2, w3, w4, w5, w6, w7, rc0, rc1, rc2, rc3)  do { \
+      TFBIG_MIX_4WAY(w0, w1, rc0); \
+      TFBIG_MIX_4WAY(w2, w3, rc1); \
+      TFBIG_MIX_4WAY(w4, w5, rc2); \
+      TFBIG_MIX_4WAY(w6, w7, rc3); \
+   } while (0)

+#define TFBIG_4WAY_4e(s)   do { \
+      TFBIG_ADDKEY_4WAY(p0, p1, p2, p3, p4, p5, p6, p7, h, t, s); \
+      TFBIG_MIX8_4WAY(p0, p1, p2, p3, p4, p5, p6, p7, 46, 36, 19, 37); \
+      TFBIG_MIX8_4WAY(p2, p1, p4, p7, p6, p5, p0, p3, 33, 27, 14, 42); \
+      TFBIG_MIX8_4WAY(p4, p1, p6, p3, p0, p5, p2, p7, 17, 49, 36, 39); \
+      TFBIG_MIX8_4WAY(p6, p1, p0, p7, p2, p5, p4, p3, 44,  9, 54, 56); \
+   } while (0)

-#define TFBIG_4e(s)   do { \
-		TFBIG_ADDKEY_4WAY(p0, p1, p2, p3, p4, p5, p6, p7, h, t, s); \
-		TFBIG_MIX8(p0, p1, p2, p3, p4, p5, p6, p7, 46, 36, 19, 37); \
-		TFBIG_MIX8(p2, p1, p4, p7, p6, p5, p0, p3, 33, 27, 14, 42); \
-		TFBIG_MIX8(p4, p1, p6, p3, p0, p5, p2, p7, 17, 49, 36, 39); \
-		TFBIG_MIX8(p6, p1, p0, p7, p2, p5, p4, p3, 44,  9, 54, 56); \
-	} while (0)
-
-#define TFBIG_4o(s)   do { \
-		TFBIG_ADDKEY_4WAY(p0, p1, p2, p3, p4, p5, p6, p7, h, t, s); \
-		TFBIG_MIX8(p0, p1, p2, p3, p4, p5, p6, p7, 39, 30, 34, 24); \
-		TFBIG_MIX8(p2, p1, p4, p7, p6, p5, p0, p3, 13, 50, 10, 17); \
-		TFBIG_MIX8(p4, p1, p6, p3, p0, p5, p2, p7, 25, 29, 39, 43); \
-		TFBIG_MIX8(p6, p1, p0, p7, p2, p5, p4, p3,  8, 35, 56, 22); \
-	} while (0)
-
+#define TFBIG_4WAY_4o(s)   do { \
+      TFBIG_ADDKEY_4WAY(p0, p1, p2, p3, p4, p5, p6, p7, h, t, s); \
+      TFBIG_MIX8_4WAY(p0, p1, p2, p3, p4, p5, p6, p7, 39, 30, 34, 24); \
+      TFBIG_MIX8_4WAY(p2, p1, p4, p7, p6, p5, p0, p3, 13, 50, 10, 17); \
+      TFBIG_MIX8_4WAY(p4, p1, p6, p3, p0, p5, p2, p7, 25, 29, 39, 43); \
+      TFBIG_MIX8_4WAY(p6, p1, p0, p7, p2, p5, p4, p3,  8, 35, 56, 22); \
+   } while (0)

 // scale buf offset by 4
 #define UBI_BIG_4WAY(etype, extra) \
@@ -357,24 +510,24 @@ do { \
  t0 = SPH_T64(bcount << 6) + (sph_u64)(extra); \
  t1 = (bcount >> 58) + ((sph_u64)(etype) << 55); \
  TFBIG_KINIT_4WAY(h0, h1, h2, h3, h4, h5, h6, h7, h8, t0, t1, t2); \
-  TFBIG_4e(0); \
-  TFBIG_4o(1); \
-  TFBIG_4e(2); \
-  TFBIG_4o(3); \
-  TFBIG_4e(4); \
-  TFBIG_4o(5); \
-  TFBIG_4e(6); \
-  TFBIG_4o(7); \
-  TFBIG_4e(8); \
-  TFBIG_4o(9); \
-  TFBIG_4e(10); \
-  TFBIG_4o(11); \
-  TFBIG_4e(12); \
-  TFBIG_4o(13); \
-  TFBIG_4e(14); \
-  TFBIG_4o(15); \
-  TFBIG_4e(16); \
-  TFBIG_4o(17); \
+  TFBIG_4WAY_4e(0); \
+  TFBIG_4WAY_4o(1); \
+  TFBIG_4WAY_4e(2); \
+  TFBIG_4WAY_4o(3); \
+  TFBIG_4WAY_4e(4); \
+  TFBIG_4WAY_4o(5); \
+  TFBIG_4WAY_4e(6); \
+  TFBIG_4WAY_4o(7); \
+  TFBIG_4WAY_4e(8); \
+  TFBIG_4WAY_4o(9); \
+  TFBIG_4WAY_4e(10); \
+  TFBIG_4WAY_4o(11); \
+  TFBIG_4WAY_4e(12); \
+  TFBIG_4WAY_4o(13); \
+  TFBIG_4WAY_4e(14); \
+  TFBIG_4WAY_4o(15); \
+  TFBIG_4WAY_4e(16); \
+  TFBIG_4WAY_4o(17); \
  TFBIG_ADDKEY_4WAY(p0, p1, p2, p3, p4, p5, p6, p7, h, t, 18); \
  h0 = _mm256_xor_si256( m0, p0 );\
  h1 = _mm256_xor_si256( m1, p1 );\
@@ -391,45 +544,142 @@ do { \
  __m256i h0, h1, h2, h3, h4, h5, h6, h7; \
  sph_u64 bcount;

-#define READ_STATE_BIG(sc)   do { \
-		h0 = (sc)->h0; \
-		h1 = (sc)->h1; \
-		h2 = (sc)->h2; \
-		h3 = (sc)->h3; \
-		h4 = (sc)->h4; \
-		h5 = (sc)->h5; \
-		h6 = (sc)->h6; \
-		h7 = (sc)->h7; \
-		bcount = sc->bcount; \
-	} while (0)
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

-#define WRITE_STATE_BIG(sc)   do { \
-		(sc)->h0 = h0; \
-		(sc)->h1 = h1; \
-		(sc)->h2 = h2; \
-		(sc)->h3 = h3; \
-		(sc)->h4 = h4; \
-		(sc)->h5 = h5; \
-		(sc)->h6 = h6; \
-		(sc)->h7 = h7; \
-		sc->bcount = bcount; \
-	} while (0)
+void skein256_8way_init( skein256_8way_context *sc )
+{
+        sc->h0 = m512_const1_64( 0xCCD044A12FDB3E13 );
+        sc->h1 = m512_const1_64( 0xE83590301A79A9EB );
+        sc->h2 = m512_const1_64( 0x55AEA0614F816E6F );
+        sc->h3 = m512_const1_64( 0x2A2767A4AE9B94DB );
+        sc->h4 = m512_const1_64( 0xEC06025E74DD7683 );
+        sc->h5 = m512_const1_64( 0xE7A436CDC4746251 );
+        sc->h6 = m512_const1_64( 0xC36FBAF9393AD185 );
+        sc->h7 = m512_const1_64( 0x3EEDBA1833EDFC13 );
+        sc->bcount = 0;
+        sc->ptr = 0;
+}

-/*
-static const sph_u64 IV256[] = {
-   SPH_C64(0xCCD044A12FDB3E13), SPH_C64(0xE83590301A79A9EB),
-   SPH_C64(0x55AEA0614F816E6F), SPH_C64(0x2A2767A4AE9B94DB),
-   SPH_C64(0xEC06025E74DD7683), SPH_C64(0xE7A436CDC4746251),
-   SPH_C64(0xC36FBAF9393AD185), SPH_C64(0x3EEDBA1833EDFC13)
-};
+void skein512_8way_init( skein512_8way_context *sc )
+{
+        sc->h0 = m512_const1_64( 0x4903ADFF749C51CE );
+        sc->h1 = m512_const1_64( 0x0D95DE399746DF03 );
+        sc->h2 = m512_const1_64( 0x8FD1934127C79BCE );
+        sc->h3 = m512_const1_64( 0x9A255629FF352CB1 );
+        sc->h4 = m512_const1_64( 0x5DB62599DF6CA7B0 );
+        sc->h5 = m512_const1_64( 0xEABE394CA9D5C3F4 );
+        sc->h6 = m512_const1_64( 0x991112C71A75B523 );
+        sc->h7 = m512_const1_64( 0xAE18A40B660FCC33 );
+        sc->bcount = 0;
+        sc->ptr = 0;
+}
+
+static void
+skein_big_core_8way( skein512_8way_context *sc, const void *data,
+                     size_t len )
+{
+   __m512i *vdata = (__m512i*)data;
+   __m512i *buf;
+   size_t ptr;
+   unsigned first;
+   DECL_STATE_BIG_8WAY
+
+   buf = sc->buf;
+   ptr = sc->ptr;
+   const int buf_size = 64;   // 64 * _m256i
+
+   if ( len <= buf_size - ptr )
+   {
+       memcpy_512( buf + (ptr>>3), vdata, len>>3 );
+       sc->ptr = ptr + len;
+       return;
+   }
+
+   READ_STATE_BIG( sc );
+   first = ( bcount == 0 ) << 7;
+   do {
+       size_t clen;
+
+       if ( ptr == buf_size )
+       {
+            bcount ++;
+            UBI_BIG_8WAY( 96 + first, 0 );
+            first = 0;
+            ptr = 0;
+       }
+       clen = buf_size - ptr;
+       if ( clen > len )
+            clen = len;
+       memcpy_512( buf + (ptr>>3), vdata, clen>>3 );
+       ptr += clen;
+       vdata += (clen>>3);
+       len -= clen;
+   } while ( len > 0 );
+   WRITE_STATE_BIG( sc );
+   sc->ptr = ptr;
+}
+
+static void
+skein_big_close_8way( skein512_8way_context *sc, unsigned ub, unsigned n,
+                      void *dst, size_t out_len )
+{
+   __m512i *buf;
+   size_t ptr;
+   unsigned et;
+   DECL_STATE_BIG_8WAY
+
+   buf = sc->buf;
+   ptr = sc->ptr;
+        const int buf_size = 64;
+
+   READ_STATE_BIG(sc);
+
+   memset_zero_512( buf + (ptr>>3), (buf_size - ptr) >> 3 );
+   et = 352 + ((bcount == 0) << 7);
+   UBI_BIG_8WAY( et, ptr );
+
+   memset_zero_512( buf, buf_size >> 3 );
+   bcount = 0;
+   UBI_BIG_8WAY( 510, 8 );
+
+   buf[0] = h0;
+   buf[1] = h1;
+   buf[2] = h2;
+   buf[3] = h3;
+   buf[4] = h4;
+   buf[5] = h5;
+   buf[6] = h6;
+   buf[7] = h7;
+
+   memcpy_512( dst, buf, out_len >> 3 );
+}
+
+void
+skein256_8way_update(void *cc, const void *data, size_t len)
+{
+   skein_big_core_8way(cc, data, len);
+}
+
+void
+skein256_8way_close(void *cc, void *dst)
+{
+        skein_big_close_8way(cc, 0, 0, dst, 32);
+}
+
+void
+skein512_8way_update(void *cc, const void *data, size_t len)
+{
+   skein_big_core_8way(cc, data, len);
+}
+
+void
+skein512_8way_close(void *cc, void *dst)
+{
+        skein_big_close_8way(cc, 0, 0, dst, 64);
+}
+
+#endif // AVX512

-static const sph_u64 IV512[] = {
-   SPH_C64(0x4903ADFF749C51CE), SPH_C64(0x0D95DE399746DF03),
-   SPH_C64(0x8FD1934127C79BCE), SPH_C64(0x9A255629FF352CB1),
-   SPH_C64(0x5DB62599DF6CA7B0), SPH_C64(0xEABE394CA9D5C3F4),
-   SPH_C64(0x991112C71A75B523), SPH_C64(0xAE18A40B660FCC33)
-};
-*/

 void skein256_4way_init( skein256_4way_context *sc )
 {
@@ -517,66 +767,30 @@ skein_big_close_4way( skein512_4way_context *sc, unsigned ub, unsigned n,
 	ptr = sc->ptr;
        const int buf_size = 64;

-	/*
-	 * At that point, if ptr == 0, then the message was empty;
-	 * otherwise, there is between 1 and 64 bytes (inclusive) which
-	 * are yet to be processed. Either way, we complete the buffer
-	 * to a full block with zeros (the Skein specification mandates
-	 * that an empty message is padded so that there is at least
-	 * one block to process).
-	 *
-	 * Once this block has been processed, we do it again, with
-	 * a block full of zeros, for the output (that block contains
-	 * the encoding of "0", over 8 bytes, then padded with zeros).
-	 */
-
 	READ_STATE_BIG(sc);

-        memset_zero_256( buf + (ptr>>3), (buf_size - ptr) >> 3 );
+   memset_zero_256( buf + (ptr>>3), (buf_size - ptr) >> 3 );
 	et = 352 + ((bcount == 0) << 7);
-        UBI_BIG_4WAY( et, ptr );
+   UBI_BIG_4WAY( et, ptr );

-        memset_zero_256( buf, buf_size >> 3 );
-        bcount = 0;
-        UBI_BIG_4WAY( 510, 8 );
+   memset_zero_256( buf, buf_size >> 3 );
+   bcount = 0;
+   UBI_BIG_4WAY( 510, 8 );

-        buf[0] = h0;
-        buf[1] = h1;
-        buf[2] = h2;
-        buf[3] = h3;
-        buf[4] = h4;
-        buf[5] = h5;
-        buf[6] = h6;
-        buf[7] = h7;
+   buf[0] = h0;
+   buf[1] = h1;
+   buf[2] = h2;
+   buf[3] = h3;
+   buf[4] = h4;
+   buf[5] = h5;
+   buf[6] = h6;
+   buf[7] = h7;

-        memcpy_256( dst, buf, out_len >> 3 );
+   memcpy_256( dst, buf, out_len >> 3 );
 }

-/*
-static const sph_u64 IV256[] = {
-	SPH_C64(0xCCD044A12FDB3E13), SPH_C64(0xE83590301A79A9EB),
-	SPH_C64(0x55AEA0614F816E6F), SPH_C64(0x2A2767A4AE9B94DB),
-	SPH_C64(0xEC06025E74DD7683), SPH_C64(0xE7A436CDC4746251),
-	SPH_C64(0xC36FBAF9393AD185), SPH_C64(0x3EEDBA1833EDFC13)
-};
-
-static const sph_u64 IV512[] = {
-	SPH_C64(0x4903ADFF749C51CE), SPH_C64(0x0D95DE399746DF03),
-	SPH_C64(0x8FD1934127C79BCE), SPH_C64(0x9A255629FF352CB1),
-	SPH_C64(0x5DB62599DF6CA7B0), SPH_C64(0xEABE394CA9D5C3F4),
-	SPH_C64(0x991112C71A75B523), SPH_C64(0xAE18A40B660FCC33)
-};
-*/
-/*
 void
-skein256_4way_init(void *cc)
-{
-	skein_big_init_4way(cc, IV256);
-}
-*/
-
-void
-skein256_4way(void *cc, const void *data, size_t len)
+skein256_4way_update(void *cc, const void *data, size_t len)
 {
 	skein_big_core_4way(cc, data, len);
 }
@@ -587,16 +801,8 @@ skein256_4way_close(void *cc, void *dst)
        skein_big_close_4way(cc, 0, 0, dst, 32);
 }

-/*
 void
-skein512_4way_init(void *cc)
-{
-	skein_big_init_4way(cc, IV512);
-}
-*/
-
-void
-skein512_4way(void *cc, const void *data, size_t len)
+skein512_4way_update(void *cc, const void *data, size_t len)
 {
 	skein_big_core_4way(cc, data, len);
 }
--- a/algo/skein/skein-hash-4way.h
+++ b/algo/skein/skein-hash-4way.h
@@ -55,29 +55,50 @@ extern "C"{
 #define SPH_SIZE_skein256   256
 #define SPH_SIZE_skein512   512

+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 typedef struct
 {
-   __m256i buf[8] __attribute__ ((aligned (64)));
+   __m512i buf[8];
+   __m512i h0, h1, h2, h3, h4, h5, h6, h7;
+   size_t ptr;
+   sph_u64 bcount;
+} sph_skein_8way_big_context __attribute__ ((aligned (128)));
+
+typedef sph_skein_8way_big_context skein512_8way_context;
+typedef sph_skein_8way_big_context skein256_8way_context;
+
+void skein512_8way_init( skein512_8way_context *sc );
+void skein512_8way_update( void *cc, const void *data, size_t len );
+void skein512_8way_close( void *cc, void *dst );
+
+void skein256_8way_init( skein256_8way_context *sc );
+void skein256_8way_update( void *cc, const void *data, size_t len );
+void skein256_8way_close( void *cc, void *dst );
+
+#endif // AVX512
+   
+typedef struct
+{
+   __m256i buf[8];
   __m256i h0, h1, h2, h3, h4, h5, h6, h7;
   size_t ptr;
 	sph_u64 bcount;
-} sph_skein_4way_big_context;
+} sph_skein_4way_big_context __attribute__ ((aligned (128)));

 typedef sph_skein_4way_big_context skein512_4way_context;
 typedef sph_skein_4way_big_context skein256_4way_context;

 void skein512_4way_init( skein512_4way_context *sc );
-void skein512_4way( void *cc, const void *data, size_t len );
+void skein512_4way_update( void *cc, const void *data, size_t len );
 void skein512_4way_close( void *cc, void *dst );
-//void sph_skein512_addbits_and_close(
-//        void *cc, unsigned ub, unsigned n, void *dst);
+#define skein512_4way skein512_4way_update

 void skein256_4way_init( skein256_4way_context *sc );
-void skein256_4way( void *cc, const void *data, size_t len );
+void skein256_4way_update( void *cc, const void *data, size_t len );
 void skein256_4way_close( void *cc, void *dst );
-//void sph_skein256_addbits_and_close(
-//	void *cc, unsigned ub, unsigned n, void *dst);
-
+#define skein256_4way skein256_4way_update

 #ifdef __cplusplus
 }
--- a/algo/skein/skein2-4way.c
+++ b/algo/skein/skein2-4way.c
@@ -1,9 +1,66 @@
-#include "skein2-gate.h"
+#include "skein-gate.h"
 #include <string.h>
 #include <stdint.h>
 #include "skein-hash-4way.h"

-#if defined(SKEIN2_4WAY)
+#if defined(SKEIN_8WAY)
+
+void skein2hash_8way( void *output, const void *input )
+{
+   skein512_8way_context ctx;
+   uint64_t hash[16*8] __attribute__ ((aligned (128)));
+
+   skein512_8way_init( &ctx );
+   skein512_8way_update( &ctx, input, 80 );
+   skein512_8way_close( &ctx, hash );
+
+   skein512_8way_init( &ctx );
+   skein512_8way_update( &ctx, hash, 64 );
+   skein512_8way_close( &ctx, output );
+}
+
+int scanhash_skein2_8way( struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done, struct thr_info *mythr )
+{
+    uint32_t hash[16*8] __attribute__ ((aligned (128)));
+    uint32_t vdata[20*8] __attribute__ ((aligned (64)));
+    uint32_t lane_hash[8] __attribute__ ((aligned (64)));
+    uint32_t *hash7 = &(hash[49]);
+    uint32_t *pdata = work->data;
+    uint32_t *ptarget = work->target;
+    const uint32_t Htarg = ptarget[7];
+    const uint32_t first_nonce = pdata[19];
+    uint32_t n = first_nonce;
+    __m512i  *noncev = (__m512i*)vdata + 9;   // aligned
+    int thr_id = mythr->id; 
+
+    mm512_bswap32_intrlv80_8x64( vdata, pdata );
+    do
+    {
+       *noncev = mm512_intrlv_blend_32( mm512_bswap_32(
+                _mm512_set_epi32( n+7, 0, n+6, 0, n+5, 0, n+4, 0,
+                                  n+3, 0, n+2, 0, n+1, 0, n  , 0 ) ), *noncev );
+
+       skein2hash_8way( hash, vdata );
+
+       for ( int lane = 0; lane < 8; lane++ )
+       if ( hash7[ lane<<1 ] <= Htarg )
+       {
+          extr_lane_8x64( lane_hash, hash, lane, 256 );
+          if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
+          {
+             pdata[19] = n + lane;
+             submit_lane_solution( work, lane_hash, mythr, lane );
+          }
+       }
+       n += 8;
+    } while ( (n < max_nonce-8) && !work_restart[thr_id].restart );
+
+    *hashes_done = n - first_nonce + 1;
+    return 0;
+}
+
+#elif defined(SKEIN_4WAY)

 void skein2hash_4way( void *output, const void *input )
 {
--- a/algo/skein/skein2-gate.c
+++ b/algo/skein/skein2-gate.c
@@ -1,17 +0,0 @@
-#include "skein2-gate.h"
-#include <stdint.h>
-#include "sph_skein.h"
-
-bool register_skein2_algo( algo_gate_t* gate )
-{
-  gate->optimizations = AVX2_OPT;
-#if defined (SKEIN2_4WAY)
-  gate->scanhash  = (void*)&scanhash_skein2_4way;
-  gate->hash      = (void*)&skein2hash_4way;
-#else
-  gate->scanhash  = (void*)&scanhash_skein2;
-  gate->hash      = (void*)&skein2hash;
-#endif
-  return true;
-};
-
--- a/algo/skein/skein2-gate.h
+++ b/algo/skein/skein2-gate.h
@@ -1,20 +0,0 @@
-#ifndef __SKEIN2GATE_H__
-#define __SKEIN2_GATE_H__
-#include "algo-gate-api.h"
-#include <stdint.h>
-
-#if defined(__AVX2__)
-  #define SKEIN2_4WAY
-#endif
-
-#if defined(SKEIN2_4WAY)
-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 );
-#endif
-
-void skein2hash( void *output, const void *input );
-int scanhash_skein2( struct work *work, uint32_t max_nonce,
-                     uint64_t *hashes_done, struct thr_info *mythr );
-#endif
-
--- a/algo/skein/skein2.c
+++ b/algo/skein/skein2.c
@@ -1,4 +1,4 @@
-#include "algo-gate-api.h"
+#include "skein-gate.h"
 #include <string.h>
 #include <stdint.h>

--- a/algo/swifftx/Swifftx_sha3.cpp
+++ b/algo/swifftx/Swifftx_sha3.cpp
@@ -0,0 +1,369 @@
+#include "Swifftx_sha3.h"
+extern "C" {
+#include "SWIFFTX.h"
+}
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+// The default salt value.
+// This is the expansion of e (Euler's number) - the 19 digits after 2.71:
+// 8281828459045235360.
+// The above in base 256, from MSB to LSB:
+BitSequence SWIF_saltValueChar[SWIF_HAIFA_SALT_SIZE] = {114, 238, 247, 26, 192, 28, 170, 160};
+
+// All the IVs here below were produced from the decimal digits of e's expansion.
+// The code can be found in 'ProduceRandomIV.c'.
+// The initial value for 224 digest size.
+const BitSequence SWIF_HAIFA_IV_224[SWIFFTX_OUTPUT_BLOCK_SIZE] = 
+{37, 242, 132,   2, 167,  81, 158, 237, 113,  77, 162,  60,  65, 236, 108, 246,
+101,  72, 190, 109,  58, 205,  99,   6, 114, 169, 104, 114,  38, 146, 121, 142,
+ 59,  98, 233,  84,  72, 227,  22, 199,  17, 102, 198, 145,  24, 178,  37,   1,
+215, 245,  66, 120, 230, 193, 113, 253, 165, 218,  66, 134,  49, 231, 124, 204,
+  0};
+
+// The initial value for 256 digest size.
+const BitSequence SWIF_HAIFA_IV_256[SWIFFTX_OUTPUT_BLOCK_SIZE] = 
+{250,  50,  42,  40,  14, 233,  53,  48, 227,  42, 237, 187, 211, 120, 209, 234,
+  27, 144,   4,  61, 243, 244,  29, 247,  37, 162,  70,  11, 231, 196,  53,   6,
+ 193, 240,  94, 126, 204, 132, 104,  46, 114,  29,   3, 104, 118, 184, 201,   3,
+  57,  77,  91, 101,  31, 155,  84, 199, 228,  39, 198,  42, 248, 198, 201, 178,
+   8};
+
+// The initial value for 384 digest size.
+const BitSequence SWIF_HAIFA_IV_384[SWIFFTX_OUTPUT_BLOCK_SIZE] = 
+{40, 145, 193, 100, 205, 171,  47,  76, 254,  10, 196,  41, 165, 207, 200,  79,
+109,  13,  75, 201,  17, 172,  64, 162, 217,  22,  88,  39,  51,  30, 220, 151,
+133,  73, 216, 233, 184, 203,  77,   0, 248,  13,  28, 199,  30, 147, 232, 242,
+227, 124, 169, 174,  14,  45,  27,  87, 254,  73,  68, 136, 135, 159,  83, 152,
+  0};
+
+// The initial value for 512 digest size.
+const BitSequence SWIF_HAIFA_IV_512[SWIFFTX_OUTPUT_BLOCK_SIZE] = 
+{195, 126, 197, 167, 157, 114,  99, 126, 208, 105, 200,  90,  71, 195, 144, 138,
+ 142, 122, 123, 116,  24, 214, 168, 173, 203, 183, 194, 210, 102, 117, 138,  42,
+ 114, 118, 132,  33,  35, 149, 143, 163, 163, 183, 243, 175,  72,  22, 201, 255,
+ 102, 243,  22, 187, 211, 167, 239,  76, 164,  70,  80, 182, 181, 212,   9, 185,
+   0};
+
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// NIST API implementation portion.
+///////////////////////////////////////////////////////////////////////////////////////////////
+
+int Swifftx::Init(int hashbitlen)
+{
+	switch(hashbitlen)
+	{
+	case 224:
+		swifftxState.hashbitlen = hashbitlen;
+		// Initializes h_0 in HAIFA:
+		memcpy(swifftxState.currOutputBlock, SWIF_HAIFA_IV_224, SWIFFTX_OUTPUT_BLOCK_SIZE);
+		break;
+	case 256:
+		swifftxState.hashbitlen = hashbitlen;
+		memcpy(swifftxState.currOutputBlock, SWIF_HAIFA_IV_256, SWIFFTX_OUTPUT_BLOCK_SIZE);
+		break;
+	case 384:
+		swifftxState.hashbitlen = hashbitlen;
+		memcpy(swifftxState.currOutputBlock, SWIF_HAIFA_IV_384, SWIFFTX_OUTPUT_BLOCK_SIZE);
+		break;
+	case 512:
+		swifftxState.hashbitlen = hashbitlen;
+		memcpy(swifftxState.currOutputBlock, SWIF_HAIFA_IV_512, SWIFFTX_OUTPUT_BLOCK_SIZE);
+		break;
+	default:
+		return BAD_HASHBITLEN;
+	}
+	
+	swifftxState.wasUpdated = false;
+	swifftxState.remainingSize = 0;
+	memset(swifftxState.remaining, 0, SWIF_HAIFA_INPUT_BLOCK_SIZE);
+	memset(swifftxState.numOfBitsChar, 0, SWIF_HAIFA_NUM_OF_BITS_SIZE);
+	// Initialize the salt with the default value.
+	memcpy(swifftxState.salt, SWIF_saltValueChar, SWIF_HAIFA_SALT_SIZE);
+
+	InitializeSWIFFTX();
+
+	return SUCCESS;
+}
+
+int Swifftx::Update(const BitSequence *data, DataLength databitlen)
+{
+	// The size of input in bytes after putting the remaining data from previous invocation.
+	int sizeOfInputAfterRemaining = 0;
+	// The input block to compression function of SWIFFTX:
+	BitSequence currInputBlock[SWIFFTX_INPUT_BLOCK_SIZE] = {0};
+	// Whether we handled a single block.
+	bool wasSingleBlockHandled = false;
+
+	swifftxState.wasUpdated = true;
+
+	// Handle an empty message as required by NIST. Since 'Final()' is oblivious to the input
+	// (but of course uses the output of the compression function from the previous round, 
+	// which is called h_{i-1} in HAIFA article), we have to do nothing here.
+	if (databitlen == 0)
+		return SUCCESS;
+
+    // If we had before an input with unaligned length, return an error
+    if (swifftxState.remainingSize % 8)
+	{
+    	return INPUT_DATA_NOT_ALIGNED;
+    }
+
+    // Convert remaining size to bytes.
+    swifftxState.remainingSize /= 8;
+
+	// As long as we have enough data combined from (remaining + data) to fill input block
+	//NASTAVENIE RUND
+	while (((databitlen / 8) + swifftxState.remainingSize) >= SWIF_HAIFA_INPUT_BLOCK_SIZE)
+	{
+		// Fill the input block with data:
+		// 1. The output of the previous block:
+		memcpy(currInputBlock, swifftxState.currOutputBlock, SWIFFTX_OUTPUT_BLOCK_SIZE);
+		// 2. The input part of the block:
+		// 2a. The remaining data from the previous 'Update()' call:
+		if (swifftxState.remainingSize)
+			memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE, swifftxState.remaining, 
+				   swifftxState.remainingSize);
+		// 2b. The input data that we have place for after the 'remaining':
+		sizeOfInputAfterRemaining = SWIFFTX_INPUT_BLOCK_SIZE - SWIFFTX_OUTPUT_BLOCK_SIZE 
+								  - ((int) swifftxState.remainingSize) - SWIF_HAIFA_NUM_OF_BITS_SIZE 
+								  - SWIF_HAIFA_SALT_SIZE;
+		memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + swifftxState.remainingSize, 
+			   data, sizeOfInputAfterRemaining);
+
+		// 3. The #bits part of the block:
+		memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + swifftxState.remainingSize 
+			 + sizeOfInputAfterRemaining,
+			   swifftxState.numOfBitsChar, SWIF_HAIFA_NUM_OF_BITS_SIZE);
+		// 4. The salt part of the block:
+		memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + swifftxState.remainingSize 
+			 + sizeOfInputAfterRemaining + SWIF_HAIFA_NUM_OF_BITS_SIZE,
+			   swifftxState.salt, SWIF_HAIFA_SALT_SIZE);
+
+		ComputeSingleSWIFFTX(currInputBlock, swifftxState.currOutputBlock, false);
+
+		// Update the #bits field with SWIF_HAIFA_INPUT_BLOCK_SIZE.
+		AddToCurrInBase256(swifftxState.numOfBitsChar, SWIF_HAIFA_INPUT_BLOCK_SIZE * 8);
+		wasSingleBlockHandled = true;
+		data += sizeOfInputAfterRemaining;
+		databitlen -= (sizeOfInputAfterRemaining * 8);
+   		swifftxState.remainingSize = 0;
+	}
+
+	// Update the swifftxState.remaining and swifftxState.remainingSize.
+    // remainingSize will be in bits after exiting 'Update()'.
+	if (wasSingleBlockHandled)
+	{		
+		swifftxState.remainingSize = (unsigned int) databitlen; // now remaining size is in bits.
+        if (swifftxState.remainingSize)
+			memcpy(swifftxState.remaining, data, (swifftxState.remainingSize + 7) / 8);
+	}
+	else
+	{
+		memcpy(swifftxState.remaining + swifftxState.remainingSize, data, 
+			   (size_t) (databitlen + 7) / 8);
+		swifftxState.remainingSize = (swifftxState.remainingSize * 8) + (unsigned short) databitlen;
+	}
+
+	return SUCCESS;
+}
+
+int Swifftx::Final(BitSequence *hashval)
+{
+    int i;
+    // Whether to add one last block. True if the padding appended to the last block overflows
+	// the block size.
+    bool toAddFinalBlock = false;
+    bool toPutOneInFinalBlock = false;
+    unsigned short oneShift = 0;
+   	// The size of the last input block before the zeroes padding. We add 1 here because we
+    // include the final '1' bit in the calculation and 7 as we round the length to bytes.
+	unsigned short sizeOfLastInputBlock = (swifftxState.remainingSize + 1 + 7) / 8;
+    // The number of bytes of zero in the padding part.
+	// The padding contains:
+	// 1. A single 1 bit.
+	// 2. As many zeroes as needed.
+	// 3. The message length in bits. Occupies SWIF_HAIFA_NUM_OF_BITS_SIZE bytes.
+	// 4. The digest size. Maximum is 512, so we need 2 bytes.
+	// If the total number achieved is negative, add an additional block, as HAIFA specifies.
+	short numOfZeroBytesInPadding = (short) SWIFFTX_INPUT_BLOCK_SIZE - SWIFFTX_OUTPUT_BLOCK_SIZE 
+								  - sizeOfLastInputBlock - (2 * SWIF_HAIFA_NUM_OF_BITS_SIZE) - 2 
+								  - SWIF_HAIFA_SALT_SIZE;
+   	// The input block to compression function of SWIFFTX:
+	BitSequence currInputBlock[SWIFFTX_INPUT_BLOCK_SIZE] = {0};
+	// The message length in base 256.
+	BitSequence messageLengthChar[SWIF_HAIFA_NUM_OF_BITS_SIZE] = {0};
+   	// The digest size used for padding:
+	unsigned char digestSizeLSB = swifftxState.hashbitlen % 256;
+	unsigned char digestSizeMSB = (swifftxState.hashbitlen - digestSizeLSB) / 256;
+
+	if (numOfZeroBytesInPadding < 1)
+		toAddFinalBlock = true;
+
+	// Fill the input block with data:
+	// 1. The output of the previous block:
+	memcpy(currInputBlock, swifftxState.currOutputBlock, SWIFFTX_OUTPUT_BLOCK_SIZE);
+	// 2a. The input part of the block, which is the remaining data from the previous 'Update()'
+    //     call, if exists and an extra '1' bit (maybe all we have is this extra 1):
+
+    // Add the last 1 in big-endian convention ...
+    if (swifftxState.remainingSize % 8 == 0)
+	{
+       swifftxState.remaining[sizeOfLastInputBlock - 1] = 0x80;
+    }
+    else 
+	{
+       swifftxState.remaining[sizeOfLastInputBlock - 1] |= (1 << (7 - (swifftxState.remainingSize % 8)));
+    }
+
+	if (sizeOfLastInputBlock)
+		memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE, swifftxState.remaining, 
+			   sizeOfLastInputBlock);
+    
+   	// Compute the message length in base 256:
+	for (i = 0; i < SWIF_HAIFA_NUM_OF_BITS_SIZE; ++i)
+        messageLengthChar[i] = swifftxState.numOfBitsChar[i];
+    if (sizeOfLastInputBlock)
+		AddToCurrInBase256(messageLengthChar, sizeOfLastInputBlock * 8);
+
+	if (!toAddFinalBlock)
+	{
+		// 2b. Put the zeroes:
+		memset(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + sizeOfLastInputBlock,
+			   0, numOfZeroBytesInPadding);
+		// 2c. Pad the message length:
+		for (i = 0; i < SWIF_HAIFA_NUM_OF_BITS_SIZE; ++i)
+			currInputBlock[SWIFFTX_OUTPUT_BLOCK_SIZE + sizeOfLastInputBlock 
+						 + numOfZeroBytesInPadding + i] = messageLengthChar[i];
+		// 2d. Pad the digest size:
+		currInputBlock[SWIFFTX_OUTPUT_BLOCK_SIZE + sizeOfLastInputBlock 
+					 + numOfZeroBytesInPadding + SWIF_HAIFA_NUM_OF_BITS_SIZE] = digestSizeMSB;
+		currInputBlock[SWIFFTX_OUTPUT_BLOCK_SIZE + sizeOfLastInputBlock 
+					 + numOfZeroBytesInPadding + SWIF_HAIFA_NUM_OF_BITS_SIZE + 1] = digestSizeLSB;
+	}
+	else
+	{
+		// 2b. Put the zeroes, if at all:
+		if ((SWIF_HAIFA_INPUT_BLOCK_SIZE - sizeOfLastInputBlock) > 0)
+		{
+			 memset(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + sizeOfLastInputBlock,
+					0, SWIF_HAIFA_INPUT_BLOCK_SIZE - sizeOfLastInputBlock);
+		}
+	}
+
+   	// 3. The #bits part of the block: 
+	memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + SWIF_HAIFA_INPUT_BLOCK_SIZE, 
+           swifftxState.numOfBitsChar, SWIF_HAIFA_NUM_OF_BITS_SIZE);
+	// 4. The salt part of the block:
+	memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + SWIF_HAIFA_INPUT_BLOCK_SIZE 
+		 + SWIF_HAIFA_NUM_OF_BITS_SIZE, 
+           swifftxState.salt, 
+		   SWIF_HAIFA_SALT_SIZE);
+
+    ComputeSingleSWIFFTX(currInputBlock, swifftxState.currOutputBlock, !toAddFinalBlock); 
+
+	// If we have to add one more block, it is now:
+	if (toAddFinalBlock)
+	{
+		// 1. The previous output block, as usual.
+		memcpy(currInputBlock, swifftxState.currOutputBlock, SWIFFTX_OUTPUT_BLOCK_SIZE);
+
+		// 2a. Instead of the input, zeroes:
+		memset(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE , 0, 
+			   SWIF_HAIFA_INPUT_BLOCK_SIZE - SWIF_HAIFA_NUM_OF_BITS_SIZE - 2);
+		// 2b. Instead of the input, the message length:
+		memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + SWIF_HAIFA_INPUT_BLOCK_SIZE 
+			 - SWIF_HAIFA_NUM_OF_BITS_SIZE - 2,
+			   messageLengthChar,
+			   SWIF_HAIFA_NUM_OF_BITS_SIZE);
+		// 2c. Instead of the input, the digest size:
+		currInputBlock[SWIFFTX_OUTPUT_BLOCK_SIZE + SWIF_HAIFA_INPUT_BLOCK_SIZE - 2] = digestSizeMSB;
+		currInputBlock[SWIFFTX_OUTPUT_BLOCK_SIZE + SWIF_HAIFA_INPUT_BLOCK_SIZE - 1] = digestSizeLSB;
+		// 3. The #bits part of the block, which is zero in case of additional block:
+		memset(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + SWIF_HAIFA_INPUT_BLOCK_SIZE,
+			   0,
+			   SWIF_HAIFA_NUM_OF_BITS_SIZE);
+		// 4. The salt part of the block:
+		memcpy(currInputBlock + SWIFFTX_OUTPUT_BLOCK_SIZE + SWIF_HAIFA_INPUT_BLOCK_SIZE 
+			 + SWIF_HAIFA_NUM_OF_BITS_SIZE, 
+               swifftxState.salt, 
+			   SWIF_HAIFA_SALT_SIZE);
+
+        ComputeSingleSWIFFTX(currInputBlock, swifftxState.currOutputBlock, true); 
+	}
+
+	// Finally, copy the result into 'hashval'. In case the digest size is not 512bit, copy the
+	// first hashbitlen of them:
+    for (i = 0; i < (swifftxState.hashbitlen / 8); ++i)
+		hashval[i] = swifftxState.currOutputBlock[i];
+
+	return SUCCESS;
+}
+
+int Swifftx::Hash(int hashbitlen, const BitSequence *data, DataLength databitlen, 
+				BitSequence *hashval)
+{
+	int result;
+	//hashState state;
+   	// The pointer to the current place in the input we take into the compression function.
+	DataLength currInputIndex = 0;
+
+    result = Swifftx::Init(hashbitlen);
+
+	if (result != SUCCESS)
+		return result;
+
+	for ( ; (databitlen / 8) >  SWIF_HAIFA_INPUT_BLOCK_SIZE; 
+         currInputIndex += SWIF_HAIFA_INPUT_BLOCK_SIZE, databitlen -= (SWIF_HAIFA_INPUT_BLOCK_SIZE * 8))
+	{
+		result = Swifftx::Update(data + currInputIndex, SWIF_HAIFA_INPUT_BLOCK_SIZE * 8); 
+		if (result != SUCCESS)
+			return result;
+	}
+
+	// The length of the last block may be shorter than (SWIF_HAIFA_INPUT_BLOCK_SIZE * 8)
+	result = Swifftx::Update(data + currInputIndex, databitlen); 
+	if (result != SUCCESS)
+	{
+		return result;
+	}
+
+    return Swifftx::Final(hashval);
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// Helper fuction implementation portion.
+///////////////////////////////////////////////////////////////////////////////////////////////
+
+void Swifftx::AddToCurrInBase256(BitSequence value[SWIF_HAIFA_NUM_OF_BITS_SIZE], 
+							   unsigned short toAdd)
+{
+	unsigned char remainder = 0;
+	short i;
+	BitSequence currValueInBase256[8] = {0};
+	unsigned short currIndex = 7;
+	unsigned short temp = 0;
+
+	do
+	{
+		remainder = toAdd % 256;
+		currValueInBase256[currIndex--] = remainder;
+		toAdd -= remainder;
+		toAdd /= 256;
+	}
+	while(toAdd != 0);
+
+	for (i = 7; i >= 0; --i)
+	{
+		temp = value[i] + currValueInBase256[i];
+		if (temp > 255)
+		{
+			value[i] = temp % 256;
+			currValueInBase256[i - 1]++;
+		}
+		else
+			value[i] = (unsigned char) temp;
+	}
+}
--- a/algo/swifftx/Swifftx_sha3.h
+++ b/algo/swifftx/Swifftx_sha3.h
@@ -0,0 +1,79 @@
+#ifndef SWIFFTX_SHA3_H
+#define SWIFFTX_SHA3_H
+
+#include "sha3_interface.h"
+#include "stdbool.h"
+#include "stdint.h"
+
+class Swifftx : public SHA3 {
+
+#define SWIFFTX_INPUT_BLOCK_SIZE 256
+#define SWIFFTX_OUTPUT_BLOCK_SIZE 65
+#define SWIF_HAIFA_SALT_SIZE 8
+#define SWIF_HAIFA_NUM_OF_BITS_SIZE 8
+#define SWIF_HAIFA_INPUT_BLOCK_SIZE (SWIFFTX_INPUT_BLOCK_SIZE - SWIFFTX_OUTPUT_BLOCK_SIZE \
+							  - SWIF_HAIFA_NUM_OF_BITS_SIZE - SWIF_HAIFA_SALT_SIZE)
+
+	typedef unsigned char BitSequence;
+//const DataLength SWIF_SALT_VALUE;
+
+#define SWIF_HAIFA_IV 0
+
+/*const BitSequence SWIF_HAIFA_IV_224[SWIFFTX_OUTPUT_BLOCK_SIZE];
+const BitSequence SWIF_HAIFA_IV_256[SWIFFTX_OUTPUT_BLOCK_SIZE];
+const BitSequence SWIF_HAIFA_IV_384[SWIFFTX_OUTPUT_BLOCK_SIZE];
+const BitSequence SWIF_HAIFA_IV_512[SWIFFTX_OUTPUT_BLOCK_SIZE];*/
+
+typedef enum 
+{ 
+	SUCCESS = 0,
+	FAIL = 1,
+	BAD_HASHBITLEN = 2,
+	BAD_SALT_SIZE = 3,
+	SET_SALT_VALUE_FAILED = 4,
+	INPUT_DATA_NOT_ALIGNED = 5
+} HashReturn;
+
+typedef struct hashState {
+	unsigned short hashbitlen;
+
+	// The data remained after the recent call to 'Update()'. 
+	BitSequence remaining[SWIF_HAIFA_INPUT_BLOCK_SIZE + 1];
+
+	// The size of the remaining data in bits.
+	// Is 0 in case there is no remaning data at all.
+	unsigned int remainingSize;
+
+	// The current output of the compression function. At the end will contain the final digest
+	// (which may be needed to be truncated, depending on hashbitlen).
+	BitSequence currOutputBlock[SWIFFTX_OUTPUT_BLOCK_SIZE];
+
+	// The value of '#bits hashed so far' field in HAIFA, in base 256.
+	BitSequence numOfBitsChar[SWIF_HAIFA_NUM_OF_BITS_SIZE];
+
+	// The salt value currently in use:
+	BitSequence salt[SWIF_HAIFA_SALT_SIZE];
+
+	// Indicates whether a single 'Update()' occured. 
+	// Ater a call to 'Update()' the key and the salt values cannot be changed.
+	bool wasUpdated;
+} hashState;
+
+private:
+int swifftxNumRounds;
+hashState swifftxState;
+
+
+public:
+int Init(int hashbitlen);
+int Update(const BitSequence *data, DataLength databitlen);
+int Final(BitSequence *hashval);
+int Hash(int hashbitlen, const BitSequence *data, DataLength databitlen, 
+				BitSequence *hashval);
+
+private:
+static void AddToCurrInBase256(BitSequence value[SWIF_HAIFA_NUM_OF_BITS_SIZE], unsigned short toAdd);
+
+};
+
+#endif
--- a/algo/swifftx/hash_interface.h
+++ b/algo/swifftx/hash_interface.h
@@ -0,0 +1,21 @@
+#pragma once
+
+#include <cstdint>
+
+namespace hash {
+
+using BitSequence = unsigned char;
+using DataLength = unsigned long long;
+
+struct hash_interface {
+    virtual ~hash_interface() = default;
+
+    virtual int Init(int hash_bitsize) = 0;
+    virtual int Update(const BitSequence *data, DataLength data_bitsize) = 0;
+    virtual int Final(BitSequence *hash) = 0;
+
+    virtual int
+    Hash(int hash_bitsize, const BitSequence *data, DataLength data_bitsize, BitSequence *hash) = 0;
+};
+
+} // namespace hash
--- a/algo/swifftx/inttypes.h
+++ b/algo/swifftx/inttypes.h
@@ -0,0 +1,39 @@
+ /*
+  inttypes.h
+
+  Contributors:
+    Created by Marek Michalkiewicz <marekm@linux.org.pl>
+
+  THIS SOFTWARE IS NOT COPYRIGHTED
+
+  This source code is offered for use in the public domain.  You may
+ use, modify or distribute it freely.
+
+ This code is distributed in the hope that it will be useful, but
+ WITHOUT ANY WARRANTY.  ALL WARRANTIES, EXPRESS OR IMPLIED ARE HEREBY
+ DISCLAIMED.  This includes but is not limited to warranties of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+*/
+
+ #ifndef __INTTYPES_H_
+ #define __INTTYPES_H_
+
+ /* Use [u]intN_t if you need exactly N bits.
+  XXX - doesn't handle the -mint8 option.  */
+
+ typedef signed char swift_int8_t;
+ typedef unsigned char swift_uint8_t;
+
+ typedef int swift_int16_t;
+ typedef unsigned int swift_uint16_t;
+
+ typedef long swift_int32_t;
+ typedef unsigned long swift_uint32_t;
+
+ typedef long long swift_int64_t;
+ typedef unsigned long long swift_uint64_t;
+
+ //typedef swift_int16_t intptr_t;
+ //typedef swift_uint16_t uintptr_t;
+
+ #endif
--- a/algo/swifftx/sha3_interface.h
+++ b/algo/swifftx/sha3_interface.h
@@ -0,0 +1,14 @@
+#pragma once
+
+#include <cstdint>
+//#include <streams/hash/hash_interface.h>
+#include "hash_interface.h"
+
+namespace sha3 {
+
+using BitSequence = hash::BitSequence;
+using DataLength = hash::DataLength;
+
+struct sha3_interface : hash::hash_interface {};
+
+} // namespace sha3
--- a/algo/swifftx/stdbool.h
+++ b/algo/swifftx/stdbool.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2000 Jeroen Ruigrok van der Werven <asmodai@FreeBSD.org>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * $FreeBSD: src/include/stdbool.h,v 1.6 2002/08/16 07:33:14 alfred Exp $
+ */
+
+#ifndef _STDBOOL_H_
+#define _STDBOOL_H_     
+
+#define __bool_true_false_are_defined   1
+
+#ifndef __cplusplus
+
+#define false   0
+#define true    1
+
+//#define bool    _Bool
+//#if __STDC_VERSION__ < 199901L && __GNUC__ < 3
+//typedef int     _Bool;
+//#endif
+typedef int bool;
+
+#endif /* !__cplusplus */
+
+#endif /* !_STDBOOL_H_ */
--- a/algo/swifftx/stdint.h
+++ b/algo/swifftx/stdint.h
@@ -0,0 +1,54 @@
+#ifndef _SWIFFT_STDINT_H
+#define _SWIFFT_STDINT_H
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+//
+// A note from SWIFFTX implementers:
+//
+// Although the submission was targeted for Microsoft Visual Studio 2005 compiler, we strived
+// to make the code as portable as possible. This is why we preferred to use the types defined
+// here, instead of Microsoft-specific types. We compiled the code with gcc to make this sure.
+// However, we couldn't use this header as is, due to VS2005 compiler objections. This is why
+// we commented out certain defines and clearly marked it.
+// To compile our code on gcc you may define SYS_STDINT.
+//
+///////////////////////////////////////////////////////////////////////////////////////////////
+
+#ifdef SYS_STDINT
+
+#include <stdint.h>
+
+#else
+
+#include "inttypes.h"
+// The following was commented out by SWIFFTX implementers:
+// __BEGIN_DECLS
+
+typedef swift_int8_t swifftx_int_least8_t;
+typedef swift_int16_t swifftx_int_least16_t;
+typedef swift_int32_t swifftx_int_least32_t;
+typedef swift_uint8_t swifftx_uint_least8_t;
+typedef swift_uint16_t swifftx_uint_least16_t;
+typedef swift_uint32_t swifftx_uint_least32_t;
+
+#ifndef __STRICT_ANSI__
+typedef swift_int64_t swifftx_int_least64_t;
+typedef swift_uint64_t swifftx_uint_least64_t;
+#endif
+
+/*typedef signed char int_fast8_t;
+typedef signed long int int_fast16_t;
+typedef signed long int int_fast32_t;
+typedef signed long long int int_fast64_t;
+
+typedef unsigned char uint_fast8_t;
+typedef unsigned long int uint_fast16_t;
+typedef unsigned long int uint_fast32_t;
+typedef unsigned long long int uint_fast64_t;*/
+
+// The following was commented out by SWIFFTX implementers:
+// #include <endian.h>
+// __END_DECLS
+#endif
+
+#endif
--- a/algo/swifftx/swifftx-4way.c
+++ b/algo/swifftx/swifftx-4way.c
@@ -0,0 +1,912 @@
+///////////////////////////////////////////////////////////////////////////////////////////////
+//
+//  SWIFFTX ANSI C OPTIMIZED 32BIT IMPLEMENTATION FOR NIST SHA-3 COMPETITION
+//
+//  SWIFFTX.c
+//
+//  October 2008
+//
+//  This is the source file of the OPTIMIZED 32BIT implementation of SWIFFTX hash function.
+//  SWIFFTX is a candidate function for SHA-3 NIST competition.
+//  More details about SWIFFTX can be found in the accompanying submission documents.
+//
+///////////////////////////////////////////////////////////////////////////////////////////////
+#include "swifftx.h"
+// See the remarks concerning compatibility issues inside stdint.h.
+#include "stdint.h"
+// Remove this while using gcc:
+//#include "stdbool.h"
+#include <memory.h>
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// Constants and static tables portion.
+///////////////////////////////////////////////////////////////////////////////////////////////
+
+// In SWIFFTX we work over Z_257, so this is the modulus and the arithmetic is performed modulo
+// this number.
+#define FIELD_SIZE 257
+
+// The size of FFT we use:
+#define N 64
+
+#define LOGN 6
+
+#define EIGHTH_N (N / 8)
+
+// The number of FFTS done on the input.
+#define M (SWIFFTX_INPUT_BLOCK_SIZE / 8)   // 32
+
+// Omega is the 128th root of unity in Z_257.
+// We choose w = 42.
+#define OMEGA 42
+
+// The size of the inner FFT lookup table:
+#define W 8
+
+// Calculates the sum and the difference of two numbers.
+//
+// Parameters:
+// - A: the first operand. After the operation stores the sum of the two operands.
+// - B: the second operand. After the operation stores the difference between the first and the
+//   second operands.
+#define ADD_SUB_4WAY( A, B ) \
+{ \
+  __m128i temp = B; \
+  B = _mm_sub_epi32( A, B ); \
+  A = _mm_add_epi32( A, temp ); \
+}
+
+
+//#define ADD_SUB(A, B) {register int temp = (B); B = ((A) - (B)); A = ((A) + (temp));}
+
+// Quickly reduces an integer modulo 257.
+//
+// Parameters:
+// - A: the input.
+
+#define Q_REDUCE( A ) ( _mm_sub_epi32( \
+                               _mm_and_epi32( A, m128_const1_32( 0xff ) ), \
+                               _mm_srli_epi32( A, 8 ) ) )
+
+//#define Q_REDUCE(A) (((A) & 0xff) - ((A) >> 8))
+
+// Since we need to do the setup only once, this is the indicator variable:
+static bool wasSetupDone = false;
+
+// This array stores the powers of omegas that correspond to the indices, which are the input
+// values. Known also as the "outer FFT twiddle factors".
+swift_int16_t multipliers[N];
+
+// This array stores the powers of omegas, multiplied by the corresponding values.
+// We store this table to save computation time.
+//
+// To calculate the intermediate value of the compression function (the first out of two
+// stages), we multiply the k-th bit of x_i by w^[(2i + 1) * k]. {x_i} is the input to the
+// compression function, i is between 0 and 31, x_i is a 64-bit value.
+// One can see the formula for this (intermediate) stage in the SWIFFT FSE 2008 paper --
+// formula (2), section 3, page 6.
+swift_int16_t fftTable[256 * EIGHTH_N];
+
+// The A's we use in SWIFFTX shall be random elements of Z_257.
+// We generated these A's from the decimal expansion of PI as follows:  we converted each
+// triple of digits into a decimal number d. If d < (257 * 3) we used (d % 257) for the next A
+// element, otherwise move to the next triple of digits in the expansion. This guarntees that
+// the A's are random, provided that PI digits are.
+const swift_int16_t As[3 * M * N] =
+{141,  78, 139,  75, 238, 205, 129, 126,  22, 245, 197, 169, 142, 118, 105,  78,
+  50, 149,  29, 208, 114,  34,  85, 117,  67, 148,  86, 256,  25,  49, 133,  93,
+  95,  36,  68, 231, 211, 102, 151, 128, 224, 117, 193,  27, 102, 187,   7, 105,
+  45, 130, 108, 124, 171, 151, 189, 128, 218, 134, 233, 165,  14, 201, 145, 134,
+  52, 203,  91,  96, 197,  69, 134, 213, 136,  93,   3, 249, 141,  16, 210,  73,
+   6,  92,  58,  74, 174,   6, 254,  91, 201, 107, 110,  76, 103,  11,  73,  16,
+  34, 209,   7, 127, 146, 254,  95, 176,  57,  13, 108, 245,  77,  92, 186, 117,
+ 124,  97, 105, 118,  34,  74, 205, 122, 235,  53,  94, 238, 210, 227, 183,  11,
+ 129, 159, 105, 183, 142, 129,  86,  21, 137, 138, 224, 223, 190, 188, 179, 188,
+ 256,  25, 217, 176,  36, 176, 238, 127, 160, 210, 155, 148, 132,   0,  54, 127,
+ 145,   6,  46,  85, 243,  95, 173, 123, 178, 207, 211, 183, 224, 173, 146,  35,
+  71, 114,  50,  22, 175,   1,  28,  19, 112, 129,  21,  34, 161, 159, 115,  52,
+   4, 193, 211,  92, 115,  49,  59, 217, 218,  96,  61,  81,  24, 202, 198,  89,
+  45, 128,   8,  51, 253,  87, 171,  35,   4, 188, 171,  10,   3, 137, 238,  73,
+  19, 208, 124, 163, 103, 177, 155, 147,  46,  84, 253, 233, 171, 241, 211, 217,
+ 159,  48,  96,  79, 237,  18, 171, 226,  99,   1,  97, 195, 216, 163, 198,  95,
+   0, 201,  65, 228,  21, 153, 124, 230,  44,  35,  44, 108,  85, 156, 249, 207,
+  26, 222, 131,   1,  60, 242, 197, 150, 181,  19, 116, 213,  75,  98, 124, 240,
+ 123, 207,  62, 255,  60, 143, 187, 157, 139,   9,  12, 104,  89,  49, 193, 146,
+ 104, 196, 181,  82, 198, 253, 192, 191, 255, 122, 212, 104,  47,  20, 132, 208,
+  46, 170,   2,  69, 234,  36,  56, 163,  28, 152, 104, 238, 162,  56,  24,  58,
+  38, 150, 193, 254, 253, 125, 173,  35,  73, 126, 247, 239, 216,   6, 199,  15,
+  90,  12,  97, 122,   9,  84, 207, 127, 219,  72,  58,  30,  29, 182,  41, 192,
+ 235, 248, 237,  74,  72, 176, 210, 252,  45,  64, 165,  87, 202, 241, 236, 223,
+ 151, 242, 119, 239,  52, 112, 169,  28,  13,  37, 160,  60, 158,  81, 133,  60,
+  16, 145, 249, 192, 173, 217, 214,  93, 141, 184,  54,  34, 161, 104, 157,  95,
+  38, 133, 218, 227, 211, 181,   9,  66, 137, 143,  77,  33, 248, 159,   4,  55,
+ 228,  48,  99, 219, 222, 184,  15,  36, 254, 256, 157, 237,  87, 139, 209, 113,
+ 232,  85, 126, 167, 197, 100, 103, 166,  64, 225, 125, 205, 117, 135,  84, 128,
+ 231, 112,  90, 241,  28,  22, 210, 147, 186,  49, 230,  21, 108,  39, 194,  47,
+ 123, 199, 107, 114,  30, 210, 250, 143,  59, 156, 131, 133, 221,  27,  76,  99,
+ 208, 250,  78,  12, 211, 141,  95,  81, 195, 106,   8, 232, 150, 212, 205, 221,
+  11, 225,  87, 219, 126, 136, 137, 180, 198,  48,  68, 203, 239, 252, 194, 235,
+ 142, 137, 174, 172, 190, 145, 250, 221, 182, 204,   1, 195, 130, 153,  83, 241,
+ 161, 239, 211, 138,  11, 169, 155, 245, 174,  49,  10, 166,  16, 130, 181, 139,
+ 222, 222, 112,  99, 124,  94,  51, 243, 133, 194, 244, 136,  35, 248, 201, 177,
+ 178, 186, 129, 102,  89, 184, 180,  41, 149,  96, 165,  72, 225, 231, 134, 158,
+ 199,  28, 249,  16, 225, 195,  10, 210, 164, 252, 138,   8,  35, 152, 213, 199,
+  82, 116,  97, 230,  63, 199, 241,  35,  79, 120,  54, 174,  67, 112,   1,  76,
+  69, 222, 194,  96,  82,  94,  25, 228, 196, 145, 155, 136, 228, 234,  46, 101,
+ 246,  51, 103, 166, 246,  75,   9, 200, 161,   4, 108,  35, 129, 168, 208, 144,
+  50,  14,  13, 220,  41, 132, 122, 127, 194,   9, 232, 234, 107,  28, 187,   8,
+  51, 141,  97, 221, 225,   9, 113, 170, 166, 102, 135,  22, 231, 185, 227, 187,
+ 110, 145, 251, 146,  76,  22, 146, 228,   7,  53,  64,  25,  62, 198, 130, 190,
+ 221, 232, 169,  64, 188, 199, 237, 249, 173, 218, 196, 191,  48, 224,   5, 113,
+ 100, 166, 160,  21, 191, 197,  61, 162, 149, 171, 240, 183, 129, 231, 123, 204,
+ 192, 179, 134,  15,  47, 161, 142, 177, 239, 234, 186, 237, 231,  53, 208,  95,
+ 146,  36, 225, 231,  89, 142,  93, 248, 137, 124,  83,  39,  69,  77,  89, 208,
+ 182,  48,  85, 147, 244, 164, 246,  68,  38, 190, 220,  35, 202,  91, 157, 151,
+ 201, 240, 185, 218,   4, 152,   2, 132, 177,  88, 190, 196, 229,  74, 220, 135,
+ 137, 196,  11,  47,   5, 251, 106, 144, 163,  60, 222, 127,  52,  57, 202, 102,
+  64, 140, 110, 206,  23, 182,  39, 245,   1, 163, 157, 186, 163,  80,   7, 230,
+  44, 249, 176, 102, 164, 125, 147, 120,  18, 191, 186, 125,  64,  65, 198, 157,
+ 164, 213,  95,  61,  13, 181, 208,  91, 242, 197, 158,  34,  98, 169,  91,  14,
+  17,  93, 157,  17,  65,  30, 183,   6, 139,  58, 255, 108, 100, 136, 209, 144,
+ 164,   6, 237,  33, 210, 110,  57, 126, 197, 136, 125, 244, 165, 151, 168,   3,
+ 143, 251, 247, 155, 136, 130,  88,  14,  74, 121, 250, 133,  21, 226, 185, 232,
+ 118, 132,  89,  64, 204, 161,   2,  70, 224, 159,  35, 204, 123, 180,  13,  52,
+ 231,  57,  25,  78,  66,  69,  97,  42, 198,  84, 176,  59,   8, 232, 125, 134,
+ 193,   2, 232, 109, 216,  69,  90, 142,  32,  38, 249,  37,  75, 180, 184, 188,
+  19,  47, 120,  87, 146,  70, 232, 120, 191,  45,  33,  38,  19, 248, 110, 110,
+  44,  64,   2,  84, 244, 228, 252, 228, 170, 123,  38, 144, 213, 144, 171, 212,
+ 243,  87, 189,  46, 128, 110,  84,  77,  65, 183,  61, 184, 101,  44, 168,  68,
+  14, 106, 105,   8, 227, 211, 166,  39, 152,  43,  52, 254, 197,  55, 119,  89,
+ 168,  65,  53, 138, 177,  56, 219,   0,  58, 121, 148,  18,  44, 100, 215, 103,
+ 145, 229, 117, 196,  91,  89, 113, 143, 172, 239, 249, 184, 154,  39, 112,  65,
+ 204,  42,  84,  38, 155, 151, 151,  16, 100,  87, 174, 162, 145, 147, 149, 186,
+ 237, 145, 134, 144, 198, 235, 213, 163,  48, 230,  24,  47,  57,  71, 127,   0,
+ 150, 219,  12,  81, 197, 150, 131,  13, 169,  63, 175, 184,  48, 235,  65, 243,
+ 149, 200, 163, 254, 202, 114, 247,  67, 143, 250, 126, 228,  80, 130, 216, 214,
+  36,   2, 230,  33, 119, 125,   3, 142, 237, 100,   3, 152, 197, 174, 244, 129,
+ 232,  30, 206, 199,  39, 210, 220,  43, 237, 221, 201,  54, 179,  42,  28, 133,
+ 246, 203, 198, 177,   0,  28, 194,  85, 223, 109, 155, 147, 221,  60, 133, 108,
+ 157, 254,  26,  75, 157, 185,  49, 142,  31, 137,  71,  43,  63,  64, 237, 148,
+ 237, 172, 159, 160, 155, 254, 234, 224, 140, 193, 114, 140,  62, 109, 136,  39,
+ 255,   8, 158, 146, 128,  49, 222,  96,  57, 209, 180, 249, 202, 127, 113, 231,
+  78, 178,  46,  33, 228, 215, 104,  31, 207, 186,  82,  41,  42,  39, 103, 119,
+ 123, 133, 243, 254, 238, 156,  90, 186,  37, 212,  33, 107, 252,  51, 177,  36,
+ 237,  76, 159, 245,  93, 214,  97,  56, 190,  38, 160,  94, 105, 222, 220, 158,
+  49,  16, 191,  52, 120,  87, 179,   2,  27, 144, 223, 230, 184,   6, 129, 227,
+  69,  47, 215, 181, 162, 139,  72, 200,  45, 163, 159,  62,   2, 221, 124,  40,
+ 159, 242,  35, 208, 179, 166,  98,  67, 178,  68, 143, 225, 178, 146, 187, 159,
+  57,  66, 176, 192, 236, 250, 168, 224, 122,  43, 159, 120, 133, 165, 122,  64,
+  87,  74, 161, 241,   9,  87,  90,  24, 255, 113, 203, 220,  57, 139, 197, 159,
+  31, 151,  27, 140,  77, 162,   7,  27,  84, 228, 187, 220,  53, 126, 162, 242,
+  84, 181, 223, 103,  86, 177, 207,  31, 140,  18, 207, 256, 201, 166,  96,  23,
+ 233, 103, 197,  84, 161,  75,  59, 149, 138, 154, 119,  92,  16,  53, 116,  97,
+ 220, 114,  35,  45,  77, 209,  40, 196,  71,  22,  81, 178, 110,  14,   3, 180,
+ 110, 129, 112,  47,  18,  61, 134,  78,  73,  79, 254, 232, 125, 180, 205,  54,
+ 220, 119,  63,  89, 181,  52,  77, 109, 151,  77,  80, 207, 144,  25,  20,   6,
+ 208,  47, 201, 206, 192,  14,  73, 176, 256, 201, 207,  87, 216,  60,  56,  73,
+  92, 243, 179, 113,  49,  59,  55, 168, 121, 137,  69, 154,  95,  57, 187,  47,
+ 129,   4,  15,  92,   6, 116,  69, 196,  48, 134,  84,  81, 111,  56,  38, 176,
+ 239,   6, 128,  72, 242, 134,  36, 221,  59,  48, 242,  68, 130, 110, 171,  89,
+  13, 220,  48,  29,   5,  75, 104, 233,  91, 129, 105, 162,  44, 113, 163, 163,
+  85, 147, 190, 111, 197,  80, 213, 153,  81,  68, 203,  33, 161, 165,  10,  61,
+ 120, 252,   0, 205,  28,  42, 193,  64,  39,  37,  83, 175,   5, 218, 215, 174,
+ 128, 121, 231,  11, 150, 145, 135, 197, 136,  91, 193,   5, 107,  88,  82,   6,
+   4, 188, 256,  70,  40,   2, 167,  57, 169, 203, 115, 254, 215, 172,  84,  80,
+ 188, 167,  34, 137,  43, 243,   2,  79, 178,  38, 188, 135, 233, 194, 208,  13,
+  11, 151, 231, 196,  12, 122, 162,  56,  17, 114, 191, 207,  90, 132,  64, 238,
+ 187,   6, 198, 176, 240,  88, 118, 236,  15, 226, 166,  22, 193, 229,  82, 246,
+ 213,  64,  37,  63,  31, 243, 252,  37, 156,  38, 175, 204, 138, 141, 211,  82,
+ 106, 217,  97, 139, 153,  56, 129, 218, 158,   9,  83,  26,  87, 112,  71,  21,
+ 250,   5,  65, 141,  68, 116, 231, 113,  10, 218,  99, 205, 201,  92, 157,   4,
+  97,  46,  49, 220,  72, 139, 103, 171, 149, 129, 193,  19,  69, 245,  43,  31,
+  58,  68,  36, 195, 159,  22,  54,  34, 233, 141, 205, 100, 226,  96,  22, 192,
+  41, 231,  24,  79, 234, 138,  30, 120, 117, 216, 172, 197, 172, 107,  86,  29,
+ 181, 151,   0,   6, 146, 186,  68,  55,  54,  58, 213, 182,  60, 231,  33, 232,
+  77, 210, 216, 154,  80,  51, 141, 122,  68, 148, 219, 122, 254,  48,  64, 175,
+  41, 115,  62, 243, 141,  81, 119, 121,   5,  68, 121,  88, 239,  29, 230,  90,
+ 135, 159,  35, 223, 168, 112,  49,  37, 146,  60, 126, 134,  42, 145, 115,  90,
+  73, 133, 211,  86, 120, 141, 122, 241, 127,  56, 130,  36, 174,  75,  83, 246,
+ 112,  45, 136, 194, 201, 115,   1, 156, 114, 167, 208,  12, 176, 147,  32, 170,
+ 251, 100, 102, 220, 122, 210,   6,  49,  75, 201,  38, 105, 132, 135, 126, 102,
+  13, 121,  76, 228, 202,  20,  61, 213, 246,  13, 207,  42, 148, 168,  37, 253,
+  34,  94, 141, 185,  18, 234, 157, 109, 104,  64, 250, 125,  49, 236,  86,  48,
+ 196,  77,  75, 237, 156, 103, 225,  19, 110, 229,  22,  68, 177,  93, 221, 181,
+ 152, 153,  61, 108, 101,  74, 247, 195, 127, 216,  30, 166, 168,  61,  83, 229,
+ 120, 156,  96, 120, 201, 124,  43,  27, 253, 250, 120, 143,  89, 235, 189, 243,
+ 150,   7, 127, 119, 149, 244,  84, 185, 134,  34, 128, 193, 236, 234, 132, 117,
+ 137,  32, 145, 184,  44, 121,  51,  76,  11, 228, 142, 251,  39,  77, 228, 251,
+  41,  58, 246, 107, 125, 187,   9, 240,  35,   8,  11, 162, 242, 220, 158, 163,
+   2, 184, 163, 227, 242,   2, 100, 101,   2,  78, 129,  34,  89,  28,  26, 157,
+  79,  31, 107, 250, 194, 156, 186,  69, 212,  66,  41, 180, 139,  42, 211, 253,
+ 256, 239,  29, 129, 104, 248, 182,  68,   1, 189,  48, 226,  36, 229,   3, 158,
+  41,  53, 241,  22, 115, 174,  16, 163, 224,  19, 112, 219, 177, 233,  42,  27,
+ 250, 134,  18,  28, 145, 122,  68,  34, 134,  31, 147,  17,  39, 188, 150,  76,
+  45,  42, 167, 249,  12,  16,  23, 182,  13,  79, 121,   3,  70, 197, 239,  44,
+  86, 177, 255,  81,  64, 171, 138, 131,  73, 110,  44, 201, 254, 198, 146,  91,
+  48,   9, 104,  31,  29, 161, 101,  31, 138, 180, 231, 233,  79, 137,  61, 236,
+ 140,  15, 249, 218, 234, 119,  99, 195, 110, 137, 237, 207,   8,  31,  45,  24,
+  90, 155, 203, 253, 192, 203,  65, 176, 210, 171, 142, 214, 220, 122, 136, 237,
+ 189, 186, 147,  40,  80, 254, 173,  33, 191,  46, 192,  26, 108, 255, 228, 205,
+  61,  76,  39, 107, 225, 126, 228, 182, 140, 251, 143, 134, 252, 168, 221,   8,
+ 185,  85,  60, 233, 147, 244,  87, 137,   8, 140,  96,  80,  53,  45, 175, 160,
+ 124, 189, 112,  37, 144,  19,  70,  17, 170, 242,   2,   3,  28,  95, 120, 199,
+ 212,  43,   9, 117,  86, 151, 101, 241, 200, 145, 241,  19, 178,  69, 204, 197,
+ 227, 166,  94,   7, 193,  45, 247, 234,  19, 187, 212, 212, 236, 125,  33,  95,
+ 198, 121, 122, 103,  77, 155, 235,  49,  25, 237, 249,  11, 162,   7, 238,  24,
+  16, 150, 129,  25, 152,  17,  42,  67, 247, 162,  77, 154,  31, 133,  55, 137,
+  79, 119, 153,  10,  86,  28, 244, 186,  41, 169, 106,  44,  10,  49, 110, 179,
+  32, 133, 155, 244,  61,  70, 131, 168, 170,  39, 231, 252,  32,  69,  92, 238,
+ 239,  35, 132, 136, 236, 167,  90,  32, 123,  88,  69,  22,  20,  89, 145, 166,
+  30, 118,  75,   4,  49,  31, 225,  54,  11,  50,  56, 191, 246,   1, 187,  33,
+ 119, 107, 139,  68,  19, 240, 131,  55,  94, 113,  31, 252,  12, 179, 121,   2,
+ 120, 252,   0,  76,  41,  80, 185,  42,  62, 121, 105, 159, 121, 109, 111,  98,
+   7, 118,  86,  29, 210,  70, 231, 179, 223, 229, 164,  70,  62,  47,   0, 206,
+ 204, 178, 168, 120, 224, 166,  99,  25, 103,  63, 246, 224, 117, 204,  75, 124,
+ 140, 133, 110, 110, 222,  88, 151, 118,  46,  37,  22, 143, 158,  40,   2,  50,
+ 153,  94, 190, 199,  13, 198, 127, 211, 180,  90, 183,  98,   0, 142, 210, 154,
+ 100, 187,  67, 231, 202, 100, 198, 235, 252, 160, 247, 124, 247,  14, 121, 221,
+  57,  88, 253, 243, 185,  89,  45, 249, 221, 194, 108, 175, 193, 119,  50, 141,
+ 223, 133, 136,  64, 176, 250, 129, 100, 124,  94, 181, 159,  99, 185, 177, 240,
+ 135,  42, 103,  52, 202, 208, 143, 186, 193, 103, 154, 237, 102,  88, 225, 161,
+  50, 188, 191, 109,  12,  87,  19, 227, 247, 183,  13,  52, 205, 170, 205, 146,
+  89, 160,  18, 105, 192,  73, 231, 225, 184, 157, 252, 220,  61,  59, 169, 183,
+ 221,  20, 141,  20, 158, 101, 245,   7, 245, 225, 118, 137,  84,  55,  19,  27,
+ 164, 110,  35,  25, 202,  94, 150,  46,  91, 152, 130,   1,   7,  46,  16, 237,
+ 171, 109,  19, 200,  65,  38,  10, 213,  70,  96, 126, 226, 185, 225, 181,  46,
+  10, 165,  11, 123,  53, 158,  22, 147,  64,  22, 227,  69, 182, 237, 197,  37,
+  39,  49, 186, 223, 139, 128,  55,  36, 166, 178, 220,  20,  98, 172, 166, 253,
+  45,   0, 120, 180, 189, 185, 158, 159, 196,   6, 214,  79, 141,  52, 156, 107,
+   5, 109, 142, 159,  33,  64, 190, 133,  95, 132,  95, 202, 160,  63, 186,  23,
+ 231, 107, 163,  33, 234,  15, 244,  77, 108,  49,  51,   7, 164,  87, 142,  99,
+ 240, 202,  47, 256, 118, 190, 196, 178, 217,  42,  39, 153,  21, 192, 232, 202,
+  14,  82, 179,  64, 233,   4, 219,  10, 133,  78,  43, 144, 146, 216, 202,  81,
+  71, 252,   8, 201,  68, 256,  85, 233, 164,  88, 176,  30,   5, 152, 126, 179,
+ 249,  84, 140, 190, 159,  54, 118,  98,   2, 159,  27, 133,  74, 121, 239, 196,
+  71, 149, 119, 135, 102,  20,  87, 112,  44,  75, 221,   3, 151, 158,   5,  98,
+ 152,  25,  97, 106,  63, 171, 240,  79, 234, 240, 230,  92,  76,  70, 173, 196,
+  36, 225, 218, 133,  64, 240, 150,  41, 146,  66, 133,  51, 134,  73, 170, 238,
+ 140,  90,  45,  89,  46, 147,  96, 169, 174, 174, 244, 151,  90,  40,  32,  74,
+  38, 154, 246,  57,  31,  14, 189, 151,  83, 243, 197, 183, 220, 185,  53, 225,
+  51, 106, 188, 208, 222, 248,  93,  13,  93, 215, 131,  25, 142, 185, 113, 222,
+ 131, 215, 149,  50, 159,  85,  32,   5, 205, 192,   2, 227,  42, 214, 197,  42,
+ 126, 182,  68, 123, 109,  36, 237, 179, 170, 199,  77, 256,   5, 128, 214, 243,
+ 137, 177, 170, 253, 179, 180, 153, 236, 100, 196, 216, 231, 198,  37, 192,  80,
+ 121, 221, 246,   1,  16, 246,  29,  78,  64, 148, 124,  38,  96, 125,  28,  20,
+  48,  51,  73, 187, 139, 208,  98, 253, 221, 188,  84, 129,   1, 205,  95, 205,
+ 117,  79,  71, 126, 134, 237,  19, 184, 137, 125, 129, 178, 223,  54, 188, 112,
+  30,   7, 225, 228, 205, 184, 233,  87, 117,  22,  58,  10,   8,  42,   2, 114,
+ 254,  19,  17,  13, 150,  92, 233, 179,  63,  12,  60, 171, 127,  35,  50,   5,
+ 195, 113, 241,  25, 249, 184, 166,  44, 221,  35, 151, 116,   8,  54, 195,  89,
+ 218, 186, 132,   5,  41,  89, 226, 177,  11,  41,  87, 172,   5,  23,  20,  59,
+ 228,  94,  76,  33, 137,  43, 151, 221,  61, 232,   4, 120,  93, 217,  80, 228,
+ 228,   6,  58,  25,  62,  84,  91,  48, 209,  20, 247, 243,  55, 106,  80,  79,
+ 235,  34,  20, 180, 146,   2, 236,  13, 236, 206, 243, 222, 204,  83, 148, 213,
+ 214, 117, 237,  98,   0,  90, 204, 168,  32,  41, 126,  67, 191,  74,  27, 255,
+  26,  75, 240, 113, 185, 105, 167, 154, 112,  67, 151,  63, 161, 134, 239, 176,
+  42,  87, 249, 130,  45, 242,  17, 100, 107, 120, 212, 218, 237,  76, 231, 162,
+ 175, 172, 118, 155,  92,  36, 124,  17, 121,  71,  13,   9,  82, 126, 147, 142,
+ 218, 148, 138,  80, 163, 106, 164, 123, 140, 129,  35,  42, 186, 154, 228, 214,
+  75,  73,   8, 253,  42, 153, 232, 164,  95,  24, 110,  90, 231, 197,  90, 196,
+  57, 164, 252, 181,  31,   7,  97, 256,  35,  77, 200, 212,  99, 179,  92, 227,
+  17, 180,  49, 176,   9, 188,  13, 182,  93,  44, 128, 219, 134,  92, 151,   6,
+  23, 126, 200, 109,  66,  30, 140, 180, 146, 134,  67, 200,   7,   9, 223, 168,
+ 186, 221,   3, 154, 150, 165,  43,  53, 138,  27,  86, 213, 235, 160,  70,   2,
+ 240,  20,  89, 212,  84, 141, 168, 246, 183, 227,  30, 167, 138, 185, 253,  83,
+  52, 143, 236,  94,  59,  65,  89, 218, 194, 157, 164, 156, 111,  95, 202, 168,
+ 245, 256, 151,  28, 222, 194,  72, 130, 217, 134, 253,  77, 246, 100,  76,  32,
+ 254, 174, 182, 193,  14, 237,  74,   1,  74,  26, 135, 216, 152, 208, 112,  38,
+ 181,  62,  25,  71,  61, 234, 254,  97, 191,  23,  92, 256, 190, 205,   6,  16,
+ 134, 147, 210, 219, 148,  59,  73, 185,  24, 247, 174, 143, 116, 220, 128, 144,
+ 111, 126, 101,  98, 130, 136, 101, 102,  69, 127,  24, 168, 146, 226, 226, 207,
+ 176, 122, 149, 254, 134, 196,  22, 151, 197,  21,  50, 205, 116, 154,  65, 116,
+ 177, 224, 127,  77, 177, 159, 225,  69, 176,  54, 100, 104, 140,   8,  11, 126,
+  11, 188, 185, 159, 107,  16, 254, 142,  80,  28,   5, 157, 104,  57, 109,  82,
+ 102,  80, 173, 242, 238, 207,  57, 105, 237, 160,  59, 189, 189, 199,  26,  11,
+ 190, 156,  97, 118,  20,  12, 254, 189, 165, 147, 142, 199,   5, 213,  64, 133,
+ 108, 217, 133,  60,  94,  28, 116, 136,  47, 165, 125,  42, 183, 143,  14, 129,
+ 223,  70, 212, 205, 181, 180,   3, 201, 182,  46,  57, 104, 239,  60,  99, 181,
+ 220, 231,  45,  79, 156,  89, 149, 143, 190, 103, 153,  61, 235,  73, 136,  20,
+  89, 243,  16, 130, 247, 141, 134,  93,  80,  68,  85,  84,   8,  72, 194,   4,
+ 242, 110,  19, 133, 199,  70, 172,  92, 132, 254,  67,  74,  36,  94,  13,  90,
+ 154, 184,   9, 109, 118, 243, 214,  71,  36,  95,   0,  90, 201, 105, 112, 215,
+  69, 196, 224, 210, 236, 242, 155, 211,  37, 134,  69, 113, 157,  97,  68,  26,
+ 230, 149, 219, 180,  20,  76, 172, 145, 154,  40, 129,   8,  93,  56, 162, 124,
+ 207, 233, 105,  19,   3, 183, 155, 134,   8, 244, 213,  78, 139,  88, 156,  37,
+  51, 152, 111, 102, 112, 250, 114, 252, 201, 241, 133,  24, 136, 153,   5,  90,
+ 210, 197, 216,  24, 131,  17, 147, 246,  13,  86,   3, 253, 179, 237, 101, 114,
+ 243, 191, 207,   2, 220, 133, 244,  53,  87, 125, 154, 158, 197,  20,   8,  83,
+  32, 191,  38, 241, 204,  22, 168,  59, 217, 123, 162,  82,  21,  50, 130,  89,
+ 239, 253, 195,  56, 253,  74, 147, 125, 234, 199, 250,  28,  65, 193,  22, 237,
+ 193,  94,  58, 229, 139, 176,  69,  42, 179, 164, 150, 168, 246, 214,  86, 174,
+  59, 117,  15,  19,  76,  37, 214, 238, 153, 226, 154,  45, 109, 114, 198, 107,
+  45,  70, 238, 196, 142, 252, 244,  71, 123, 136, 134, 188,  99, 132,  25,  42,
+ 240,   0, 196,  33,  26, 124, 256, 145,  27, 102, 153,  35,  28, 132, 221, 167,
+ 138, 133,  41, 170,  95, 224,  40, 139, 239, 153,   1, 106, 255, 106, 170, 163,
+ 127,  44, 155, 232, 194, 119, 232, 117, 239, 143, 108,  41,   3,   9, 180, 256,
+ 144, 113, 133, 200,  79,  69, 128, 216,  31,  50, 102, 209, 249, 136, 150, 154,
+ 182,  51, 228,  39, 127, 142,  87,  15,  94,  92, 187, 245,  31, 236,  64,  58,
+ 114,  11,  17, 166, 189, 152, 218,  34, 123,  39,  58,  37, 153,  91,  63, 121,
+  31,  34,  12, 254, 106,  96, 171,  14, 155, 247, 214,  69,  24,  98,   3, 204,
+ 202, 194, 207,  30, 253,  44, 119,  70,  14,  96,  82, 250,  63,   6, 232,  38,
+  89, 144, 102, 191,  82, 254,  20, 222,  96, 162, 110,   6, 159,  58, 200, 226,
+  98, 128,  42,  70,  84, 247, 128, 211, 136,  54, 143, 166,  60, 118,  99, 218,
+  27, 193,  85,  81, 219, 223,  46,  41,  23, 233, 152, 222,  36, 236,  54, 181,
+  56,  50,   4, 207, 129,  92,  78,  88, 197, 251, 131, 105,  31, 172,  38, 131,
+  19, 204, 129,  47, 227, 106, 202, 183,  23,   6,  77, 224, 102, 147,  11, 218,
+ 131, 132,  60, 192, 208, 223, 236,  23, 103, 115,  89,  18, 185, 171,  70, 174,
+ 139,   0, 100, 160, 221,  11, 228,  60,  12, 122, 114,  12, 157, 235, 148,  57,
+  83,  62, 173, 131, 169, 126,  85,  99,  93, 243,  81,  80,  29, 245, 206,  82,
+ 236, 227, 166,  14, 230, 213, 144,  97,  27, 111,  99, 164, 105, 150,  89, 111,
+ 252, 118, 140, 232, 120, 183, 137, 213, 232, 157, 224,  33, 134, 118, 186,  80,
+ 159,   2, 186, 193,  54, 242,  25, 237, 232, 249, 226, 213,  90, 149,  90, 160,
+ 118,  69,  64,  37,  10, 183, 109, 246,  30,  52, 219,  69, 189,  26, 116, 220,
+  50, 244, 243, 243, 139, 137, 232,  98,  38,  45, 256, 143, 171, 101,  73, 238,
+ 123,  45, 194, 167, 250, 123,  12,  29, 136, 237, 141,  21,  89,  96, 199,  44,
+   8, 214, 208,  17, 113,  41, 137,  26, 166, 155,  89,  85,  54,  58,  97, 160,
+  50, 239,  58,  71,  21, 157, 139,  12,  37, 198, 182, 131, 149, 134,  16, 204,
+ 164, 181, 248, 166,  52, 216, 136, 201,  37, 255, 187, 240,   5, 101, 147, 231,
+  14, 163, 253, 134, 146, 216,   8,  54, 224,  90, 220, 195,  75, 215, 186,  58,
+  71, 204, 124, 105, 239,  53,  16,  85,  69, 163, 195, 223,  33,  38,  69,  88,
+  88, 203,  99,  55, 176,  13, 156, 204, 236,  99, 194, 134,  75, 247, 126, 129,
+ 160, 124, 233, 206, 139, 144, 154,  45, 233,  51, 206,  61,  60,  55, 205, 107,
+  84, 108,  96, 188, 203,  31,  89,  20, 115, 144, 137,  90, 237,  78, 231, 185,
+ 120, 217,   1, 176, 169,  30, 155, 176, 100, 113,  53,  42, 193, 108,  14, 121,
+ 176, 158, 137,  92, 178,  44, 110, 249, 108, 234,  94, 101, 128,  12, 250, 173,
+  72, 202, 232,  66, 139, 152, 189,  18,  32, 197,   9, 238, 246,  55, 119, 183,
+ 196, 119, 113, 247, 191, 100, 200, 245,  46,  16, 234, 112, 136, 116, 232,  48,
+ 176, 108,  11, 237,  14, 153,  93, 177, 124,  72,  67, 121, 135, 143,  45,  18,
+  97, 251, 184, 172, 136,  55, 213,   8, 103,  12, 221, 212,  13, 160, 116,  91,
+ 237, 127, 218, 190, 103, 131,  77,  82,  36, 100,  22, 252,  79,  69,  54,  26,
+  65, 182, 115, 142, 247,  20,  89,  81, 188, 244,  27, 120, 240, 248,  13, 230,
+  67, 133,  32, 201, 129,  87,   9, 245,  66,  88, 166,  34,  46, 184, 119, 218,
+ 144, 235, 163,  40, 138, 134, 127, 217,  64, 227, 116,  67,  55, 202, 130,  48,
+ 199,  42, 251, 112, 124, 153, 123, 194, 243,  49, 250,  12,  78, 157, 167, 134,
+ 210,  73, 156, 102,  21,  88, 216, 123,  45,  11, 208,  18,  47, 187,  20,  43,
+   3, 180, 124,   2, 136, 176,  77, 111, 138, 139,  91, 225, 126,   8,  74, 255,
+  88, 192, 193, 239, 138, 204, 139, 194, 166, 130, 252, 184, 140, 168,  30, 177,
+ 121,  98, 131, 124,  69, 171,  75,  49, 184,  34,  76, 122, 202, 115, 184, 253,
+ 120, 182,  33, 251,   1,  74, 216, 217, 243, 168,  70, 162, 119, 158, 197, 198,
+  61,  89,   7,   5,  54, 199, 211, 170,  23, 226,  44, 247, 165, 195,   7, 225,
+  91,  23,  50,  15,  51, 208, 106,  94,  12,  31,  43, 112, 146, 139, 246, 182,
+ 113,   1,  97,  15,  66,   2,  51,  76, 164, 184, 237, 200, 218, 176,  72,  98,
+  33, 135,  38, 147, 140, 229,  50,  94,  81, 187, 129,  17, 238, 168, 146, 203,
+ 181,  99, 164,   3, 104,  98, 255, 189, 114, 142,  86, 102, 229, 102,  80, 129,
+  64,  84,  79, 161,  81, 156, 128, 111, 164, 197,  18,  15,  55, 196, 198, 191,
+  28, 113, 117,  96, 207, 253,  19, 158, 231,  13,  53, 130, 252, 211,  58, 180,
+ 212, 142,   7, 219,  38,  81,  62, 109, 167, 113,  33,  56,  97, 185, 157, 130,
+ 186, 129, 119, 182, 196,  26,  54, 110,  65, 170, 166, 236,  30,  22, 162,   0,
+ 106,  12, 248,  33,  48,  72, 159,  17,  76, 244, 172, 132,  89, 171, 196,  76,
+ 254, 166,  76, 218, 226,   3,  52, 220, 238, 181, 179, 144, 225,  23,   3, 166,
+ 158,  35, 228, 154, 204,  23, 203,  71, 134, 189,  18, 168, 236, 141, 117, 138,
+   2, 132,  78,  57, 154,  21, 250, 196, 184,  40, 161,  40,  10, 178, 134, 120,
+ 132, 123, 101,  82, 205, 121,  55, 140, 231,  56, 231,  71, 206, 246, 198, 150,
+ 146, 192,  45, 105, 242,   1, 125,  18, 176,  46, 222, 122,  19,  80, 113, 133,
+ 131, 162,  81,  51,  98, 168, 247, 161, 139,  39,  63, 162,  22, 153, 170,  92,
+  91, 130, 174, 200,  45, 112,  99, 164, 132, 184, 191, 186, 200, 167,  86, 145,
+ 167, 227, 130,  44,  12, 158, 172, 249, 204,  17,  54, 249,  16, 200,  21, 174,
+  67, 223, 105, 201,  50,  36, 133, 203, 244, 131, 228,  67,  29, 195,  91,  91,
+  55, 107, 167, 154, 170, 137, 218, 183, 169,  61,  99, 175, 128,  23, 142, 183,
+  66, 255,  59, 187,  66,  85, 212, 109, 168,  82,  16,  43,  67, 139, 114, 176,
+ 216, 255, 130,  94, 152,  79, 183,  64, 100,  23, 214,  82,  34, 230,  48,  15,
+ 242, 130,  50, 241,  81,  32,   5, 125, 183, 182, 184,  99, 248, 109, 159, 210,
+ 226,  61, 119, 129,  39, 149,  78, 214, 107,  78, 147, 124, 228,  18, 143, 188,
+  84, 180, 233, 119,  64,  39, 158, 133, 177, 168,   6, 150,  80, 117, 150,  56,
+  49,  72,  49,  37,  30, 242,  49, 142,  33, 156,  34,  44,  44,  72,  58,  22,
+ 249,  46, 168,  80,  25, 196,  64, 174,  97, 179, 244, 134, 213, 105,  63, 151,
+  21,  90, 168,  90, 245,  28, 157,  65, 250, 232, 188,  27,  99, 160, 156, 127,
+  68, 193,  10,  80, 205,  36, 138, 229,  12, 223,  70, 169, 251,  41,  48,  94,
+  41, 177,  99, 256, 158,   0,   6,  83, 231, 191, 120, 135, 157, 146, 218, 213,
+ 160,   7,  47, 234,  98, 211,  79, 225, 179,  95, 175, 105, 185,  79, 115,   0,
+ 104,  14,  65, 124,  15, 188,  52,   9, 253,  27, 132, 137,  13, 127,  75, 238,
+ 185, 253,  33,   8,  52, 157, 164,  68, 232, 188,  69,  28, 209, 233,   5, 129,
+ 216,  90, 252, 212,  33, 200, 222,   9, 112,  15,  43,  36, 226, 114,  15, 249,
+ 217,   8, 148,  22, 147,  23, 143,  67, 222, 116, 235, 250, 212, 210,  39, 142,
+ 108,  64, 209,  83,  73,  66,  99,  34,  17,  29,  45, 151, 244, 114,  28, 241,
+ 144, 208, 146, 179, 132,  89, 217, 198, 252, 219, 205, 165,  75, 107,  11, 173,
+  76,   6, 196, 247, 152, 216, 248,  91, 209, 178,  57, 250, 174,  60,  79, 123,
+  18, 135,   9, 241, 230, 159, 184,  68, 156, 251, 215,   9, 113, 234,  75, 235,
+ 103, 194, 205, 129, 230,  45,  96,  73, 157,  20, 200, 212, 212, 228, 161,   7,
+ 231, 228, 108,  43, 198,  87, 140, 140,   4, 182, 164,   3,  53, 104, 250, 213,
+  85,  38,  89,  61,  52, 187,  35, 204,  86, 249, 100,  71, 248, 213, 163, 215,
+  66, 106, 252, 129,  40, 111,  47,  24, 186, 221,  85, 205, 199, 237, 122, 181,
+  32,  46, 182, 135,  33, 251, 142,  34, 208, 242, 128, 255,   4, 234,  15,  33,
+ 167, 222,  32, 186, 191,  34, 255, 244,  98, 240, 228, 204,  30, 142,  32,  70,
+  69,  83, 110, 151,  10, 243, 141,  21, 223,  69,  61,  37,  59, 209, 102, 114,
+ 223,  33, 129, 254, 255, 103,  86, 247, 235,  72, 126, 177, 102, 226, 102,  30,
+ 149, 221,  62, 247, 251, 120, 163, 173,  57, 202, 204,  24,  39, 106, 120, 143,
+ 202, 176, 191, 147,  37,  38,  51, 133,  47, 245, 157, 132, 154,  71, 183, 111,
+  30, 180,  18, 202,  82,  96, 170,  91, 157, 181, 212, 140, 256,   8, 196, 121,
+ 149,  79,  66, 127, 113,  78,   4, 197,  84, 256, 111, 222, 102,  63, 228, 104,
+ 136, 223,  67, 193,  93, 154, 249,  83, 204, 101, 200, 234,  84, 252, 230, 195,
+  43, 140, 120, 242,  89,  63, 166, 233, 209,  94,  43, 170, 126,   5, 205,  78,
+ 112,  80, 143, 151, 146, 248, 137, 203,  45, 183,  61,   1, 155,   8, 102,  59,
+  68, 212, 230,  61, 254, 191, 128, 223, 176, 123, 229,  27, 146, 120,  96, 165,
+ 213,  12, 232,  40, 186, 225,  66, 105, 200, 195, 212, 110, 237, 238, 151,  19,
+  12, 171, 150,  82,   7, 228,  79,  52,  15,  78,  62,  43,  21, 154, 114,  21,
+  12, 212, 256, 232, 125, 127,   5,  51,  37, 252, 136,  13,  47, 195, 168, 191,
+ 231,  55,  57, 251, 214, 116,  15,  86, 210,  41, 249, 242, 119,  27, 250, 203,
+ 107,  69,  90,  43, 206, 154, 127,  54, 100,  78, 187,  54, 244, 177, 234, 167,
+ 202, 136, 209, 171,  69, 114, 133, 173,  26, 139,  78, 141, 128,  32, 124,  39,
+  45, 218,  96,  68,  90,  44,  67,  62,  83, 190, 188, 256, 103,  42, 102,  64,
+ 249,   0, 141,  11,  61,  69,  70,  66, 233, 237,  29, 200, 251, 157,  71,  51,
+  64, 133, 113,  76,  35, 125,  76, 137, 217, 145,  35,  69, 226, 180,  56, 249,
+ 156, 163, 176, 237,  81,  54,  85, 169, 115, 211, 129,  70, 248,  40, 252, 192,
+ 194, 101, 247,   8, 181, 124, 217, 191, 194,  93,  99, 127, 117, 177, 144, 151,
+ 228, 121,  32,  11,  89,  81,  26,  29, 183,  76, 249, 132, 179,  70,  34, 102,
+  20,  66,  87,  63, 124, 205, 174, 177,  87, 219,  73, 218,  91,  87, 176,  72,
+  15, 211,  47,  61, 251, 165,  39, 247, 146,  70, 150,  57,   1, 212,  36, 162,
+  39,  38,  16, 216,   3,  50, 116, 200,  32, 234,  77, 181, 155,  19,  90, 188,
+  36,   6, 254,  46,  46, 203,  25, 230, 181, 196,   4, 151, 225,  65, 122, 216,
+ 168,  86, 158, 131, 136,  16,  49, 102, 233,  64, 154,  88, 228,  52, 146,  69,
+  93, 157, 243, 121,  70, 209, 126, 213,  88, 145, 236,  65,  70,  96, 204,  47,
+  10, 200,  77,   8, 103, 150,  48, 153,   5,  37,  52, 235, 209,  31, 181, 126,
+  83, 142, 224, 140,   6,  32, 200, 171, 160, 179, 115, 229,  75, 194, 208,  39,
+  59, 223,  52, 247,  38, 197, 135,   1,   6, 189, 106, 114, 168,   5, 211, 222,
+  44,  63,  90, 160, 116, 172, 170, 133, 125, 138,  39, 131,  23, 178,  10, 214,
+  36,  93,  28,  59,  68,  17, 123,  25, 255, 184, 204, 102, 194, 214, 129,  94,
+ 159, 245, 112, 141,  62,  11,  61, 197, 124, 221, 205,  11,  79,  71, 201,  54,
+  58, 150,  29, 121,  87,  46, 240, 201,  68,  20, 194, 209,  47, 152, 158, 174,
+ 193, 164, 120, 255, 216, 165, 247,  58,  85, 130, 220,  23, 122, 223, 188,  98,
+  21,  70,  72, 170, 150, 237,  76, 143, 112, 238, 206, 146, 215, 110,   4, 250,
+  68,  44, 174, 177,  30,  98, 143, 241, 180, 127, 113,  48,   0,   1, 179, 199,
+  59, 106, 201, 114,  29,  86, 173, 133, 217,  44, 200, 141, 107, 172,  16,  60,
+  82,  58, 239,  94, 141, 234, 186, 235, 109, 173, 249, 139, 141,  59, 100, 248,
+  84, 144,  49, 160,  51, 207, 164, 103,  74,  97, 146, 202, 193, 125, 168, 134,
+ 236, 111, 135, 121,  59, 145, 168, 200, 181, 173, 109,   2, 255,   6,   9, 245,
+  90, 202, 214, 143, 121,  65,  85, 232, 132,  77, 228,  84,  26,  54, 184,  15,
+ 161,  29, 177,  79,  43,   0, 156, 184, 163, 165,  62,  90, 179,  93,  45, 239,
+   1,  16, 120, 189, 127,  47,  74, 166,  20, 214, 233, 226,  89, 217, 229,  26,
+ 156,  53, 162,  60,  21,   3, 192,  72, 111,  51,  53, 101, 181, 208,  88,  82,
+ 179, 160, 219, 113, 240, 108,  43, 224, 162, 147,  62,  14,  95,  81, 205,   4,
+ 160, 177, 225, 115,  29,  69, 235, 168, 148,  29, 128, 114, 124, 129, 172, 165,
+ 215, 231, 214,  86, 160,  44, 157,  91, 248, 183,  73, 164,  56, 181, 162,  92,
+ 141, 118, 127, 240, 196,  77,   0,   9, 244,  79, 250, 100, 195,  25, 255,  85,
+  94,  35, 212, 137, 107,  34, 110,  20, 200, 104,  17,  32, 231,  43, 150, 159,
+ 231, 216, 223, 190, 226, 109, 162, 197,  87,  92, 224,  11, 111,  73,  60, 225,
+ 238,  73, 246, 169,  19, 217, 119,  38, 121, 118,  70,  82,  99, 241, 110,  67,
+  31,  76, 146, 215, 124, 240,  31, 103, 139, 224,  75, 160,  31,  78,  93,   4,
+  64,   9, 103, 223,   6, 227, 119,  85, 116,  81,  21,  43,  46, 206, 234, 132,
+  85,  99,  22, 131, 135,  97,  86,  13, 234, 188,  21,  14,  89, 169, 207, 238,
+ 219, 177, 190,  72, 157,  41, 114, 140,  92, 141, 186,   1,  63, 107, 225, 184,
+ 118, 150, 153, 254, 241, 106, 120, 210, 104, 144, 151, 161,  88, 206, 125, 164,
+  15, 211, 173,  49, 146, 241,  71,  36,  58, 201,  46,  27,  33, 187,  91, 162,
+ 117,  19, 210, 213, 187,  97, 193,  50, 190, 114, 217,  60,  61, 167, 207, 213,
+ 213,  53, 135,  34, 156,  91, 115, 119,  46,  99, 242,   1,  90,  52, 198, 227,
+ 201,  91, 216, 146, 210,  82, 121,  38,  73, 133, 182, 193, 132, 148, 246,  75,
+ 109, 157, 179, 113, 176, 134, 205, 159, 148,  58, 103, 171, 132, 156, 133, 147,
+ 161, 231,  39, 100, 175,  97, 125,  28, 183, 129, 135, 191, 202, 181,  29, 218,
+  43, 104, 148, 203, 189, 204,   4, 182, 169,   1, 134, 122, 141, 202,  13, 187,
+ 177, 112, 162,  35, 231,   6,   8, 241,  99,   6, 191,  45, 113, 113, 101, 104};
+
+// The S-Box we use for further linearity breaking.
+// We created it by taking the digits of decimal expansion of e.
+// The code that created it can be found in 'ProduceRandomSBox.c'.
+unsigned char SBox[256] = {
+//0     1    2      3     4    5     6     7      8    9     A      B    C     D     E     F
+0x7d, 0xd1, 0x70, 0x0b, 0xfa, 0x39, 0x18, 0xc3, 0xf3, 0xbb, 0xa7, 0xd4, 0x84, 0x25, 0x3b, 0x3c,   // 0
+0x2c, 0x15, 0x69, 0x9a, 0xf9, 0x27, 0xfb, 0x02, 0x52, 0xba, 0xa8, 0x4b, 0x20, 0xb5, 0x8b, 0x3a,   // 1
+0x88, 0x8e, 0x26, 0xcb, 0x71, 0x5e, 0xaf, 0xad, 0x0c, 0xac, 0xa1, 0x93, 0xc6, 0x78, 0xce, 0xfc,   // 2
+0x2a, 0x76, 0x17, 0x1f, 0x62, 0xc2, 0x2e, 0x99, 0x11, 0x37, 0x65, 0x40, 0xfd, 0xa0, 0x03, 0xc1,   // 3
+0xca, 0x48, 0xe2, 0x9b, 0x81, 0xe4, 0x1c, 0x01, 0xec, 0x68, 0x7a, 0x5a, 0x50, 0xf8, 0x0e, 0xa3,   // 4
+0xe8, 0x61, 0x2b, 0xa2, 0xeb, 0xcf, 0x8c, 0x3d, 0xb4, 0x95, 0x13, 0x08, 0x46, 0xab, 0x91, 0x7b,   // 5
+0xea, 0x55, 0x67, 0x9d, 0xdd, 0x29, 0x6a, 0x8f, 0x9f, 0x22, 0x4e, 0xf2, 0x57, 0xd2, 0xa9, 0xbd,   // 6
+0x38, 0x16, 0x5f, 0x4c, 0xf7, 0x9e, 0x1b, 0x2f, 0x30, 0xc7, 0x41, 0x24, 0x5c, 0xbf, 0x05, 0xf6,   // 7
+0x0a, 0x31, 0xa5, 0x45, 0x21, 0x33, 0x6b, 0x6d, 0x6c, 0x86, 0xe1, 0xa4, 0xe6, 0x92, 0x9c, 0xdf,   // 8
+0xe7, 0xbe, 0x28, 0xe3, 0xfe, 0x06, 0x4d, 0x98, 0x80, 0x04, 0x96, 0x36, 0x3e, 0x14, 0x4a, 0x34,   // 9
+0xd3, 0xd5, 0xdb, 0x44, 0xcd, 0xf5, 0x54, 0xdc, 0x89, 0x09, 0x90, 0x42, 0x87, 0xff, 0x7e, 0x56,   // A
+0x5d, 0x59, 0xd7, 0x23, 0x75, 0x19, 0x97, 0x73, 0x83, 0x64, 0x53, 0xa6, 0x1e, 0xd8, 0xb0, 0x49,   // B
+0x3f, 0xef, 0xbc, 0x7f, 0x43, 0xf0, 0xc9, 0x72, 0x0f, 0x63, 0x79, 0x2d, 0xc0, 0xda, 0x66, 0xc8,   // C
+0x32, 0xde, 0x47, 0x07, 0xb8, 0xe9, 0x1d, 0xc4, 0x85, 0x74, 0x82, 0xcc, 0x60, 0x51, 0x77, 0x0d,   // D
+0xaa, 0x35, 0xed, 0x58, 0x7c, 0x5b, 0xb9, 0x94, 0x6e, 0x8d, 0xb1, 0xc5, 0xb7, 0xee, 0xb6, 0xae,   // E
+0x10, 0xe0, 0xd6, 0xd9, 0xe5, 0x4f, 0xf1, 0x12, 0x00, 0xd0, 0xf4, 0x1a, 0x6f, 0x8a, 0xb3, 0xb2 }; // F
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+//
+//	Helper functions definition portion.
+//
+///////////////////////////////////////////////////////////////////////////////////////////////
+
+// Don't vectorize, move decl to header file
+
+// Translates an input array with values in base 257 to output array with values in base 256.
+// Returns the carry bit.
+//
+// Parameters:
+// - input: the input array of size EIGHTH_N. Each value in the array is a number in Z_257.
+//          The MSB is assumed to be the last one in the array.
+// - output: the input array encoded in base 256.
+//
+// Returns:
+// - The carry bit (MSB).
+swift_int16_t TranslateToBase256(swift_int32_t input[EIGHTH_N], unsigned char output[EIGHTH_N]);
+
+// Translates an input integer into the range (-FIELD_SIZE / 2) <= result <= (FIELD_SIZE / 2).
+//
+// Parameters:
+// - x: the input integer.
+//
+// Returns:
+// - The result, which equals (x MOD FIELD_SIZE), such that |result| <= (FIELD_SIZE / 2).
+int Center(int x);
+
+// Calculates bit reversal permutation.
+//
+// Parameters:
+// - input: the input to reverse.
+// - numOfBits: the number of bits in the input to reverse.
+//
+// Returns:
+// - The resulting number, which is obtained from the input by reversing its bits.
+int ReverseBits(int input, int numOfBits);
+
+// Initializes the FFT fast lookup table.
+// Shall be called only once.
+void InitializeSWIFFTX();
+
+// Calculates the FFT.
+//
+// Parameters:
+// - input: the input to the FFT.
+// - output: the resulting output.
+void FFT(const unsigned char input[EIGHTH_N], swift_int32_t *output);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// Helper functions implementation portion.
+///////////////////////////////////////////////////////////////////////////////////////////////
+
+// Don't vectorize, delete this copy.
+
+swift_int16_t TranslateToBase256(swift_int32_t input[EIGHTH_N], unsigned char output[EIGHTH_N])
+{
+	swift_int32_t pairs[EIGHTH_N / 2];
+	int i;
+
+	for (i = 0; i < EIGHTH_N; i += 2)
+	{
+		// input[i] + 257 * input[i + 1]
+		pairs[i >> 1] = input[i] + input[i + 1] + (input[i + 1] << 8);
+	}
+
+	for (i = (EIGHTH_N / 2) - 1; i > 0; --i)
+	{
+		int j;
+
+		for (j = i - 1; j < (EIGHTH_N / 2) - 1; ++j)
+		{
+			// pairs[j + 1] * 513, because 257^2 = 513 % 256^2.
+			register swift_int32_t temp = pairs[j] + pairs[j + 1] + (pairs[j + 1] << 9);
+			pairs[j] = temp & 0xffff;
+			pairs[j + 1] += (temp >> 16);
+		}
+	}
+
+	for (i = 0; i < EIGHTH_N; i += 2)
+	{
+		output[i] = (unsigned char) (pairs[i >> 1] & 0xff);
+		output[i + 1] = (unsigned char) ((pairs[i >> 1] >> 8) & 0xff);
+	}
+
+	return (pairs[EIGHTH_N/2 - 1] >> 16);
+}
+
+int Center(int x)
+{
+	int result = x % FIELD_SIZE;
+
+	if (result > (FIELD_SIZE / 2))
+		result -= FIELD_SIZE;
+
+	if (result < (FIELD_SIZE / -2))
+		result += FIELD_SIZE;
+
+	return result;
+}
+
+int ReverseBits(int input, int numOfBits)
+{
+	register int reversed = 0;
+
+	for (input |= numOfBits; input > 1; input >>= 1)
+		reversed = (reversed << 1) | (input & 1);
+
+	return reversed;
+}
+
+void InitializeSWIFFTX()
+{
+	int i, j, k, x;
+	// The powers of OMEGA
+	int omegaPowers[2 * N];
+	omegaPowers[0] = 1;
+
+	if (wasSetupDone)
+		return;
+
+	for (i = 1; i < (2 * N); ++i)
+	{
+		omegaPowers[i] = Center(omegaPowers[i - 1] * OMEGA);
+	}
+
+	for (i = 0; i < (N / W); ++i)
+	{
+		for (j = 0; j < W; ++j)
+		{
+			multipliers[(i << 3) + j] = omegaPowers[ReverseBits(i, N / W) * (2 * j + 1)];
+		}
+	}
+
+	for (x = 0; x < 256; ++x)
+	{
+		for (j = 0; j < 8; ++j)
+		{
+			register int temp = 0;
+			for (k = 0; k < 8; ++k)
+			{
+				temp += omegaPowers[(EIGHTH_N * (2 * j + 1) * ReverseBits(k, W)) % (2 * N)]
+					  * ((x >> k) & 1);
+			}
+
+			fftTable[(x << 3) + j] = Center(temp);
+		}
+	}
+
+	wasSetupDone = true;
+}
+
+// input should be deinterleaved in contiguos memory
+// output and F are 4x32
+// multipliers & fftTable are scalar 16
+
+
+void FFT_4way(const unsigned char input[EIGHTH_N], swift_int32_t *output)
+{
+	swift_int16_t *mult = multipliers;
+   m128_swift_int32_t F[64];
+
+   for (int i = 0; i < 8; i++)
+   {
+      int j = i<<3;
+
+// Need to isolate bytes in input, 8 bytes per lane.
+// Each iteration of the loop process one input vector
+// Each lane reads a different index to ffttable.
+
+// deinterleave the input!
+
+// load table with 4 lanes from different indexes into fftTable
+// extract bytes into m128 4x16
+// mutiply by vectorized mult
+
+// input[lane][byte]
+
+      __m128i table;
+      table = _mm_set_epi32( fftTable[ input[3][i] ],
+                             fftTable[ input[2][i] ],
+                             fftTable[ input[1][i] ],
+                             fftTable[ input[0][i] ] );
+
+      F[i  ] = _mm_mullo_epi32( mm128_const1_32( mult[j+0] ), table );
+
+      table = _mm_set_epi32( fftTable[ input[3][i+1] ]
+                             fftTable[ input[2][i+1] ]
+                             fftTable[ input[1][i+1] ]
+                             fftTable[ input[0][i+1] ] );
+
+      F[i+8] = _mm_mullo_epi32( mm128_const1_32( mult[j+0] ), table );
+
+
+      m128_swift_int16_t *table = &( fftTable[input[i] << 3] );
+
+      F[i   ] = _mm_mullo_epi32( mm128_const1_32( mult[j+0] ),
+                                 mm128_const1_32( table[0] ) );
+      F[i+ 8] = _mm_mullo_epi32( mm128_const1_32( mult[j+1] ),
+                                 mm128_const1_32( table[1] ) );
+      F[i+16] = _mm_mullo_epi32( mm128_const1_32( mult[j+2] ),
+                                 mm128_const1_32( table[2] ) );
+      F[i+24] = _mm_mullo_epi32( mm128_const1_32( mult[j+3] ),
+                                 mm128_const1_32( table[3] ) );
+      F[i+32] = _mm_mullo_epi32( mm128_const1_32( mult[j+4] ),
+                                 mm128_const1_32( table[4] ) );
+      F[i+40] = _mm_mullo_epi32( mm128_const1_32( mult[j+5] ),
+                                 mm128_const1_32( table[5] ) );
+      F[i+48] = _mm_mullo_epi32( mm128_const1_32( mult[j+6] ),
+                                 mm128_const1_32( table[6] ) );
+      F[i+56] = _mm_mullo_epi32( mm128_const1_32( mult[j+7] ),
+                                 mm128_const1_32( table[7] ) );
+   }
+
+
+   for ( int i = 0; i < 8; i++ )
+   {
+      int j = i<<3;
+      ADD_SUB_4WAY( F[j  ], F[j+1] );
+      ADD_SUB_4WAY( F[j+2], F[j+3] );
+      ADD_SUB_4WAY( F[j+4], F[j+5] );
+      ADD_SUB_4WAY( F[j+6], F[j+7] );
+
+      F[j+3] = _mm_slli_epi32( F[j+3], 4 );
+      F[j+7] = _mm_slli_epi32( F[j+7], 4 );
+
+      ADD_SUB_4WAY( F[j  ], F[j+2] );
+      ADD_SUB_4WAY( F[j+1], F[j+3] );
+      ADD_SUB_4WAY( F[j+4], F[j+6] );
+      ADD_SUB_4WAY( F[j+5], F[j+7] );
+
+      F[j+5] = _mm_slli_epi32( F[j+5], 2 );
+      F[j+6] = _mm_slli_epi32( F[j+6], 4 );
+      F[j+7] = _mm_slli_epi32( F[j+7], 6 );
+
+      ADD_SUB_4WAY( F[j  ], F[j+4] );
+      ADD_SUB_4WAY( F[j+1], F[j+5] );
+      ADD_SUB_4WAY( F[j+2], F[j+6] );
+      ADD_SUB_4WAY( F[j+3], F[j+7] );
+
+      output[i   ] = Q_REDUCE_4WAY( F[j  ] );
+      output[i+ 8] = Q_REDUCE_4WAY( F[j+1] );
+      output[i+16] = Q_REDUCE_4WAY( F[j+2] );
+      output[i+24] = Q_REDUCE_4WAY( F[j+3] );
+      output[i+32] = Q_REDUCE_4WAY( F[j+4] );
+      output[i+40] = Q_REDUCE_4WAY( F[j+5] );
+      output[i+48] = Q_REDUCE_4WAY( F[j+6] );
+      output[i+56] = Q_REDUCE_4WAY( F[j+7] );
+   }
+}
+
+// Calculates the FFT part of SWIFFT.
+// We divided the SWIFFT calculation into two, because that way we could save 2 computations of
+// the FFT part, since in the first stage of SWIFFTX the difference between the first 3 SWIFFTs
+// is only the A's part.
+//
+// Parameters:
+// - input: the input to FFT.
+// - m: the input size divided by 8. The function performs m FFTs.
+// - output: will store the result.
+void SWIFFTFFT(const unsigned char *input, int m, swift_int32_t *output)
+{
+	int i;
+
+	for (i = 0;
+		 i < m;
+		 i++, input += EIGHTH_N, output += N)
+	{
+		FFT(input, output);
+	}
+}
+
+// Calculates the 'sum' part of SWIFFT, including the base change at the end.
+// We divided the SWIFFT calculation into two, because that way we could save 2 computations of
+// the FFT part, since in the first stage of SWIFFTX the difference between the first 3 SWIFFTs
+// is only the A's part.
+//
+// Parameters:
+// - input: the input. Of size 64 * m.
+// - m: the input size divided by 64.
+// - output: will store the result.
+// - a: the coefficients in the sum. Of size 64 * m.
+void SWIFFTSum(const swift_int32_t *input, int m, unsigned char *output, const swift_int16_t *a)
+{
+	int i, j;
+	swift_int32_t result[N];
+	register swift_int16_t carry = 0;
+
+	for (j = 0; j < N; ++j)
+	{
+		register swift_int32_t sum = 0;
+		const register swift_int32_t *f = input + j;
+		const register swift_int16_t *k = a + j;
+
+		for (i = 0; i < m; i++, f += N,k += N)
+		{
+			sum += (*f) * (*k);
+		}
+
+		result[j] = sum;
+	}
+
+	for (j = 0; j < N; ++j)
+	{
+		result[j] = ((FIELD_SIZE << 22) + result[j]) % FIELD_SIZE;
+	}
+
+	for (j = 0; j < 8; ++j)
+	{
+		int register carryBit = TranslateToBase256(result + (j << 3), output + (j << 3));
+		carry |= carryBit << j;
+	}
+
+	output[N] = carry;
+}
+
+
+// On entry input is interleaved 4x64. SIZE is *4 lanes / 8 bytes,
+// multiply by 2.
+
+
+void ComputeSingleSWIFFTX_4way( unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
+                          unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE],
+						  bool doSmooth)
+{
+	int i;
+	// Will store the result of the FFT parts:
+   m128_swift_int32_t fftOut[N * M];
+//   swift_int32_t fftOut[N * M];
+	unsigned char intermediate[N * 3 + 8];
+	unsigned char carry0,carry1,carry2;
+
+	// Do the three SWIFFTS while remembering the three carry bytes (each carry byte gets
+	// overriden by the following SWIFFT):
+
+	// 1. Compute the FFT of the input - the common part for the first 3 SWIFFTs:
+	SWIFFTFFT(input, M, fftOut);
+
+	// 2. Compute the sums of the 3 SWIFFTs, each using a different set of coefficients:
+
+	// 2a. The first SWIFFT:
+	SWIFFTSum(fftOut, M, intermediate, As);
+	// Remember the carry byte:
+	carry0 = intermediate[N];
+
+	// 2b. The second one:
+	SWIFFTSum(fftOut, M, intermediate + N, As + (M * N));
+	carry1 = intermediate[2 * N];
+
+	// 2c. The third one:
+	SWIFFTSum(fftOut, M, intermediate + (2 * N), As + 2 * (M * N));
+	carry2 = intermediate[3 * N];
+
+	//2d. Put three carry bytes in their place
+	intermediate[3 * N] = carry0;
+	intermediate[(3 * N) + 1] = carry1;
+	intermediate[(3 * N) + 2] = carry2;
+
+	// Padding  intermediate output with 5 zeroes.
+	memset(intermediate + (3 * N) + 3, 0, 5);
+
+	// Apply the S-Box:
+	for (i = 0; i < (3 * N) + 8; ++i)
+	{
+		intermediate[i] = SBox[intermediate[i]];
+	}
+
+	// 3. The final and last SWIFFT:
+	SWIFFTFFT(intermediate, 3 * (N/8) + 1, fftOut);
+	SWIFFTSum(fftOut,       3 * (N/8) + 1, output, As);
+
+	if (doSmooth)
+	{
+		unsigned char sum[N];
+		register int i, j;
+		memset(sum, 0, N);
+
+		for (i = 0; i < (N + 1) * 8; ++i)
+		{
+			register const swift_int16_t *AsRow;
+			register int AShift;
+
+			if  (!(output[i >> 3] & (1 << (i & 7))))
+			{
+				continue;
+			}
+
+			AsRow = As + N * M + (i & ~(N - 1)) ;
+			AShift = i & 63;
+
+			for (j = AShift; j < N; ++j)
+			{
+				sum[j] += AsRow[j - AShift];
+			}
+
+			for(j = 0; j < AShift; ++j)
+			{
+				sum[j] -= AsRow[N - AShift + j];
+			}
+		}
+
+		for (i = 0; i < N; ++i)
+		{
+			output[i] = sum[i];
+		}
+
+		output[N] = 0;
+	}
+}
--- a/algo/swifftx/swifftx.c
+++ b/algo/swifftx/swifftx.c
--- a/algo/swifftx/swifftx.c.bak
+++ b/algo/swifftx/swifftx.c.bak
--- a/algo/swifftx/swifftx.h
+++ b/algo/swifftx/swifftx.h
@@ -0,0 +1,78 @@
+///////////////////////////////////////////////////////////////////////////////////////////////
+//
+//  SWIFFTX ANSI C OPTIMIZED 32BIT IMPLEMENTATION FOR NIST SHA-3 COMPETITION
+//
+//  SWIFFTX.h
+//
+//  October 2008
+//
+//	This file is the exact copy from the reference implementation.
+//
+///////////////////////////////////////////////////////////////////////////////////////////////
+#ifndef __SWIFFTX__
+#define __SWIFFTX__
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+// See the remarks concerning compatibility issues inside stdint.h.
+//#include <stdint.h>
+#include <stdbool.h>
+#include "stdint.h"
+//#include "stdbool.h"
+//#include "SHA3swift.h"
+
+// The size of SWIFFTX input in bytes.
+#define SWIFFTX_INPUT_BLOCK_SIZE 256
+
+// The size of output block in bytes. The compression function of SWIFFT outputs a block of
+// this size (i.e., this is the size of the resulting hash value).
+#define SWIFFTX_OUTPUT_BLOCK_SIZE 65
+
+// Computes the result of a single SWIFFT operation.
+// This is the simple implementation, where our main concern is to show our design principles.
+// It is made more efficient in the optimized version, by using FFT instead of DFT, and
+// through other speed-up techniques.
+//
+// Parameters:
+// - input: the input string. Consists of 8*m input bytes, where each octet passes the DFT
+//   processing.
+// - m: the length of the input in bytes.
+// - output: the resulting hash value of SWIFFT, of size 65 bytes (520 bit). This is the
+//	 result of summing the dot products of the DFTS with the A's after applying the base
+//	 change transformation
+// - A: the A's coefficients to work with (since every SWIFFT in SWIFFTX uses different As).
+//   A single application of SWIFFT uses 64*m A's.
+void ComputeSingleSWIFFT(unsigned char *input, unsigned short m,
+					  	 unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE],
+						 const swift_int16_t *a);
+
+// Computes the result of a single SWIFFTX operation.
+// NOTE: for simplicity we use 'ComputeSingleSWIFFT()' as a subroutine. This is only to show
+// the design idea. In the optimized versions we don't do this for efficiency concerns, since
+// there we compute the first part (which doesn't involve the A coefficients) only once for all
+// of the 3 invocations of SWIFFT. This enables us to introduce a significant speedup.
+//
+// Parameters:
+// - input: the input input of 256 bytes (2048 bit).
+// - output: the resulting hash value of SWIFFT, of size 64 bytes (512 bit).
+// - doSMooth: if true, a final smoothing stage is performed and the output is of size 512 bits.
+//
+// Returns:
+// - Success value.
+void ComputeSingleSWIFFTX( unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
+                           unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE] );
+
+void ComputeSingleSWIFFTX_smooth( unsigned char input[SWIFFTX_INPUT_BLOCK_SIZE],
+	            unsigned char output[SWIFFTX_OUTPUT_BLOCK_SIZE], bool doSmooth);
+
+// Calculates the powers of OMEGA and generates the bit reversal permutation.
+// You must call this function before doing SWIFFT/X, otherwise you will get zeroes everywhere.
+void InitializeSWIFFTX();
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // __SWIFFTX__
--- a/algo/x11/timetravel-4way.c
+++ b/algo/x11/timetravel-4way.c
@@ -45,12 +45,12 @@ void init_tt8_4way_ctx()

 void timetravel_4way_hash(void *output, const void *input)
 {
-   uint64_t hash0[8] __attribute__ ((aligned (64)));
-   uint64_t hash1[8] __attribute__ ((aligned (64)));
-   uint64_t hash2[8] __attribute__ ((aligned (64)));
-   uint64_t hash3[8] __attribute__ ((aligned (64)));
-   uint64_t vhashX[8*4] __attribute__ ((aligned (64)));
-   uint64_t vhashY[8*4] __attribute__ ((aligned (64)));
+   uint64_t hash0[10] __attribute__ ((aligned (64)));
+   uint64_t hash1[10] __attribute__ ((aligned (64)));
+   uint64_t hash2[10] __attribute__ ((aligned (64)));
+   uint64_t hash3[10] __attribute__ ((aligned (64)));
+   uint64_t vhashX[10*4] __attribute__ ((aligned (64)));
+   uint64_t vhashY[10*4] __attribute__ ((aligned (64)));
   uint64_t *vhashA, *vhashB;
   tt8_4way_ctx_holder ctx __attribute__ ((aligned (64)));
   uint32_t dataLen = 64;
--- a/algo/x11/timetravel10-4way.c
+++ b/algo/x11/timetravel10-4way.c
@@ -51,12 +51,12 @@ void init_tt10_4way_ctx()

 void timetravel10_4way_hash(void *output, const void *input)
 {
-   uint64_t hash0[8] __attribute__ ((aligned (64)));
-   uint64_t hash1[8] __attribute__ ((aligned (64)));
-   uint64_t hash2[8] __attribute__ ((aligned (64)));
-   uint64_t hash3[8] __attribute__ ((aligned (64)));
-   uint64_t vhashX[8*4] __attribute__ ((aligned (64)));
-   uint64_t vhashY[8*4] __attribute__ ((aligned (64)));
+   uint64_t hash0[10] __attribute__ ((aligned (64)));
+   uint64_t hash1[10] __attribute__ ((aligned (64)));
+   uint64_t hash2[10] __attribute__ ((aligned (64)));
+   uint64_t hash3[10] __attribute__ ((aligned (64)));
+   uint64_t vhashX[10*4] __attribute__ ((aligned (64)));
+   uint64_t vhashY[10*4] __attribute__ ((aligned (64)));
   uint64_t *vhashA, *vhashB;
   tt10_4way_ctx_holder ctx __attribute__ ((aligned (64)));
   uint32_t dataLen = 64;
--- a/algo/x12/x12-4way.c
+++ b/algo/x12/x12-4way.c
@@ -108,7 +108,7 @@ void x12_4way_hash( void *state, const void *input )
     intrlv_2x128( vhash, hash2, hash3, 512 );
     luffa_2way_init( &ctx.luffa, 512 );
     luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     intrlv_2x128( hash2, hash3, vhash, 512 );
+     dintrlv_2x128( hash2, hash3, vhash, 512 );

     // 8 Cubehash
     cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
--- a/algo/x13/x13bcd-4way.c
+++ b/algo/x13/x13bcd-4way.c
@@ -12,7 +12,6 @@
 #include "algo/skein/skein-hash-4way.h"
 #include "algo/jh/jh-hash-4way.h"
 #include "algo/keccak/keccak-hash-4way.h"
-//#include "algo/luffa/luffa-hash-2way.h"
 #include "algo/cubehash/cubehash_sse2.h"
 #include "algo/shavite/sph_shavite.h"
 #include "algo/simd/simd-hash-2way.h"
@@ -28,7 +27,6 @@ typedef struct {
    skein512_4way_context   skein;
    jh512_4way_context      jh;
    keccak512_4way_context  keccak;
-//    luffa_2way_context      luffa;
    cubehashParam           cube;
    sph_shavite512_context  shavite;
    simd_2way_context       simd;
@@ -49,7 +47,6 @@ void init_x13bcd_4way_ctx()
     skein512_4way_init( &x13bcd_4way_ctx.skein );
     jh512_4way_init( &x13bcd_4way_ctx.jh );
     keccak512_4way_init( &x13bcd_4way_ctx.keccak );
-//     luffa_2way_init( &x13bcd_4way_ctx.luffa, 512 );
     cubehashInit( &x13bcd_4way_ctx.cube, 512, 16, 32 );
     sph_shavite512_init( &x13bcd_4way_ctx.shavite );
     simd_2way_init( &x13bcd_4way_ctx.simd, 512 );
@@ -72,8 +69,6 @@ void x13bcd_4way_hash( void *state, const void *input )
     // Blake
     memcpy( &ctx.blake, &x13bcd_ctx_mid, sizeof(x13bcd_ctx_mid) );
     blake512_4way( &ctx.blake, input + (64<<2), 16 );
-
-//     blake512_4way( &ctx.blake, input, 80 );
     blake512_4way_close( &ctx.blake, vhash );

     // Bmw
@@ -127,17 +122,6 @@ void x13bcd_4way_hash( void *state, const void *input )
     sm3_4way_close( &ctx.sm3, sm3_vhash );
     dintrlv_4x32( hash0, hash1, hash2, hash3, sm3_vhash, 512 );

-/*     
-     // Luffa
-     intrlv_2x128( vhash, hash0, hash1, 512 );
-     luffa_2way_update_close( &ctx.luffa, vhash, vhash, 64 );
-     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 );
-     dintrlv_2x128( hash2, hash3, vhash, 512 );
-*/
-     
     // Cubehash
     cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*) hash0, 64 );
     memcpy( &ctx.cube, &x13bcd_4way_ctx.cube, sizeof(cubehashParam) );
@@ -185,26 +169,6 @@ void x13bcd_4way_hash( void *state, const void *input )
     update_final_echo( &ctx.echo, (BitSequence *)hash3,
                       (const BitSequence *) hash3, 512 );

-/*
-     intrlv_4x32( vhash, hash0, hash1, hash2, hash3, 512 );
-
-     // SM3 parallel 32 bit
-     uint32_t sm3_vhash[32*4] __attribute__ ((aligned (64)));
-     memset( sm3_vhash, 0, sizeof sm3_vhash );
-     uint32_t sm3_hash0[32] __attribute__ ((aligned (32)));
-     memset( sm3_hash0, 0, sizeof sm3_hash0 );
-     uint32_t sm3_hash1[32] __attribute__ ((aligned (32)));
-     memset( sm3_hash1, 0, sizeof sm3_hash1 );
-     uint32_t sm3_hash2[32] __attribute__ ((aligned (32)));
-     memset( sm3_hash2, 0, sizeof sm3_hash2 );
-     uint32_t sm3_hash3[32] __attribute__ ((aligned (32)));
-     memset( sm3_hash3, 0, sizeof sm3_hash3 );
-
-     sm3_4way( &ctx.sm3, vhash, 64 );
-     sm3_4way_close( &ctx.sm3, sm3_vhash );
-     dintrlv_4x32( hash0, hash1, hash2, hash3, sm3_vhash, 512 );
-*/
-
     // Hamsi parallel 4x32x2
     intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
     hamsi512_4way( &ctx.hamsi, vhash, 64 );
--- a/algo/x16/x16r-4way.c
+++ b/algo/x16/x16r-4way.c
@@ -275,34 +275,31 @@ int scanhash_x16r_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 endiandata[20] __attribute__((aligned(64)));
+   uint32_t bedata1[2] __attribute__((aligned(64)));
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
   const uint32_t Htarg = ptarget[7];
   const uint32_t first_nonce = pdata[19];
   uint32_t n = first_nonce;
-   int thr_id = mythr->id;  // thr_id arg is deprecated
    __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
+   int thr_id = mythr->id;
   volatile uint8_t *restart = &(work_restart[thr_id].restart);

-   casti_m256i( endiandata, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
-   casti_m256i( endiandata, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
-   casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
-
-   if ( s_ntime != endiandata[17] )
-   {
-      uint32_t ntime = swab32(pdata[17]);
-      x16_r_s_getAlgoString( (const uint8_t*) (&endiandata[1]), hashOrder );
-      s_ntime = ntime;
-      if ( opt_debug && !thr_id )
-              applog( LOG_DEBUG, "hash order %s (%08x)", hashOrder, ntime );
-   }
-
   if ( opt_benchmark )
      ptarget[7] = 0x0cff;

-   uint64_t *edata = (uint64_t*)endiandata;
-   intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
+
+   bedata1[0] = bswap_32( pdata[1] );
+   bedata1[1] = bswap_32( pdata[2] );
+   const uint32_t ntime = bswap_32( pdata[17] );
+   if ( s_ntime != ntime )
+   {
+      x16_r_s_getAlgoString( (const uint8_t*)bedata1, hashOrder );
+      s_ntime = ntime;
+      if ( opt_debug && !thr_id )
+              applog( LOG_DEBUG, "hash order %s (%08x)", hashOrder, ntime );
+   }

   do
   {
@@ -312,14 +309,15 @@ int scanhash_x16r_4way( struct work *work, uint32_t max_nonce,
      x16r_4way_hash( hash, vdata );
      pdata[19] = n;

-      for ( int i = 0; i < 4; i++ )  if ( (hash+(i<<3))[7] <= Htarg )
-      if( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark )
+      for ( int i = 0; i < 4; i++ )
+      if ( unlikely( (hash+(i<<3))[7] <= Htarg ) )
+      if( likely( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark ) )
      {
         pdata[19] = n+i;
         submit_lane_solution( work, hash+(i<<3), mythr, i );
      }
      n += 4;
-   } while ( (  n < max_nonce ) && !(*restart) );
+   } while ( likely( ( n < max_nonce ) && !(*restart) ) );

   *hashes_done = n - first_nonce + 1;
   return 0;
--- a/algo/x16/x16rt-4way.c
+++ b/algo/x16/x16rt-4way.c
@@ -24,7 +24,6 @@
 #include "algo/sha/sha-hash-4way.h"

 static __thread uint32_t s_ntime = UINT32_MAX;
-static __thread bool s_implemented = false;
 static __thread char hashOrder[X16R_HASH_FUNC_COUNT + 1] = { 0 };

 union _x16rt_4way_context_overlay
@@ -64,26 +63,8 @@ void x16rt_4way_hash( void* output, const void* input )

   dintrlv_4x64( hash0, hash1, hash2, hash3, input, 640 );
 
-/*
-   void *in = (void*) input;
-   uint32_t *in32 = (uint32_t*) hash0;
-   uint32_t ntime = in32[17];
-   if ( s_ntime == UINT32_MAX )
-   {
-      uint32_t _ALIGN(64) timeHash[8];
-      x16rt_getTimeHash(ntime, &timeHash);
-      x16rt_getAlgoString(&timeHash[0], hashOrder);
-   }
-*/
-
   // Input data is both 64 bit interleaved (input)
-   // and deinterleaved in inp0-3.
-   // If First function uses 64 bit data it is not required to interleave inp
-   // first. It may use the inerleaved data dmost convenient, ie 4way 64 bit.
-   // All other functions assume data is deinterleaved in hash0-3
-   // All functions must exit with data deinterleaved in hash0-3.
-   // Alias in0-3 points to either inp0-3 or hash0-3 according to
-   // its hashOrder position. Size is also set accordingly.
+   // and deinterleaved in inp0-3. First function has no need re-interleave.
   for ( int i = 0; i < 16; i++ )
   {
      const char elem = hashOrder[i];
@@ -290,44 +271,31 @@ int scanhash_x16rt_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 endiandata[20] __attribute__((aligned(64)));
   uint32_t _ALIGN(64) timeHash[4*8];
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
   const uint32_t Htarg = ptarget[7];
   const uint32_t first_nonce = pdata[19];
   uint32_t n = first_nonce;
-   int thr_id = mythr->id;  // thr_id arg is deprecated
+   int thr_id = mythr->id;  
    __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
   volatile uint8_t *restart = &(work_restart[thr_id].restart);

-   casti_m256i( endiandata, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
-   casti_m256i( endiandata, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
-   casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
+   if ( opt_benchmark )
+      ptarget[7] = 0x0cff;

-   uint32_t ntime = swab32( pdata[17] );
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
+   
+   uint32_t ntime = bswap_32( pdata[17] );
   if ( s_ntime != ntime )
   {
      x16rt_getTimeHash( ntime, &timeHash );
      x16rt_getAlgoString( &timeHash[0], hashOrder );
      s_ntime = ntime;
-      s_implemented = true;
      if ( opt_debug && !thr_id )
          applog( LOG_INFO, "hash order: %s time: (%08x) time hash: (%08x)",
                               hashOrder, ntime, timeHash );
   }
-   if ( !s_implemented )
-   {
-      applog( LOG_WARNING, "s not implemented");
-      sleep(1);
-      return 0;
-   }
-
-   if ( opt_benchmark )
-      ptarget[7] = 0x0cff;
-
-   uint64_t *edata = (uint64_t*)endiandata;
-   intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );

   do
   {
--- a/algo/x16/x16rv2-4way.c
+++ b/algo/x16/x16rv2-4way.c
@@ -331,35 +331,32 @@ int scanhash_x16rv2_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 endiandata[20] __attribute__((aligned(64)));
+   uint32_t bedata1[2] __attribute__((aligned(64)));
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
   const uint32_t Htarg = ptarget[7];
   const uint32_t first_nonce = pdata[19];
   uint32_t n = first_nonce;
-   int thr_id = mythr->id;  // thr_id arg is deprecated
+   int thr_id = mythr->id; 
    __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
   volatile uint8_t *restart = &(work_restart[thr_id].restart);

-   casti_m256i( endiandata, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
-   casti_m256i( endiandata, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
-   casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
+   if ( opt_benchmark )
+      ptarget[7] = 0x0fff;
+   
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );

-   if ( s_ntime != endiandata[17] )
+   bedata1[0] = bswap_32( pdata[1] );
+   bedata1[1] = bswap_32( pdata[2] );
+   const uint32_t ntime = bswap_32(pdata[17]);
+   if ( s_ntime != ntime )
   {
-      uint32_t ntime = swab32(pdata[17]);
-      x16_r_s_getAlgoString( (const uint8_t*) (&endiandata[1]), hashOrder );
+      x16_r_s_getAlgoString( (const uint8_t*)bedata1, hashOrder );
      s_ntime = ntime;
      if ( opt_debug && !thr_id )
              applog( LOG_DEBUG, "hash order %s (%08x)", hashOrder, ntime );
   }

-   if ( opt_benchmark )
-      ptarget[7] = 0x0cff;
-
-   uint64_t *edata = (uint64_t*)endiandata;
-   intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
-
   do
   {
      *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
@@ -368,14 +365,15 @@ int scanhash_x16rv2_4way( struct work *work, uint32_t max_nonce,
      x16rv2_4way_hash( hash, vdata );
      pdata[19] = n;

-      for ( int i = 0; i < 4; i++ )  if ( (hash+(i<<3))[7] <= Htarg )
-      if( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark )
+      for ( int i = 0; i < 4; i++ )
+      if ( unlikely( (hash+(i<<3))[7] <= Htarg ) )
+      if( likely( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark ) )
      {
         pdata[19] = n+i;
         submit_lane_solution( work, hash+(i<<3), mythr, i );
      }
      n += 4;
-   } while ( (  n < max_nonce ) && !(*restart) );
+   } while ( likely( (  n < max_nonce ) && !(*restart) ) );

   *hashes_done = n - first_nonce + 1;
   return 0;
--- a/algo/x16/x21s-4way.c
+++ b/algo/x16/x21s-4way.c
@@ -368,7 +368,7 @@ int scanhash_x21s_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 endiandata[20] __attribute__((aligned(64)));
+   uint32_t bedata1[2] __attribute__((aligned(64)));
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
   const uint32_t Htarg = ptarget[7];
@@ -378,25 +378,22 @@ int scanhash_x21s_4way( struct work *work, uint32_t max_nonce,
    __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
   volatile uint8_t *restart = &(work_restart[thr_id].restart);

-   casti_m256i( endiandata, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
-   casti_m256i( endiandata, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
-   casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
+   if ( opt_benchmark )
+    ptarget[7] = 0x0cff;
+ 
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );

-   if ( s_ntime != endiandata[17] )
+   bedata1[0] = bswap_32( pdata[1] );
+   bedata1[1] = bswap_32( pdata[2] );
+   uint32_t ntime = bswap_32( pdata[17] );
+   if ( s_ntime != ntime )
   {
-      uint32_t ntime = swab32(pdata[17]);
-      x16_r_s_getAlgoString( (const uint8_t*) (&endiandata[1]), hashOrder );
+      x16_r_s_getAlgoString( (const uint8_t*)bedata1, hashOrder );
      s_ntime = ntime;
      if ( opt_debug && !thr_id )
              applog( LOG_DEBUG, "hash order %s (%08x)", hashOrder, ntime );
   }

-   if ( opt_benchmark )
-      ptarget[7] = 0x0cff;
-
-   uint64_t *edata = (uint64_t*)endiandata;
-   intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
-
   do
   {
      *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
--- a/algo/x17/sonoa-4way.c
+++ b/algo/x17/sonoa-4way.c
@@ -803,52 +803,40 @@ void sonoa_4way_hash( void *state, const void *input )
     haval256_5_4way_close( &ctx.haval, state );
 }

-int scanhash_sonoa_4way( struct work *work, uint32_t max_nonce,
+int scanhash_sonoa_4way( struct work *work, const uint32_t max_nonce,
 	            uint64_t *hashes_done, struct thr_info *mythr )
 {
     uint32_t hash[4*16] __attribute__ ((aligned (64)));
     uint32_t vdata[24*4] __attribute__ ((aligned (64)));
     uint32_t lane_hash[8] __attribute__ ((aligned (32)));
-     uint32_t *hash7 = &(hash[7<<2]);
+     uint32_t *hash7 = &( hash[7<<2] );
     uint32_t *pdata = work->data;
-     uint32_t *ptarget = work->target;
-     uint32_t n = pdata[19];
+     const uint32_t *ptarget = work->target;
     const uint32_t first_nonce = pdata[19];
-     __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
     const uint32_t Htarg = ptarget[7];
-     int thr_id = mythr->id;
-     uint64_t htmax[] = {          0,        0xF,       0xFF,
-                               0xFFF,     0xFFFF, 0x10000000  };
-     uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
-                          0xFFFFF000, 0xFFFF0000,          0  };
+     uint32_t n = first_nonce;
+     __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
+     const int thr_id = mythr->id;

-     // Need big endian data
     mm256_bswap32_intrlv80_4x64( vdata, pdata );
-     for ( int m=0; m < 6; m++ ) if ( Htarg <= htmax[m] )
+     do
     {
-        uint32_t mask = masks[m];
-        do
+        *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
+                _mm256_set_epi32( n+3, 0,n+2, 0,n+1, 0, n, 0 ) ), *noncev );
+        sonoa_4way_hash( hash, vdata );
+
+        for ( int lane = 0; lane < 4; lane++ )
+        if ( unlikely( hash7[ lane ] <= Htarg ) )
        {
-           *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
-                             _mm256_set_epi32( n+3, 0,n+2, 0,n+1, 0, n, 0 ) ),
-                                                *noncev );
-           sonoa_4way_hash( hash, vdata );
-
-           for ( int lane = 0; lane < 4; lane++ )
-           if ( ( ( hash7[ lane ] & mask ) == 0 ) )
+           extr_lane_4x32( lane_hash, hash, lane, 256 );
+           if ( likely( fulltest( lane_hash, ptarget ) && !opt_benchmark ) )
           {
-              extr_lane_4x32( lane_hash, hash, lane, 256 );
-              if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
-              {
-                 pdata[19] = n + lane;
-                 submit_lane_solution( work, lane_hash, mythr, lane );
-              }
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
           }
-           n += 4;
-        } while ( ( n < max_nonce - 4 ) && !work_restart[thr_id].restart );
-        break;
-     }
-
+        }
+        n += 4;
+     } while ( ( n < max_nonce - 4 ) && !work_restart[thr_id].restart );
     *hashes_done = n - first_nonce + 1;
     return 0;
 }
--- a/algo/x17/x17-4way.c
+++ b/algo/x17/x17-4way.c
@@ -205,50 +205,40 @@ void x17_4way_hash( void *state, const void *input )
 int scanhash_x17_4way( struct work *work, uint32_t max_nonce,
                       uint64_t *hashes_done, struct thr_info *mythr )
 {
-     uint32_t hash[4*16] __attribute__ ((aligned (64)));
-     uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-     uint32_t lane_hash[8] __attribute__ ((aligned (32)));
-     uint32_t *hash7 = &(hash[7<<2]);
-     uint32_t *pdata = work->data;
-     const uint32_t *ptarget = work->target;
-     const uint32_t first_nonce = pdata[19];
-     __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
-     uint32_t n = first_nonce;
-     const int thr_id = mythr->id;
-     const uint32_t Htarg = ptarget[7];
-     uint64_t htmax[] = {          0,        0xF,       0xFF,
-                               0xFFF,     0xFFFF, 0x10000000  };
-     uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
-                          0xFFFFF000, 0xFFFF0000,          0  };
+   uint32_t hash[4*16] __attribute__ ((aligned (64)));
+   uint32_t vdata[24*4] __attribute__ ((aligned (64)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash7 = &(hash[7<<2]);
+   uint32_t *pdata = work->data;
+   const uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
+   uint32_t n = first_nonce;
+   const int thr_id = mythr->id;
+   const uint32_t Htarg = ptarget[7];

-     // Need big endian data
-     mm256_bswap32_intrlv80_4x64( vdata, pdata );
-     for ( int m = 0; m < 6; m++ ) if ( Htarg <= htmax[m] )
-     {
-        const uint32_t mask = masks[ m ];
-        do
-        {
-           *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
-	              _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
-           x17_4way_hash( hash, vdata );
+   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 );
+      x17_4way_hash( hash, vdata );

-           for ( int lane = 0; lane < 4; lane++ )
-           if ( ( hash7[ lane ] & mask ) == 0 )
-           {
-              extr_lane_4x32( lane_hash, hash, lane, 256 );
-              if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
-              {
-                 pdata[19] = n + lane;
-                 submit_lane_solution( work, lane_hash, mythr, lane );
-              }
-           }
-           n += 4;
-        } while ( ( n < max_nonce - 4 ) && !work_restart[thr_id].restart );
-        break;
-     }
+      for ( int lane = 0; lane < 4; lane++ )
+      if unlikely( ( hash7[ lane ] <= Htarg ) )
+      {
+         extr_lane_4x32( lane_hash, hash, lane, 256 );
+         if ( likely( fulltest( lane_hash, ptarget ) && !opt_benchmark ) )
+         {
+            pdata[19] = n + lane;
+            submit_lane_solution( work, lane_hash, mythr, lane );
+         }
+      }
+      n += 4;
+   } while ( likely( ( n < max_nonce - 4 ) && !work_restart[thr_id].restart ) );

-     *hashes_done = n - first_nonce + 1;
-     return 0;
+   *hashes_done = n - first_nonce + 1;
+   return 0;
 }

 #endif
--- a/algo/x22/x22i-4way.c
+++ b/algo/x22/x22i-4way.c
@@ -0,0 +1,311 @@
+#include "x22i-gate.h"
+
+#if defined(X22I_4WAY)
+
+#include "algo/blake/blake-hash-4way.h"
+#include "algo/bmw/bmw-hash-4way.h"
+#include "algo/echo/aes_ni/hash_api.h"
+#include "algo/groestl/aes_ni/hash-groestl.h"
+#include "algo/skein/skein-hash-4way.h"
+#include "algo/jh/jh-hash-4way.h"
+#include "algo/keccak/keccak-hash-4way.h"
+#include "algo/luffa/luffa-hash-2way.h"
+#include "algo/cubehash/cube-hash-2way.h"
+#include "algo/shavite/shavite-hash-2way.h"
+#include "algo/simd/simd-hash-2way.h"
+#include "algo/shavite/sph_shavite.h"
+#include "algo/hamsi/hamsi-hash-4way.h"
+#include "algo/fugue/sph_fugue.h"
+#include "algo/shabal/shabal-hash-4way.h"
+#include "algo/whirlpool/sph_whirlpool.h"
+#include "algo/sha/sha-hash-4way.h"
+#include "algo/haval/haval-hash-4way.h"
+#include "algo/tiger/sph_tiger.h"
+#include "algo/lyra2/lyra2.h"
+#include "algo/gost/sph_gost.h"
+#include "algo/swifftx/swifftx.h"
+
+union _x22i_4way_ctx_overlay
+{
+    blake512_4way_context   blake;
+    bmw512_4way_context     bmw;
+    hashState_groestl       groestl;
+    hashState_echo          echo;
+    skein512_4way_context   skein;
+    jh512_4way_context      jh;
+    keccak512_4way_context  keccak;
+    luffa_2way_context      luffa;
+    cube_2way_context       cube;
+    shavite512_2way_context shavite;
+    simd_2way_context       simd;
+    hamsi512_4way_context   hamsi;
+    sph_fugue512_context    fugue;
+    shabal512_4way_context  shabal;
+    sph_whirlpool_context   whirlpool;
+    sha512_4way_context     sha512;
+    haval256_5_4way_context haval;
+    sph_tiger_context       tiger;
+    sph_gost512_context     gost;
+    sha256_4way_context     sha256;
+};
+typedef union _x22i_4way_ctx_overlay x22i_ctx_overlay;
+
+void x22i_4way_hash( void *output, const void *input )
+{
+   uint64_t hash0[8*4] __attribute__ ((aligned (64)));
+   uint64_t hash1[8*4] __attribute__ ((aligned (64)));
+   uint64_t hash2[8*4] __attribute__ ((aligned (64)));
+   uint64_t hash3[8*4] __attribute__ ((aligned (64)));
+   uint64_t vhash[8*4] __attribute__ ((aligned (64)));
+   uint64_t vhashA[8*4] __attribute__ ((aligned (64)));
+   uint64_t vhashB[8*4] __attribute__ ((aligned (64)));
+
+//   unsigned char hash[64 * 4] __attribute__((aligned(64))) = {0};
+   unsigned char hashA0[64]    __attribute__((aligned(64))) = {0};
+   unsigned char hashA1[64]    __attribute__((aligned(32))) = {0};
+   unsigned char hashA2[64]    __attribute__((aligned(32))) = {0};
+   unsigned char hashA3[64]    __attribute__((aligned(32))) = {0};
+   x22i_ctx_overlay ctx;
+
+   blake512_4way_init( &ctx.blake );
+   blake512_4way( &ctx.blake, input, 80 );
+   blake512_4way_close( &ctx.blake, vhash );
+
+   bmw512_4way_init( &ctx.bmw );
+   bmw512_4way( &ctx.bmw, vhash, 64 );
+   bmw512_4way_close( &ctx.bmw, vhash );
+
+   dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
+   
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)hash0,
+                                  (const char*)hash0, 512 );
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)hash1,
+                                  (const char*)hash1, 512 );
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)hash2,
+                                  (const char*)hash2, 512 );
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)hash3,
+                                  (const char*)hash3, 512 );
+
+   intrlv_4x64_512( vhash, hash0, hash1, hash2, hash3 );
+
+   skein512_4way_init( &ctx.skein );
+   skein512_4way( &ctx.skein, vhash, 64 );
+   skein512_4way_close( &ctx.skein, vhash );
+
+   jh512_4way_init( &ctx.jh );
+   jh512_4way( &ctx.jh, vhash, 64 );
+   jh512_4way_close( &ctx.jh, vhash );
+
+   keccak512_4way_init( &ctx.keccak );
+   keccak512_4way( &ctx.keccak, vhash, 64 );
+   keccak512_4way_close( &ctx.keccak, vhash );
+
+   rintrlv_4x64_2x128( vhashA, vhashB, vhash, 512 );
+
+   luffa_2way_init( &ctx.luffa, 512 );
+   luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, 64 );
+   luffa_2way_init( &ctx.luffa, 512 );
+   luffa_2way_update_close( &ctx.luffa, vhashB, vhashB, 64 );
+
+   cube_2way_init( &ctx.cube, 512, 16, 32 );
+   cube_2way_update_close( &ctx.cube, vhashA, vhashA, 64 );
+   cube_2way_init( &ctx.cube, 512, 16, 32 );
+   cube_2way_update_close( &ctx.cube, vhashB, vhashB, 64 );
+
+   shavite512_2way_init( &ctx.shavite );
+   shavite512_2way_update_close( &ctx.shavite, vhashA, vhashA, 64 );
+   shavite512_2way_init( &ctx.shavite );
+   shavite512_2way_update_close( &ctx.shavite, vhashB, vhashB, 64 );
+
+   simd_2way_init( &ctx.simd, 512 );
+   simd_2way_update_close( &ctx.simd, vhashA, vhashA, 512 );
+   simd_2way_init( &ctx.simd, 512 );
+   simd_2way_update_close( &ctx.simd, vhashB, 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,
+                            (const BitSequence*)hash0, 512 );
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)hash1,
+                            (const BitSequence*)hash1, 512 );
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)hash2,
+                            (const BitSequence*)hash2, 512 );
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)hash3,
+                            (const BitSequence*)hash3, 512 );
+
+
+   intrlv_4x64_512( vhash, hash0, hash1, hash2, hash3 );
+
+   hamsi512_4way_init( &ctx.hamsi );
+   hamsi512_4way( &ctx.hamsi, vhash, 64 );
+   hamsi512_4way_close( &ctx.hamsi, vhash );
+
+   dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
+
+   sph_fugue512_init( &ctx.fugue );
+   sph_fugue512( &ctx.fugue, hash0, 64 );
+   sph_fugue512_close( &ctx.fugue, hash0 );
+   sph_fugue512_init( &ctx.fugue );
+   sph_fugue512( &ctx.fugue, hash1, 64 );
+   sph_fugue512_close( &ctx.fugue, hash1 );
+   sph_fugue512_init( &ctx.fugue );
+   sph_fugue512( &ctx.fugue, hash2, 64 );
+   sph_fugue512_close( &ctx.fugue, hash2 );
+   sph_fugue512_init( &ctx.fugue );
+   sph_fugue512( &ctx.fugue, hash3, 64 );
+   sph_fugue512_close( &ctx.fugue, hash3 );
+
+   intrlv_4x32_512( vhash, hash0, hash1, hash2, hash3 );
+
+   shabal512_4way_init( &ctx.shabal );
+   shabal512_4way( &ctx.shabal, vhash, 64 );
+   shabal512_4way_close( &ctx.shabal, vhash );
+
+   dintrlv_4x32_512( &hash0[8], &hash1[8], &hash2[8], &hash3[8], vhash );
+
+   sph_whirlpool_init( &ctx.whirlpool );
+   sph_whirlpool( &ctx.whirlpool, &hash0[8], 64 );
+   sph_whirlpool_close( &ctx.whirlpool, &hash0[16] );
+   sph_whirlpool_init( &ctx.whirlpool );
+   sph_whirlpool( &ctx.whirlpool, &hash1[8], 64 );
+   sph_whirlpool_close( &ctx.whirlpool, &hash1[16] );
+   sph_whirlpool_init( &ctx.whirlpool );
+   sph_whirlpool( &ctx.whirlpool, &hash2[8], 64 );
+   sph_whirlpool_close( &ctx.whirlpool, &hash2[16] );
+   sph_whirlpool_init( &ctx.whirlpool );
+   sph_whirlpool( &ctx.whirlpool, &hash3[8], 64 );
+   sph_whirlpool_close( &ctx.whirlpool, &hash3[16] );
+
+   intrlv_4x64_512( vhash, &hash0[16], &hash1[16], &hash2[16], &hash3[16] );
+
+   sha512_4way_init( &ctx.sha512 );
+   sha512_4way( &ctx.sha512, vhash, 64 );
+   sha512_4way_close( &ctx.sha512, vhash );
+
+   dintrlv_4x64_512( &hash0[24], &hash1[24], &hash2[24], &hash3[24], vhash );
+
+//	InitializeSWIFFTX();
+	ComputeSingleSWIFFTX((unsigned char*)hash0, (unsigned char*)hashA0);
+   ComputeSingleSWIFFTX((unsigned char*)hash1, (unsigned char*)hashA1);
+   ComputeSingleSWIFFTX((unsigned char*)hash2, (unsigned char*)hashA2);
+   ComputeSingleSWIFFTX((unsigned char*)hash3, (unsigned char*)hashA3);
+
+   intrlv_4x32_512( vhashA, hashA0, hashA1, hashA2, hashA3 );
+
+   memset( vhash, 0, 64*4 );
+
+   haval256_5_4way_init( &ctx.haval );
+   haval256_5_4way( &ctx.haval, vhashA, 64 );
+   haval256_5_4way_close( &ctx.haval, vhash );
+
+   dintrlv_4x32_512( hash0, hash1, hash2, hash3, vhash );
+     
+	memset( hashA0, 0, 64 );
+   memset( hashA1, 0, 64 );
+   memset( hashA2, 0, 64 );
+   memset( hashA3, 0, 64 );
+
+   sph_tiger_init(&ctx.tiger);
+   sph_tiger (&ctx.tiger, (const void*) hash0, 64);
+   sph_tiger_close(&ctx.tiger, (void*) hashA0);
+   sph_tiger_init(&ctx.tiger);
+   sph_tiger (&ctx.tiger, (const void*) hash1, 64);
+   sph_tiger_close(&ctx.tiger, (void*) hashA1);
+   sph_tiger_init(&ctx.tiger);
+   sph_tiger (&ctx.tiger, (const void*) hash2, 64);
+   sph_tiger_close(&ctx.tiger, (void*) hashA2);
+   sph_tiger_init(&ctx.tiger);
+	sph_tiger (&ctx.tiger, (const void*) hash3, 64);
+	sph_tiger_close(&ctx.tiger, (void*) hashA3);
+
+	memset( hash0, 0, 64 );
+   memset( hash1, 0, 64 );
+   memset( hash2, 0, 64 );
+   memset( hash3, 0, 64 );
+
+   LYRA2RE( (void*) hash0, 32, (const void*) hashA0, 32, (const void*) hashA0,
+            32, 1, 4, 4 );
+   LYRA2RE( (void*) hash1, 32, (const void*) hashA1, 32, (const void*) hashA1,
+            32, 1, 4, 4 );
+   LYRA2RE( (void*) hash2, 32, (const void*) hashA2, 32, (const void*) hashA2,
+            32, 1, 4, 4 );
+   LYRA2RE( (void*) hash3, 32, (const void*) hashA3, 32, (const void*) hashA3,
+            32, 1, 4, 4 );
+
+   sph_gost512_init(&ctx.gost);
+   sph_gost512 (&ctx.gost, (const void*) hash0, 64);
+   sph_gost512_close(&ctx.gost, (void*) hash0);
+   sph_gost512_init(&ctx.gost);
+   sph_gost512 (&ctx.gost, (const void*) hash1, 64);
+   sph_gost512_close(&ctx.gost, (void*) hash1);
+   sph_gost512_init(&ctx.gost);
+   sph_gost512 (&ctx.gost, (const void*) hash2, 64);
+   sph_gost512_close(&ctx.gost, (void*) hash2);
+	sph_gost512_init(&ctx.gost);
+	sph_gost512 (&ctx.gost, (const void*) hash3, 64);
+	sph_gost512_close(&ctx.gost, (void*) hash3);
+
+   intrlv_4x32_512( vhash, hash0, hash1, hash2, hash3 );
+
+   sha256_4way_init( &ctx.sha256 );
+   sha256_4way( &ctx.sha256, vhash, 64 );
+   sha256_4way_close( &ctx.sha256, output );
+   
+//	memcpy(output, hash, 32);
+}
+
+
+int scanhash_x22i_4way( struct work* work, uint32_t max_nonce,
+                   uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t hash[4*16] __attribute__ ((aligned (64)));
+   uint32_t vdata[24*4] __attribute__ ((aligned (64)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash7 = &(hash[7<<2]);
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
+   uint32_t n = first_nonce;
+   const int thr_id = mythr->id;
+   const uint32_t Htarg = ptarget[7];
+
+   if (opt_benchmark)
+      ((uint32_t*)ptarget)[7] = 0x08ff;
+   
+   InitializeSWIFFTX();
+
+   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 );
+      x22i_4way_hash( hash, vdata );
+
+      for ( int lane = 0; lane < 4; lane++ )
+      if unlikely( ( hash7[ lane ] <= Htarg ) )
+      {
+         extr_lane_4x32( lane_hash, hash, lane, 256 );
+         if ( likely( fulltest( lane_hash, ptarget ) && !opt_benchmark ) )
+         {
+            pdata[19] = n + lane;
+            submit_lane_solution( work, lane_hash, mythr, lane );
+         }
+      }
+      n += 4;
+   } while ( likely( ( n < max_nonce - 4 ) && !work_restart[thr_id].restart ) );
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+#endif  // X22I_4WAY
--- a/algo/x22/x22i-gate.c
+++ b/algo/x22/x22i-gate.c
@@ -0,0 +1,28 @@
+#include "x22i-gate.h"
+
+bool register_x22i_algo( algo_gate_t* gate )
+{
+#if defined (X22I_4WAY)
+  gate->scanhash  = (void*)&scanhash_x22i_4way;
+  gate->hash      = (void*)&x22i_4way_hash;
+#else
+  gate->scanhash  = (void*)&scanhash_x22i;
+  gate->hash      = (void*)&x22i_hash;
+#endif
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | SHA_OPT;
+  return true;
+};
+
+bool register_x25x_algo( algo_gate_t* gate )
+{
+#if defined (X22I_4WAY)
+  gate->scanhash  = (void*)&scanhash_x25x_4way;
+  gate->hash      = (void*)&x25x_4way_hash;
+#else
+  gate->scanhash  = (void*)&scanhash_x25x;
+  gate->hash      = (void*)&x25x_hash;
+#endif
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | SHA_OPT;
+  return true;
+};
+
--- a/algo/x22/x22i-gate.h
+++ b/algo/x22/x22i-gate.h
@@ -0,0 +1,35 @@
+#ifndef X22I_GATE_H__
+#define X22I_GATE_H__ 1
+
+#include "algo-gate-api.h"
+#include "simd-utils.h"
+#include <stdint.h>
+#include <unistd.h>
+
+#if defined(__AVX2__) && defined(__AES__)
+  #define X22I_4WAY
+#endif
+
+bool register_x22i__algo( algo_gate_t* gate );
+
+#if defined(X22I_4WAY)
+
+void x22i_4way_hash( void *state, const void *input );
+int scanhash_x22i_4way( struct work *work, uint32_t max_nonce,
+                        uint64_t *hashes_done, struct thr_info *mythr );
+
+void x25x_4way_hash( void *state, const void *input );
+int scanhash_x25x_4way( struct work *work, uint32_t max_nonce,
+                        uint64_t *hashes_done, struct thr_info *mythr );
+
+#endif
+
+void x22i_hash( void *state, const void *input );
+int scanhash_x22i( struct work *work, uint32_t max_nonce,
+                   uint64_t *hashes_done, struct thr_info *mythr );
+
+void x25x_hash( void *state, const void *input );
+int scanhash_x25x( struct work *work, uint32_t max_nonce,
+                   uint64_t *hashes_done, struct thr_info *mythr );
+
+#endif  // X22I_GATE_H__
--- a/algo/x22/x22i.c
+++ b/algo/x22/x22i.c
@@ -0,0 +1,202 @@
+#include "algo/blake/sph_blake.h"
+#include "algo/bmw/sph_bmw.h"
+#if defined(__AES__)
+  #include "algo/echo/aes_ni/hash_api.h"
+  #include "algo/groestl/aes_ni/hash-groestl.h"
+#else
+  #include "algo/groestl/sph_groestl.h"
+  #include "algo/echo/sph_echo.h"
+#endif
+#include "algo/skein/sph_skein.h"
+#include "algo/jh/sph_jh.h"
+#include "algo/keccak/sph_keccak.h"
+#include "algo/luffa/luffa_for_sse2.h"
+#include "algo/cubehash/cubehash_sse2.h"
+#include "algo/shavite/sph_shavite.h"
+#include "algo/simd/nist.h"
+#include "algo/hamsi/sph_hamsi.h"
+#include "algo/fugue/sph_fugue.h"
+#include "algo/shabal/sph_shabal.h"
+#include "algo/whirlpool/sph_whirlpool.h"
+#include <openssl/sha.h>
+#include "algo/haval/sph-haval.h"
+#include "algo/tiger/sph_tiger.h"
+#include "algo/lyra2/lyra2.h"
+#include "algo/gost/sph_gost.h"
+#include "algo/swifftx/swifftx.h"
+#include "x22i-gate.h"
+
+union _x22i_context_overlay
+{
+        sph_blake512_context blake;
+        sph_bmw512_context bmw;
+#if defined(__AES__)
+        hashState_groestl       groestl;
+        hashState_echo          echo;
+#else
+        sph_groestl512_context  groestl;
+        sph_echo512_context     echo;
+#endif
+        sph_jh512_context       jh;
+        sph_keccak512_context   keccak;
+        sph_skein512_context    skein;
+        hashState_luffa         luffa;
+        cubehashParam           cube;
+        sph_shavite512_context  shavite;
+        hashState_sd            simd;
+        sph_hamsi512_context    hamsi;
+        sph_fugue512_context    fugue;
+        sph_shabal512_context   shabal;
+        sph_whirlpool_context   whirlpool;
+        SHA512_CTX              sha512;
+        sph_haval256_5_context  haval;
+        sph_tiger_context       tiger;
+        sph_gost512_context     gost;
+        SHA256_CTX              sha256;
+};
+typedef union _x22i_context_overlay x22i_context_overlay;
+
+void x22i_hash( void *output, const void *input )
+{
+   unsigned char hash[64 * 4] __attribute__((aligned(64))) = {0};
+   unsigned char hash2[65]    __attribute__((aligned(64))) = {0};
+   x22i_context_overlay ctx;
+
+	sph_blake512_init(&ctx.blake);
+	sph_blake512(&ctx.blake, input, 80);
+	sph_blake512_close(&ctx.blake, hash);
+
+	sph_bmw512_init(&ctx.bmw);
+	sph_bmw512(&ctx.bmw, (const void*) hash, 64);
+	sph_bmw512_close(&ctx.bmw, hash);
+
+#if defined(__AES__)
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)hash,
+                                  (const char*)hash, 512 );
+#else
+   sph_groestl512_init( &ctx.groestl );
+   sph_groestl512( &ctx.groestl, hash, 64 );
+   sph_groestl512_close( &ctx.groestl, hash );
+#endif
+   
+	sph_skein512_init(&ctx.skein);
+	sph_skein512(&ctx.skein, (const void*) hash, 64);
+	sph_skein512_close(&ctx.skein, hash);
+
+	sph_jh512_init(&ctx.jh);
+	sph_jh512(&ctx.jh, (const void*) hash, 64);
+	sph_jh512_close(&ctx.jh, hash);
+
+	sph_keccak512_init(&ctx.keccak);
+	sph_keccak512(&ctx.keccak, (const void*) hash, 64);
+	sph_keccak512_close(&ctx.keccak, hash);
+
+   init_luffa( &ctx.luffa, 512 );
+   update_and_final_luffa( &ctx.luffa, (BitSequence*)hash,
+                                (const BitSequence*)hash, 64 );
+
+   cubehashInit( &ctx.cube, 512, 16, 32 );
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash,
+                              (const byte*)hash, 64 );
+
+	sph_shavite512_init(&ctx.shavite);
+	sph_shavite512(&ctx.shavite, (const void*) hash, 64);
+	sph_shavite512_close(&ctx.shavite, hash);
+
+   init_sd( &ctx.simd, 512 );
+   update_final_sd( &ctx.simd, (BitSequence*)hash,
+                         (const BitSequence*)hash, 512 );
+
+#if defined(__AES__)
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)hash,
+                            (const BitSequence*)hash, 512 );
+#else
+   sph_echo512_init( &ctx.echo );
+   sph_echo512( &ctx.echo, hash, 64 );
+   sph_echo512_close( &ctx.echo, hash );
+#endif
+
+	sph_hamsi512_init(&ctx.hamsi);
+	sph_hamsi512(&ctx.hamsi, (const void*) hash, 64);
+	sph_hamsi512_close(&ctx.hamsi, hash);
+
+	sph_fugue512_init(&ctx.fugue);
+	sph_fugue512(&ctx.fugue, (const void*) hash, 64);
+	sph_fugue512_close(&ctx.fugue, hash);
+
+	sph_shabal512_init(&ctx.shabal);
+	sph_shabal512(&ctx.shabal, (const void*) hash, 64);
+	sph_shabal512_close(&ctx.shabal, &hash[64]);
+
+	sph_whirlpool_init(&ctx.whirlpool);
+	sph_whirlpool (&ctx.whirlpool, (const void*) &hash[64], 64);
+	sph_whirlpool_close(&ctx.whirlpool, &hash[128]);
+
+   SHA512_Init( &ctx.sha512 );
+   SHA512_Update(  &ctx.sha512, (const void*) &hash[128], 64);
+   SHA512_Final( (void*) &hash[192], &ctx.sha512 );
+
+	ComputeSingleSWIFFTX((unsigned char*)hash, (unsigned char*)hash2);
+
+	memset(hash, 0, 64);
+	sph_haval256_5_init(&ctx.haval);
+	sph_haval256_5(&ctx.haval,(const void*) hash2, 64);
+	sph_haval256_5_close(&ctx.haval,hash);
+
+	memset(hash2, 0, 64);
+	sph_tiger_init(&ctx.tiger);
+	sph_tiger (&ctx.tiger, (const void*) hash, 64);
+	sph_tiger_close(&ctx.tiger, (void*) hash2);
+
+	memset(hash, 0, 64);
+	LYRA2RE((void*) hash, 32, (const void*) hash2, 32, (const void*) hash2, 32, 1, 4, 4);
+
+	sph_gost512_init(&ctx.gost);
+	sph_gost512 (&ctx.gost, (const void*) hash, 64);
+	sph_gost512_close(&ctx.gost, (void*) hash);
+
+   SHA256_Init( &ctx.sha256 );
+   SHA256_Update(  &ctx.sha256, (const void*) hash, 64 );
+   SHA256_Final( (unsigned char*) hash, &ctx.sha256 );
+
+	memcpy(output, hash, 32);
+}
+
+int scanhash_x22i( struct work* work, uint32_t max_nonce,
+                   uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t endiandata[20] __attribute__((aligned(64)));
+   uint32_t hash[8] __attribute__((aligned(64)));
+	uint32_t *pdata = work->data;
+	uint32_t *ptarget = work->target;
+	const uint32_t first_nonce = pdata[19];
+   const uint32_t Htarg = ptarget[7];
+   uint32_t n = first_nonce;
+   const int thr_id = mythr->id;
+
+	if (opt_benchmark)
+		((uint32_t*)ptarget)[7] = 0x08ff;
+
+	for (int k=0; k < 20; k++)
+		be32enc(&endiandata[k], pdata[k]);
+
+   InitializeSWIFFTX();
+
+   do
+   {
+       pdata[19] = ++n;
+       be32enc( &endiandata[19], n );
+
+       x22i_hash( hash, endiandata );
+
+       if ( hash[7] < Htarg )
+       if ( fulltest( hash, ptarget ) && !opt_benchmark )
+           submit_solution( work, hash, mythr );
+    } while ( n < max_nonce && !work_restart[thr_id].restart );
+
+	 *hashes_done = pdata[19] - first_nonce;
+	 return 0;
+}
+
--- a/algo/x22/x25x-4way.c
+++ b/algo/x22/x25x-4way.c
@@ -0,0 +1,402 @@
+#include "x22i-gate.h"
+
+#if defined(X22I_4WAY)
+
+#include "algo/blake/blake-hash-4way.h"
+#include "algo/bmw/bmw-hash-4way.h"
+#include "algo/skein/skein-hash-4way.h"
+#include "algo/jh/jh-hash-4way.h"
+#include "algo/keccak/keccak-hash-4way.h"
+#include "algo/hamsi/hamsi-hash-4way.h"
+#include "algo/shabal/shabal-hash-4way.h"
+#include "algo/sha/sha-hash-4way.h"
+#include "algo/haval/haval-hash-4way.h"
+#include "algo/blake/blake2s-hash-4way.h"
+#include "algo/echo/aes_ni/hash_api.h"
+#include "algo/groestl/aes_ni/hash-groestl.h"
+#include "algo/luffa/luffa_for_sse2.h"
+#include "algo/cubehash/cubehash_sse2.h"
+#include "algo/shavite/sph_shavite.h"
+#include "algo/simd/nist.h"
+#include "algo/fugue/sph_fugue.h"
+#include "algo/whirlpool/sph_whirlpool.h"
+#include "algo/tiger/sph_tiger.h"
+#include "algo/lyra2/lyra2.h"
+#include "algo/gost/sph_gost.h"
+#include "algo/swifftx/swifftx.h"
+#include "algo/panama/sph_panama.h"
+#include "algo/lanehash/lane.h"
+
+union _x25x_4way_ctx_overlay
+{
+    blake512_4way_context   blake;
+    bmw512_4way_context     bmw;
+    hashState_groestl       groestl;
+    hashState_echo          echo;
+    skein512_4way_context   skein;
+    jh512_4way_context      jh;
+    keccak512_4way_context  keccak;
+    hashState_luffa         luffa;
+    cubehashParam           cube;
+    sph_shavite512_context  shavite;
+    hashState_sd            simd;
+    hamsi512_4way_context   hamsi;
+    sph_fugue512_context    fugue;
+    shabal512_4way_context  shabal;
+    sph_whirlpool_context   whirlpool;
+    sha512_4way_context     sha512;
+    haval256_5_4way_context haval;
+    sph_tiger_context       tiger;
+    sph_gost512_context     gost;
+    sha256_4way_context     sha256;
+    sph_panama_context      panama;
+     blake2s_4way_state           blake2s;
+};
+typedef union _x25x_4way_ctx_overlay x25x_4way_ctx_overlay;
+
+void x25x_shuffle( void *hash )
+{
+   // Simple shuffle algorithm, instead of just reversing
+   #define X25X_SHUFFLE_BLOCKS (24 * 64 / 2)
+   #define X25X_SHUFFLE_ROUNDS 12
+
+   static const uint16_t x25x_round_const[X25X_SHUFFLE_ROUNDS] =
+   {
+      0x142c, 0x5830, 0x678c, 0xe08c, 0x3c67, 0xd50d, 0xb1d8, 0xecb2,
+      0xd7ee, 0x6783, 0xfa6c, 0x4b9c
+   };
+
+   uint16_t* block_pointer = (uint16_t*)hash;
+   for ( int r = 0; r < X25X_SHUFFLE_ROUNDS; r++ )
+   {
+      for ( int i = 0; i < X25X_SHUFFLE_BLOCKS; i++ )
+      {
+         uint16_t block_value = block_pointer[ X25X_SHUFFLE_BLOCKS - i - 1 ];
+         block_pointer[i] ^= block_pointer[ block_value % X25X_SHUFFLE_BLOCKS ]
+                                + ( x25x_round_const[r] << (i % 16) );
+      }
+   }
+
+   #undef X25X_SHUFFLE_BLOCKS
+   #undef X25X_SHUFFLE_ROUNDS
+}
+
+void x25x_4way_hash( void *output, const void *input )
+{
+   unsigned char hash0[25][64] __attribute__((aligned(64))) = {0};
+   unsigned char hash1[25][64] __attribute__((aligned(64))) = {0};
+   unsigned char hash2[25][64] __attribute__((aligned(64))) = {0};
+   unsigned char hash3[25][64] __attribute__((aligned(64))) = {0};
+   uint64_t vhash[8*4] __attribute__ ((aligned (64)));
+   unsigned char vhashA[24][64*4] __attribute__ ((aligned (64)));
+   x25x_4way_ctx_overlay ctx __attribute__ ((aligned (64)));
+
+   blake512_4way_init( &ctx.blake );
+   blake512_4way( &ctx.blake, input, 80 );
+   blake512_4way_close( &ctx.blake, vhash );
+   dintrlv_4x64_512( &hash0[0], &hash1[0], &hash2[0], &hash3[0], vhash );
+
+   bmw512_4way_init( &ctx.bmw );
+   bmw512_4way( &ctx.bmw, vhash, 64 );
+   bmw512_4way_close( &ctx.bmw, vhash );
+   dintrlv_4x64_512( &hash0[1], &hash1[1], &hash2[1], &hash3[1], vhash );
+
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)&hash0[2],
+                                  (const char*)&hash0[1], 512 );
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)&hash1[2],
+                                  (const char*)&hash1[1], 512 );
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)&hash2[2],
+                                  (const char*)&hash2[1], 512 );
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)&hash3[2],
+                                  (const char*)&hash3[1], 512 );
+   
+   intrlv_4x64_512( vhash, &hash0[2], &hash1[2], &hash2[2], &hash3[2] );
+
+   skein512_4way_init( &ctx.skein );
+   skein512_4way( &ctx.skein, vhash, 64 );
+   skein512_4way_close( &ctx.skein, vhash );
+   dintrlv_4x64_512( &hash0[3], &hash1[3], &hash2[3], &hash3[3], vhash );
+
+   jh512_4way_init( &ctx.jh );
+   jh512_4way( &ctx.jh, vhash, 64 );
+   jh512_4way_close( &ctx.jh, vhash );
+   dintrlv_4x64_512( &hash0[4], &hash1[4], &hash2[4], &hash3[4], vhash );
+
+   keccak512_4way_init( &ctx.keccak );
+   keccak512_4way( &ctx.keccak, vhash, 64 );
+   keccak512_4way_close( &ctx.keccak, vhash );
+   dintrlv_4x64_512( &hash0[5], &hash1[5], &hash2[5], &hash3[5], vhash );
+   
+   init_luffa( &ctx.luffa, 512 );
+   update_and_final_luffa( &ctx.luffa, (BitSequence*)&hash0[6],
+                                (const BitSequence*)&hash0[5], 64 );
+   init_luffa( &ctx.luffa, 512 );
+   update_and_final_luffa( &ctx.luffa, (BitSequence*)&hash1[6],
+                                (const BitSequence*)&hash1[5], 64 );
+   init_luffa( &ctx.luffa, 512 );
+   update_and_final_luffa( &ctx.luffa, (BitSequence*)&hash2[6],
+                                (const BitSequence*)&hash2[5], 64 );
+   init_luffa( &ctx.luffa, 512 );
+   update_and_final_luffa( &ctx.luffa, (BitSequence*)&hash3[6],
+                                (const BitSequence*)&hash3[5], 64 );
+
+   cubehashInit( &ctx.cube, 512, 16, 32 );
+   cubehashUpdateDigest( &ctx.cube, (byte*) &hash0[7],
+                              (const byte*)&hash0[6], 64 );
+   cubehashInit( &ctx.cube, 512, 16, 32 );
+   cubehashUpdateDigest( &ctx.cube, (byte*) &hash1[7],
+                              (const byte*)&hash1[6], 64 );
+   cubehashInit( &ctx.cube, 512, 16, 32 );
+   cubehashUpdateDigest( &ctx.cube, (byte*) &hash2[7],
+                              (const byte*)&hash2[6], 64 );
+   cubehashInit( &ctx.cube, 512, 16, 32 );
+   cubehashUpdateDigest( &ctx.cube, (byte*) &hash3[7],
+                              (const byte*)&hash3[6], 64 );
+
+	sph_shavite512_init(&ctx.shavite);
+	sph_shavite512(&ctx.shavite, (const void*) &hash0[7], 64);
+	sph_shavite512_close(&ctx.shavite, &hash0[8]);
+   sph_shavite512_init(&ctx.shavite);
+   sph_shavite512(&ctx.shavite, (const void*) &hash1[7], 64);
+   sph_shavite512_close(&ctx.shavite, &hash1[8]);
+   sph_shavite512_init(&ctx.shavite);
+   sph_shavite512(&ctx.shavite, (const void*) &hash2[7], 64);
+   sph_shavite512_close(&ctx.shavite, &hash2[8]);
+   sph_shavite512_init(&ctx.shavite);
+   sph_shavite512(&ctx.shavite, (const void*) &hash3[7], 64);
+   sph_shavite512_close(&ctx.shavite, &hash3[8]);
+
+   init_sd( &ctx.simd, 512 );
+   update_final_sd( &ctx.simd, (BitSequence*)&hash0[9],
+                         (const BitSequence*)&hash0[8], 512 );
+   init_sd( &ctx.simd, 512 );
+   update_final_sd( &ctx.simd, (BitSequence*)&hash1[9],
+                         (const BitSequence*)&hash1[8], 512 );
+   init_sd( &ctx.simd, 512 );
+   update_final_sd( &ctx.simd, (BitSequence*)&hash2[9],
+                         (const BitSequence*)&hash2[8], 512 );
+   init_sd( &ctx.simd, 512 );
+   update_final_sd( &ctx.simd, (BitSequence*)&hash3[9],
+                         (const BitSequence*)&hash3[8], 512 );
+
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)&hash0[10],
+                            (const BitSequence*)&hash0[9], 512 );
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)&hash1[10],
+                            (const BitSequence*)&hash1[9], 512 );
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)&hash2[10],
+                            (const BitSequence*)&hash2[9], 512 );
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)&hash3[10],
+                            (const BitSequence*)&hash3[9], 512 );
+
+   intrlv_4x64_512( vhash, &hash0[10], &hash1[10], &hash2[10], &hash3[10] );
+
+   hamsi512_4way_init( &ctx.hamsi );
+   hamsi512_4way( &ctx.hamsi, vhash, 64 );
+   hamsi512_4way_close( &ctx.hamsi, vhash );
+   dintrlv_4x64_512( &hash0[11], &hash1[11], &hash2[11], &hash3[11], vhash );
+
+	sph_fugue512_init(&ctx.fugue);
+	sph_fugue512(&ctx.fugue, (const void*) &hash0[11], 64);
+	sph_fugue512_close(&ctx.fugue, &hash0[12]);
+   sph_fugue512_init(&ctx.fugue);
+   sph_fugue512(&ctx.fugue, (const void*) &hash1[11], 64);
+   sph_fugue512_close(&ctx.fugue, &hash1[12]);
+   sph_fugue512_init(&ctx.fugue);
+   sph_fugue512(&ctx.fugue, (const void*) &hash2[11], 64);
+   sph_fugue512_close(&ctx.fugue, &hash2[12]);
+   sph_fugue512_init(&ctx.fugue);
+   sph_fugue512(&ctx.fugue, (const void*) &hash3[11], 64);
+   sph_fugue512_close(&ctx.fugue, &hash3[12]);
+
+   intrlv_4x32_512( vhash, &hash0[12], &hash1[12], &hash2[12], &hash3[12] );
+
+   shabal512_4way_init( &ctx.shabal );
+   shabal512_4way( &ctx.shabal, vhash, 64 );
+   shabal512_4way_close( &ctx.shabal, vhash );
+   dintrlv_4x32_512( &hash0[13], &hash1[13], &hash2[13], &hash3[13], vhash );
+
+	sph_whirlpool_init(&ctx.whirlpool);
+	sph_whirlpool (&ctx.whirlpool, (const void*) &hash0[13], 64);
+	sph_whirlpool_close(&ctx.whirlpool, &hash0[14]);
+   sph_whirlpool_init(&ctx.whirlpool);
+   sph_whirlpool (&ctx.whirlpool, (const void*) &hash1[13], 64);
+   sph_whirlpool_close(&ctx.whirlpool, &hash1[14]);
+   sph_whirlpool_init(&ctx.whirlpool);
+   sph_whirlpool (&ctx.whirlpool, (const void*) &hash2[13], 64);
+   sph_whirlpool_close(&ctx.whirlpool, &hash2[14]);
+   sph_whirlpool_init(&ctx.whirlpool);
+   sph_whirlpool (&ctx.whirlpool, (const void*) &hash3[13], 64);
+   sph_whirlpool_close(&ctx.whirlpool, &hash3[14]);
+
+   intrlv_4x64_512( vhash, &hash0[14], &hash1[14], &hash2[14], &hash3[14] );
+
+   sha512_4way_init( &ctx.sha512 );
+   sha512_4way( &ctx.sha512, vhash, 64 );
+   sha512_4way_close( &ctx.sha512, vhash );
+   dintrlv_4x64_512( &hash0[15], &hash1[15], &hash2[15], &hash3[15], vhash );
+
+
+   ComputeSingleSWIFFTX((unsigned char*)&hash0[12], (unsigned char*)&hash0[16]);
+   ComputeSingleSWIFFTX((unsigned char*)&hash1[12], (unsigned char*)&hash1[16]);
+   ComputeSingleSWIFFTX((unsigned char*)&hash2[12], (unsigned char*)&hash2[16]);
+   ComputeSingleSWIFFTX((unsigned char*)&hash3[12], (unsigned char*)&hash3[16]);
+
+   intrlv_4x32_512( &vhashA, &hash0[16], &hash1[16], &hash2[16], &hash3[16] );
+
+   memset( vhash, 0, 64*4 );
+   
+   haval256_5_4way_init( &ctx.haval );
+   haval256_5_4way( &ctx.haval, vhashA, 64 );
+   haval256_5_4way_close( &ctx.haval, vhash );
+   dintrlv_4x32_512( &hash0[17], &hash1[17], &hash2[17], &hash3[17], vhash );
+
+	sph_tiger_init(&ctx.tiger);
+	sph_tiger (&ctx.tiger, (const void*) &hash0[17], 64);
+	sph_tiger_close(&ctx.tiger, (void*) &hash0[18]);
+   sph_tiger_init(&ctx.tiger);
+   sph_tiger (&ctx.tiger, (const void*) &hash1[17], 64);
+   sph_tiger_close(&ctx.tiger, (void*) &hash1[18]);
+   sph_tiger_init(&ctx.tiger);
+   sph_tiger (&ctx.tiger, (const void*) &hash2[17], 64);
+   sph_tiger_close(&ctx.tiger, (void*) &hash2[18]);
+   sph_tiger_init(&ctx.tiger);
+   sph_tiger (&ctx.tiger, (const void*) &hash3[17], 64);
+   sph_tiger_close(&ctx.tiger, (void*) &hash3[18]);
+
+	LYRA2RE( (void*)&hash0[19], 32, (const void*)&hash0[18], 32,
+            (const void*)&hash0[18], 32, 1, 4, 4 );
+   LYRA2RE( (void*)&hash1[19], 32, (const void*)&hash1[18], 32,
+            (const void*)&hash1[18], 32, 1, 4, 4 );
+   LYRA2RE( (void*)&hash2[19], 32, (const void*)&hash2[18], 32,
+            (const void*)&hash2[18], 32, 1, 4, 4 );
+   LYRA2RE( (void*)&hash3[19], 32, (const void*)&hash3[18], 32,
+            (const void*)&hash3[18], 32, 1, 4, 4 );
+
+	sph_gost512_init(&ctx.gost);
+	sph_gost512 (&ctx.gost, (const void*) &hash0[19], 64);
+	sph_gost512_close(&ctx.gost, (void*) &hash0[20]);
+   sph_gost512_init(&ctx.gost);
+   sph_gost512 (&ctx.gost, (const void*) &hash1[19], 64);
+   sph_gost512_close(&ctx.gost, (void*) &hash1[20]);
+   sph_gost512_init(&ctx.gost);
+   sph_gost512 (&ctx.gost, (const void*) &hash2[19], 64);
+   sph_gost512_close(&ctx.gost, (void*) &hash2[20]);
+   sph_gost512_init(&ctx.gost);
+   sph_gost512 (&ctx.gost, (const void*) &hash3[19], 64);
+   sph_gost512_close(&ctx.gost, (void*) &hash3[20]);
+
+   intrlv_4x32_512( vhashA, &hash0[20], &hash1[20], &hash2[20], &hash3[20] );
+   memset( vhash, 0, 64*4 );
+
+   sha256_4way_init( &ctx.sha256 );
+   sha256_4way( &ctx.sha256, vhashA, 64 );
+   sha256_4way_close( &ctx.sha256, vhash );
+   dintrlv_4x32_512( &hash0[21], &hash1[21], &hash2[21], &hash3[21], vhash );
+
+   sph_panama_init(&ctx.panama);
+   sph_panama (&ctx.panama, (const void*) &hash0[21], 64 );
+   sph_panama_close(&ctx.panama, (void*) &hash0[22]);
+   sph_panama_init(&ctx.panama);
+   sph_panama (&ctx.panama, (const void*) &hash1[21], 64 );
+   sph_panama_close(&ctx.panama, (void*) &hash1[22]);
+   sph_panama_init(&ctx.panama);
+   sph_panama (&ctx.panama, (const void*) &hash2[21], 64 );
+   sph_panama_close(&ctx.panama, (void*) &hash2[22]);
+   sph_panama_init(&ctx.panama);
+   sph_panama (&ctx.panama, (const void*) &hash3[21], 64 );
+   sph_panama_close(&ctx.panama, (void*) &hash3[22]);
+
+   laneHash(512, (const BitSequence*)&hash0[22], 512, (BitSequence*)&hash0[23]);
+   laneHash(512, (const BitSequence*)&hash1[22], 512, (BitSequence*)&hash1[23]);
+   laneHash(512, (const BitSequence*)&hash2[22], 512, (BitSequence*)&hash2[23]);
+   laneHash(512, (const BitSequence*)&hash3[22], 512, (BitSequence*)&hash3[23]);
+
+   x25x_shuffle( hash0 );
+   x25x_shuffle( hash1 );
+   x25x_shuffle( hash2 );
+   x25x_shuffle( hash3 );
+
+   intrlv_4x32_512( &vhashA[ 0], &hash0[ 0], &hash1[ 0], &hash2[ 0], &hash3[ 0] );
+   intrlv_4x32_512( &vhashA[ 1], &hash0[ 1], &hash1[ 1], &hash2[ 1], &hash3[ 1] );
+   intrlv_4x32_512( &vhashA[ 2], &hash0[ 2], &hash1[ 2], &hash2[ 2], &hash3[ 2] );
+   intrlv_4x32_512( &vhashA[ 3], &hash0[ 3], &hash1[ 3], &hash2[ 3], &hash3[ 3] );
+   intrlv_4x32_512( &vhashA[ 4], &hash0[ 4], &hash1[ 4], &hash2[ 4], &hash3[ 4] );
+   intrlv_4x32_512( &vhashA[ 5], &hash0[ 5], &hash1[ 5], &hash2[ 5], &hash3[ 5] );
+   intrlv_4x32_512( &vhashA[ 6], &hash0[ 6], &hash1[ 6], &hash2[ 6], &hash3[ 6] );
+   intrlv_4x32_512( &vhashA[ 7], &hash0[ 7], &hash1[ 7], &hash2[ 7], &hash3[ 7] );
+   intrlv_4x32_512( &vhashA[ 8], &hash0[ 8], &hash1[ 8], &hash2[ 8], &hash3[ 8] );
+   intrlv_4x32_512( &vhashA[ 9], &hash0[ 9], &hash1[ 9], &hash2[ 9], &hash3[ 9] );
+   intrlv_4x32_512( &vhashA[10], &hash0[10], &hash1[10], &hash2[10], &hash3[10] );
+   intrlv_4x32_512( &vhashA[11], &hash0[11], &hash1[11], &hash2[11], &hash3[11] );
+   intrlv_4x32_512( &vhashA[12], &hash0[12], &hash1[12], &hash2[12], &hash3[12] );
+   intrlv_4x32_512( &vhashA[13], &hash0[13], &hash1[13], &hash2[13], &hash3[13] );
+   intrlv_4x32_512( &vhashA[14], &hash0[14], &hash1[14], &hash2[14], &hash3[14] );
+   intrlv_4x32_512( &vhashA[15], &hash0[15], &hash1[15], &hash2[15], &hash3[15] );
+   intrlv_4x32_512( &vhashA[16], &hash0[16], &hash1[16], &hash2[16], &hash3[16] );
+   intrlv_4x32_512( &vhashA[17], &hash0[17], &hash1[17], &hash2[17], &hash3[17] );
+   intrlv_4x32_512( &vhashA[18], &hash0[18], &hash1[18], &hash2[18], &hash3[18] );
+   intrlv_4x32_512( &vhashA[19], &hash0[19], &hash1[19], &hash2[19], &hash3[19] );
+   intrlv_4x32_512( &vhashA[20], &hash0[20], &hash1[20], &hash2[20], &hash3[20] );
+   intrlv_4x32_512( &vhashA[21], &hash0[21], &hash1[21], &hash2[21], &hash3[21] );
+   intrlv_4x32_512( &vhashA[22], &hash0[22], &hash1[22], &hash2[22], &hash3[22] );
+   intrlv_4x32_512( &vhashA[23], &hash0[23], &hash1[23], &hash2[23], &hash3[23] );
+
+   blake2s_4way_init( &ctx.blake2s, 32 );
+   blake2s_4way_full_blocks( &ctx.blake2s, vhash, vhashA, 64*24 );
+
+   dintrlv_4x32( &hash0[24], &hash1[24], &hash2[24], &hash3[24], vhash, 256 );
+     
+	memcpy(output,    &hash0[24], 32);
+   memcpy(output+32, &hash1[24], 32);
+   memcpy(output+64, &hash2[24], 32);
+   memcpy(output+96, &hash3[24], 32);
+}
+
+int scanhash_x25x_4way( struct work* work, uint32_t max_nonce,
+                   uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t hash[4*16] __attribute__ ((aligned (64)));
+   uint32_t vdata[24*4] __attribute__ ((aligned (64)));
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
+   uint32_t n = first_nonce;
+   const int thr_id = mythr->id;
+   const uint32_t Htarg = ptarget[7];
+
+   if (opt_benchmark)
+      ((uint32_t*)ptarget)[7] = 0x08ff;
+
+   InitializeSWIFFTX();
+
+   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 );
+      x25x_4way_hash( hash, vdata );
+
+      for ( int i = 0; i < 4; i++ )
+      if ( unlikely( (hash+(i<<3))[7] <= Htarg ) )
+      if( likely( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark ) )
+      {
+         pdata[19] = n+i;
+         submit_lane_solution( work, hash+(i<<3), mythr, i );
+      }
+      n += 4;
+   } while ( likely( ( n < max_nonce - 4 ) && !work_restart[thr_id].restart ) );
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+#endif
--- a/algo/x22/x25x.c
+++ b/algo/x22/x25x.c
@@ -0,0 +1,236 @@
+#include "x22i-gate.h"
+#include "algo/blake/sph_blake.h"
+#include "algo/bmw/sph_bmw.h"
+#if defined(__AES__)
+  #include "algo/echo/aes_ni/hash_api.h"
+  #include "algo/groestl/aes_ni/hash-groestl.h"
+#else
+  #include "algo/groestl/sph_groestl.h"
+  #include "algo/echo/sph_echo.h"
+#endif
+#include "algo/skein/sph_skein.h"
+#include "algo/jh/sph_jh.h"
+#include "algo/keccak/sph_keccak.h"
+#include "algo/luffa/luffa_for_sse2.h"
+#include "algo/cubehash/cubehash_sse2.h"
+#include "algo/shavite/sph_shavite.h"
+#include "algo/simd/nist.h"
+#include "algo/hamsi/sph_hamsi.h"
+#include "algo/fugue/sph_fugue.h"
+#include "algo/shabal/sph_shabal.h"
+#include "algo/whirlpool/sph_whirlpool.h"
+#include <openssl/sha.h>
+#include "algo/haval/sph-haval.h"
+#include "algo/tiger/sph_tiger.h"
+#include "algo/lyra2/lyra2.h"
+#include "algo/gost/sph_gost.h"
+#include "algo/swifftx/swifftx.h"
+#include "algo/blake/sph-blake2s.h"
+#include "algo/panama/sph_panama.h"
+#include "algo/lanehash/lane.h"
+
+union _x25x_context_overlay
+{
+        sph_blake512_context    blake;
+        sph_bmw512_context      bmw;
+#if defined(__AES__)
+        hashState_groestl       groestl;
+        hashState_echo          echo;
+#else
+        sph_groestl512_context  groestl;
+        sph_echo512_context     echo;
+#endif
+        sph_jh512_context       jh;
+        sph_keccak512_context   keccak;
+        sph_skein512_context    skein;
+        hashState_luffa         luffa;
+        cubehashParam           cube;
+        sph_shavite512_context  shavite;
+        hashState_sd            simd;
+        sph_hamsi512_context    hamsi;
+        sph_fugue512_context    fugue;
+        sph_shabal512_context   shabal;
+        sph_whirlpool_context   whirlpool;
+        SHA512_CTX              sha512;
+        sph_haval256_5_context  haval;
+        sph_tiger_context       tiger;
+        sph_gost512_context     gost;
+        SHA256_CTX              sha256;
+        sph_panama_context      panama;
+        blake2s_state           blake2s;
+};
+typedef union _x25x_context_overlay x25x_context_overlay;
+
+void x25x_hash( void *output, const void *input )
+{
+   unsigned char hash[25][64] __attribute__((aligned(64))) = {0};
+   x25x_context_overlay ctx;
+
+	sph_blake512_init(&ctx.blake);
+	sph_blake512(&ctx.blake, input, 80);
+	sph_blake512_close(&ctx.blake, &hash[0] );
+
+	sph_bmw512_init(&ctx.bmw);
+	sph_bmw512(&ctx.bmw, (const void*) &hash[0], 64);
+	sph_bmw512_close(&ctx.bmw, &hash[1]);
+
+#if defined(__AES__)
+   init_groestl( &ctx.groestl, 64 );
+   update_and_final_groestl( &ctx.groestl, (char*)&hash[2],
+                                  (const char*)&hash[1], 512 );
+#else
+   sph_groestl512_init( &ctx.groestl );
+   sph_groestl512( &ctx.groestl, &hash[1], 64 );
+   sph_groestl512_close( &ctx.groestl, &hash[2] );
+#endif
+   
+	sph_skein512_init(&ctx.skein);
+	sph_skein512(&ctx.skein, (const void*) &hash[2], 64);
+	sph_skein512_close(&ctx.skein, &hash[3]);
+
+	sph_jh512_init(&ctx.jh);
+	sph_jh512(&ctx.jh, (const void*) &hash[3], 64);
+	sph_jh512_close(&ctx.jh, &hash[4]);
+
+	sph_keccak512_init(&ctx.keccak);
+	sph_keccak512(&ctx.keccak, (const void*) &hash[4], 64);
+	sph_keccak512_close(&ctx.keccak, &hash[5]);
+
+   init_luffa( &ctx.luffa, 512 );
+   update_and_final_luffa( &ctx.luffa, (BitSequence*)&hash[6],
+                                (const BitSequence*)&hash[5], 64 );
+
+   cubehashInit( &ctx.cube, 512, 16, 32 );
+   cubehashUpdateDigest( &ctx.cube, (byte*) &hash[7],
+                              (const byte*)&hash[6], 64 );
+
+	sph_shavite512_init(&ctx.shavite);
+	sph_shavite512(&ctx.shavite, (const void*) &hash[7], 64);
+	sph_shavite512_close(&ctx.shavite, &hash[8]);
+
+   init_sd( &ctx.simd, 512 );
+   update_final_sd( &ctx.simd, (BitSequence*)&hash[9],
+                         (const BitSequence*)&hash[8], 512 );
+
+#if defined(__AES__)
+   init_echo( &ctx.echo, 512 );
+   update_final_echo ( &ctx.echo, (BitSequence*)&hash[10],
+                            (const BitSequence*)&hash[9], 512 );
+#else
+   sph_echo512_init( &ctx.echo );
+   sph_echo512( &ctx.echo, &hash[9], 64 );
+   sph_echo512_close( &ctx.echo, &hash[10] );
+#endif
+
+	sph_hamsi512_init(&ctx.hamsi);
+	sph_hamsi512(&ctx.hamsi, (const void*) &hash[10], 64);
+	sph_hamsi512_close(&ctx.hamsi, &hash[11]);
+
+	sph_fugue512_init(&ctx.fugue);
+	sph_fugue512(&ctx.fugue, (const void*) &hash[11], 64);
+	sph_fugue512_close(&ctx.fugue, &hash[12]);
+
+	sph_shabal512_init(&ctx.shabal);
+	sph_shabal512(&ctx.shabal, (const void*) &hash[12], 64);
+	sph_shabal512_close(&ctx.shabal, &hash[13]);
+
+	sph_whirlpool_init(&ctx.whirlpool);
+	sph_whirlpool (&ctx.whirlpool, (const void*) &hash[13], 64);
+	sph_whirlpool_close(&ctx.whirlpool, &hash[14]);
+
+   SHA512_Init( &ctx.sha512 );
+   SHA512_Update(  &ctx.sha512, (const void*) &hash[14], 64);
+   SHA512_Final( (void*) &hash[15], &ctx.sha512 );
+
+   ComputeSingleSWIFFTX((unsigned char*)&hash[12], (unsigned char*)&hash[16]);
+
+	sph_haval256_5_init(&ctx.haval);
+	sph_haval256_5(&ctx.haval,(const void*) &hash[16], 64);
+	sph_haval256_5_close(&ctx.haval,&hash[17]);
+
+	sph_tiger_init(&ctx.tiger);
+	sph_tiger (&ctx.tiger, (const void*) &hash[17], 64);
+	sph_tiger_close(&ctx.tiger, (void*) &hash[18]);
+
+	LYRA2RE( (void*)&hash[19], 32, (const void*)&hash[18], 32,
+            (const void*)&hash[18], 32, 1, 4, 4 );
+
+	sph_gost512_init(&ctx.gost);
+	sph_gost512 (&ctx.gost, (const void*) &hash[19], 64);
+	sph_gost512_close(&ctx.gost, (void*) &hash[20]);
+
+   SHA256_Init( &ctx.sha256 );
+   SHA256_Update(  &ctx.sha256, (const void*) &hash[20], 64 );
+   SHA256_Final( (unsigned char*) &hash[21], &ctx.sha256 );
+
+   sph_panama_init(&ctx.panama);
+   sph_panama (&ctx.panama, (const void*) &hash[21], 64 );
+   sph_panama_close(&ctx.panama, (void*) &hash[22]);
+
+   laneHash(512, (const BitSequence*) &hash[22], 512, (BitSequence*) &hash[23]);
+
+   // Simple shuffle algorithm, instead of just reversing
+   #define X25X_SHUFFLE_BLOCKS (24 * 64 / 2)
+   #define X25X_SHUFFLE_ROUNDS 12
+
+   static const uint16_t x25x_round_const[X25X_SHUFFLE_ROUNDS] =
+   {
+      0x142c, 0x5830, 0x678c, 0xe08c, 0x3c67, 0xd50d, 0xb1d8, 0xecb2,
+      0xd7ee, 0x6783, 0xfa6c, 0x4b9c
+   };
+
+   uint16_t* block_pointer = (uint16_t*)hash;
+   for ( int r = 0; r < X25X_SHUFFLE_ROUNDS; r++ )
+   {
+      for ( int i = 0; i < X25X_SHUFFLE_BLOCKS; i++ )
+      {
+         uint16_t block_value = block_pointer[ X25X_SHUFFLE_BLOCKS - i - 1 ];
+         block_pointer[i] ^= block_pointer[ block_value % X25X_SHUFFLE_BLOCKS ]
+                                + ( x25x_round_const[r] << (i % 16) );
+      }
+   }
+
+   #undef X25X_SHUFFLE_BLOCKS
+   #undef X25X_SHUFFLE_ROUNDS
+
+   blake2s_simple( (uint8_t*)&hash[24], (const void*)(&hash[0]), 64 * 24 );
+   
+	memcpy(output, &hash[24], 32);
+}
+
+int scanhash_x25x( struct work* work, uint32_t max_nonce,
+                   uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t endiandata[20] __attribute__((aligned(64)));
+   uint32_t hash[8] __attribute__((aligned(64)));
+	uint32_t *pdata = work->data;
+	uint32_t *ptarget = work->target;
+	const uint32_t first_nonce = pdata[19];
+   const uint32_t Htarg = ptarget[7];
+   uint32_t n = first_nonce;
+   const int thr_id = mythr->id;
+
+	if (opt_benchmark)
+		((uint32_t*)ptarget)[7] = 0x08ff;
+
+	for (int k=0; k < 20; k++)
+		be32enc(&endiandata[k], pdata[k]);
+
+   InitializeSWIFFTX();
+
+   do
+   {
+       pdata[19] = ++n;
+       be32enc( &endiandata[19], n );
+
+       x25x_hash( hash, endiandata );
+
+       if ( hash[7] < Htarg )
+       if ( fulltest( hash, ptarget ) && !opt_benchmark )
+           submit_solution( work, hash, mythr );
+    } while ( n < max_nonce && !work_restart[thr_id].restart );
+
+	 *hashes_done = pdata[19] - first_nonce;
+	 return 0;
+}
+
--- a/algo/yespower/.yespower-opt.c.swp
+++ b/algo/yespower/.yespower-opt.c.swp
--- a/algo/yespower/yespower-blake2b.c
+++ b/algo/yespower/yespower-blake2b.c
@@ -49,6 +49,7 @@
 * no slowdown from the prefixes is generally observed on AMD CPUs supporting
 * XOP, some slowdown is sometimes observed on Intel CPUs with AVX.
 */
+/*
 #ifdef __XOP__
 #warning "Note: XOP is enabled.  That's great."
 #elif defined(__AVX__)
@@ -60,6 +61,7 @@
 #else
 #warning "Note: building generic code for non-x86.  That's OK."
 #endif
+*/

 /*
 * The SSE4 code version has fewer instructions than the generic SSE2 version,
--- a/build-allarch.sh
+++ b/build-allarch.sh
@@ -16,7 +16,8 @@ mv cpuminer cpuminer-avx512

 make clean || echo clean
 rm -f config.status
-CFLAGS="-O3 -march=core-avx2 -Wall" ./configure --with-curl
+# GCC 9 doesn't include AES with core-avx2
+CFLAGS="-O3 -march=core-avx2 -maes -Wall" ./configure --with-curl
 make -j 16
 strip -s cpuminer.exe
 mv cpuminer.exe cpuminer-avx2.exe
@@ -25,7 +26,7 @@ mv cpuminer cpuminer-avx2

 make clean || echo clean
 rm -f config.status
-CFLAGS="-O3 -march=corei7-avx -Wall" ./configure --with-curl
+CFLAGS="-O3 -march=corei7-avx -maes -Wall" ./configure --with-curl
 make -j 16
 strip -s cpuminer.exe
 mv cpuminer.exe cpuminer-aes-avx.exe
--- a/build-allarch.sh.bak
+++ b/build-allarch.sh.bak
@@ -0,0 +1,86 @@
+#!/bin/bash
+#
+# This script is not intended for users, it is only used for compile testing
+# during develpment. Howver the information contained my provide cimpilation
+# tips to users.
+
+make distclean || echo clean
+rm -f config.status
+./autogen.sh || echo done
+CFLAGS="-O3 -march=skylake-avx512 -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe cpuminer-avx512.exe
+strip -s cpuminer
+mv cpuminer cpuminer-avx512
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -march=core-avx2 -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe cpuminer-avx2.exe
+strip -s cpuminer
+mv cpuminer cpuminer-avx2
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -march=corei7-avx -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe cpuminer-aes-avx.exe
+strip -s cpuminer
+mv cpuminer cpuminer-aes-avx
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -maes -msse4.2 -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe cpuminer-aes-sse42.exe
+strip -s cpuminer
+mv cpuminer cpuminer-aes-sse42
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -march=corei7 -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe cpuminer-sse42.exe
+strip -s cpuminer
+mv cpuminer cpuminer-sse42
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -march=core2 -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe cpuminer-ssse3.exe
+strip -s cpuminer
+mv cpuminer cpuminer-ssse3
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -msse2 -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe cpuminer-sse2.exe
+strip -s cpuminer
+mv cpuminer cpuminer-sse2
+
+make clean || echo done
+rm -f config.status
+CFLAGS="-O3 -march=znver1 -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe cpuminer-zen.exe
+strip -s cpuminer
+mv cpuminer cpuminer-zen
+
+make clean || echo done
+rm -f config.status
+CFLAGS="-O3 -march=native -Wall" ./configure --with-curl
+make -j 16
+strip -s cpuminer.exe
+strip -s cpuminer
+
--- a/20
+++ b/20
@@ -1,6 +1,6 @@
 #! /bin/sh
 # Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.69 for cpuminer-opt 3.9.9.1.
+# Generated by GNU Autoconf 2.69 for cpuminer-opt 3.10.0.
 #
 #
 # 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='3.9.9.1'
-PACKAGE_STRING='cpuminer-opt 3.9.9.1'
+PACKAGE_VERSION='3.10.0'
+PACKAGE_STRING='cpuminer-opt 3.10.0'
 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 3.9.9.1 to adapt to many kinds of systems.
+\`configure' configures cpuminer-opt 3.10.0 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 3.9.9.1:";;
+     short | recursive ) echo "Configuration of cpuminer-opt 3.10.0:";;
   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 3.9.9.1
+cpuminer-opt configure 3.10.0
 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 3.9.9.1, which was
+It was created by cpuminer-opt $as_me 3.10.0, 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='3.9.9.1'
+ VERSION='3.10.0'


 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 3.9.9.1, which was
+This file was extended by cpuminer-opt $as_me 3.10.0, 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 3.9.9.1
+cpuminer-opt config.status 3.10.0
 configured by $0, generated by GNU Autoconf 2.69,
  with options \\"\$ac_cs_config\\"

--- a/configure.ac
+++ b/configure.ac
@@ -1,4 +1,4 @@
-AC_INIT([cpuminer-opt], [3.9.9.1])
+AC_INIT([cpuminer-opt], [3.10.0])

 AC_PREREQ([2.59c])
 AC_CANONICAL_SYSTEM
--- a/cpu-miner.c
+++ b/cpu-miner.c
@@ -53,6 +53,8 @@
 #if HAVE_SYS_PARAM_H
 #include <sys/param.h>
 #endif
+
+// GCC 9 warning sysctl.h is deprecated
 #include <sys/sysctl.h>
 #endif
 #endif
@@ -157,8 +159,6 @@ uint32_t accepted_share_count = 0;
 uint32_t rejected_share_count = 0;
 uint32_t solved_block_count = 0;
 double *thr_hashrates;
-double *thr_hashcount;
-double global_hashcount = 0;
 double global_hashrate = 0;
 double stratum_diff = 0.;
 double net_diff = 0.;
@@ -875,7 +875,7 @@ static uint32_t last_block_height = 0;
 static double   last_targetdiff = 0.;
 static double   ref_rate_hi = 0.;
 static double   ref_rate_lo = 1e100;
-#if !(defined(__WINDOWS__) || defined(__WIN64))
+#if !(defined(__WINDOWS__) || defined(_WIN64) || defined(_WIN32))
 static uint32_t hi_temp = 0;
 #endif
 //static uint32_t stratum_errors = 0;
@@ -976,10 +976,11 @@ void report_summary_log( bool force )
                       accepts, accepted_share_count );
   applog2( LOG_INFO,"Rejected         %6d       %6d",
                       rejects, rejected_share_count );
-//   applog2( LOG_INFO,"Blocks solved                  %6d",
-//                              solved_block_count );
+   if ( solved_block_count )
+      applog2( LOG_INFO,"Blocks solved                 %6d",
+                         solved_block_count );

-#if !(defined(__WINDOWS__) || defined(__WIN64))
+#if !(defined(__WINDOWS__) || defined(_WIN64) || defined(_WIN32))

   int temp = cpu_temp(0);
   char tempstr[32];
@@ -1004,7 +1005,7 @@ static int share_result( int result, struct work *null_work,
                         const char *reason )
 {
   double share_time = 0., share_ratio = 0.;
-   double hashcount = 0., hashrate = 0.;
+   double hashrate = 0.;
   int latency = 0;
   struct share_stats_t my_stats = {0};
   struct timeval ack_time, latency_tv, et;
@@ -1065,11 +1066,7 @@ static int share_result( int result, struct work *null_work,
   pthread_mutex_lock( &stats_lock );

   for ( int i = 0; i < opt_n_threads; i++ )
-   {
-       hashcount += thr_hashcount[i];
       hashrate += thr_hashrates[i];
-   }
-   global_hashcount = hashcount;
   global_hashrate = hashrate;
   
   if ( result ) 
@@ -1342,8 +1339,8 @@ static bool submit_upstream_work( CURL *curl, struct work *work )
      if ( work->height && work->height <= net_blocks )
      {
         if (opt_debug)
-	    applog(LOG_WARNING, "block %u was already solved", work->height);
-	 return true;
+ 	        applog(LOG_WARNING, "block %u was already solved", work->height);
+	      return true;
      }
   }

@@ -1735,12 +1732,12 @@ err_out:
 	return false;
 }

-// Treat 32 byte hash string as 256 bit integer and convert to double precision
-// floating point number.
+// Convert little endian 256 bit unsigned integer to
+// double precision floating point.
 static inline double u256_to_double( const uint64_t* u )
 {
   const double f = 4294967296.0 * 4294967296.0;  // 2**64
-   return u[0] + f * ( u[1] + f * ( u[2] + f * u[3] ) );
+   return ( ( u[3] * f + u[2] ) * f + u[1] ) * f + u[0];
 }

 void work_set_target_ratio( struct work* work, uint32_t* hash )
@@ -2203,7 +2200,6 @@ static void *miner_thread( void *userdata )
       if ( diff.tv_usec || diff.tv_sec )
       {
          pthread_mutex_lock( &stats_lock );
-          thr_hashcount[thr_id] = hashes_done;
          thr_hashrates[thr_id] =
          hashes_done / ( diff.tv_sec + diff.tv_usec * 1e-6 );
          pthread_mutex_unlock( &stats_lock );
@@ -2232,30 +2228,19 @@ static void *miner_thread( void *userdata )
          }
       }
       // display hashrate
-       if ( !opt_quiet )
+       if ( opt_hash_meter )
       {
-          char hc[16];
          char hr[16];
-          char hc_units[2] = {0,0};
          char hr_units[2] = {0,0};
-          double hashcount;
          double hashrate;
-          if ( opt_hash_meter )
+
+          hashrate  = thr_hashrates[thr_id];
+          if ( hashrate != 0. )
          {
-             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 );
-             }
+             scale_hash_for_display( &hashrate,  hr_units );
+             sprintf( hr, "%.2f", hashrate );
+             applog( LOG_INFO, "CPU #%d: %s %sh/s",
+                               thr_id, hr, hr_units );
          }
       }

@@ -2265,35 +2250,23 @@ static void *miner_thread( void *userdata )
            && thr_id == opt_n_threads - 1 )
       {
          double hashrate  = 0.;
-          double hashcount = 0.;
          for ( i = 0; i < opt_n_threads; i++ )
-          {
              hashrate  += thr_hashrates[i];
-              hashcount += thr_hashcount[i];
-          }
-          if ( hashcount != 0. )
+
+          if ( hashrate != 0. )
          {
-             global_hashcount = hashcount;
             global_hashrate  = hashrate;
             if ( opt_benchmark )
             {
-                char hc[16];
-                char hc_units[2] = {0,0};
                char hr[16];
                char hr_units[2] = {0,0};
-	             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 );
-#if ((defined(_WIN64) || defined(__WINDOWS__)))
-                applog( LOG_NOTICE, "Total: %s %sH, %s %sH/s",
-                                  hc, hc_units, hr, hr_units );
+#if ((defined(_WIN64) || defined(__WINDOWS__)) || defined(_WIN32))
+                applog( LOG_NOTICE, "Total: %s %sH/s", hr, hr_units );
 #else
-                applog( LOG_NOTICE, "Total: %s %sH, %s %sH/s, %dC",
-                         hc, hc_units, hr, hr_units, (uint32_t)cpu_temp(0) );
+                applog( LOG_NOTICE, "Total: %s %sH/s, CPU temp: %dC",
+                                     hr, hr_units, (uint32_t)cpu_temp(0) );
 #endif
             }
 	       }
@@ -2612,10 +2585,14 @@ void std_stratum_gen_work( struct stratum_ctx *sctx, struct work *g_work )
   if ( ( stratum_diff != sctx->job.diff )
     || ( last_block_height != sctx->block_height ) )
   {
-       double hr = global_hashrate;
-       char hr_units[4] = {0};
-       char block_ttf[32];
-       char share_ttf[32];
+       double hr = 0.;
+
+       pthread_mutex_lock( &stats_lock );
+
+       for ( int i = 0; i < opt_n_threads; i++ )
+          hr += thr_hashrates[i];
+       global_hashrate = hr;
+       pthread_mutex_unlock( &stats_lock );
  
       if ( stratum_diff != sctx->job.diff )
          applog( LOG_BLUE, "New stratum difficulty" );
@@ -2626,17 +2603,25 @@ void std_stratum_gen_work( struct stratum_ctx *sctx, struct work *g_work )
       stratum_diff = sctx->job.diff;
       last_block_height = stratum.block_height;
       last_targetdiff = g_work->targetdiff;
-  
-       sprintf_et( block_ttf, net_diff * diff_to_hash / hr );
-       sprintf_et( share_ttf, last_targetdiff * diff_to_hash / hr );
-       scale_hash_for_display ( &hr, hr_units );

       applog2( LOG_INFO, "%s %s block %d", short_url,
                algo_names[opt_algo], stratum.block_height );
       applog2( LOG_INFO, "Diff: net %g, stratum %g, target %g",
                net_diff, stratum_diff, last_targetdiff );
-       applog2( LOG_INFO, "TTF @ %.2f %sh/s: block %s, share %s",
-                hr, hr_units, block_ttf, share_ttf );
+
+       if ( hr > 0. )
+       {
+          char hr_units[4] = {0};
+          char block_ttf[32];
+          char share_ttf[32];
+
+          sprintf_et( block_ttf, net_diff * diff_to_hash / hr );
+          sprintf_et( share_ttf, last_targetdiff * diff_to_hash / hr );
+          scale_hash_for_display ( &hr, hr_units );
+
+          applog2( LOG_INFO, "TTF @ %.2f %sh/s: block %s, share %s",
+                              hr, hr_units, block_ttf, share_ttf );
+       }
   }     
 }

@@ -3356,12 +3341,14 @@ bool check_cpu_capability ()
     bool cpu_has_avx2   = has_avx2();
     bool cpu_has_sha    = has_sha();
     bool cpu_has_avx512 = has_avx512();
+     bool cpu_has_vaes   = has_vaes();
     bool sw_has_aes    = false;
     bool sw_has_sse42  = false;
     bool sw_has_avx    = false;
     bool sw_has_avx2   = false;
     bool sw_has_avx512 = false;
     bool sw_has_sha    = false;
+     bool sw_has_vaes   = false;
     set_t algo_features = algo_gate.optimizations;
     bool algo_has_sse2   = set_incl( SSE2_OPT,    algo_features );
     bool algo_has_aes    = set_incl( AES_OPT,     algo_features );
@@ -3369,12 +3356,14 @@ bool check_cpu_capability ()
     bool algo_has_avx2   = set_incl( AVX2_OPT,    algo_features );
     bool algo_has_avx512 = set_incl( AVX512_OPT,  algo_features );
     bool algo_has_sha    = set_incl( SHA_OPT,     algo_features );
+     bool algo_has_vaes   = set_incl( VAES_OPT,    algo_features );
     bool use_aes;
     bool use_sse2;
     bool use_sse42;
     bool use_avx2;
     bool use_avx512;
     bool use_sha;
+     bool use_vaes;
     bool use_none;

     #ifdef __AES__
@@ -3389,12 +3378,16 @@ bool check_cpu_capability ()
     #ifdef __AVX2__
         sw_has_avx2 = true;
     #endif
-     #if (defined(__AVX512F__) && defined(__AVX51DQF__) && defined(__AVX51BW__) && defined(__AVX512VL__))
+     #if (defined(__AVX512F__) && defined(__AVX512DQ__) && defined(__AVX512BW__) && defined(__AVX512VL__))
         sw_has_avx512 = true;
     #endif
     #ifdef __SHA__
         sw_has_sha = true;
     #endif
+     #ifdef __VAES__
+         sw_has_vaes = true;
+     #endif
+         

 //     #if !((__AES__) || (__SSE2__))
 //         printf("Neither __AES__ nor __SSE2__ defined.\n");
@@ -3421,6 +3414,7 @@ bool check_cpu_capability ()
     if ( cpu_has_avx2   )    printf( " AVX2"   );
     if ( cpu_has_avx512 )    printf( " AVX512" );
     if ( cpu_has_sha    )    printf( " SHA"    );
+     if ( cpu_has_vaes   )    printf( " VAES"   );

     printf(".\nSW features: SSE2");
     if ( sw_has_aes    )     printf( " AES"    );
@@ -3429,18 +3423,20 @@ bool check_cpu_capability ()
     if ( sw_has_avx2   )     printf( " AVX2"   );
     if ( sw_has_avx512 )     printf( " AVX512" );
     if ( sw_has_sha    )     printf( " SHA"    );
+     if ( sw_has_vaes   )     printf( " VAES"   );
    

     printf(".\nAlgo features:");
     if ( algo_features == EMPTY_SET ) printf( " None" );
     else
     {
-        if ( algo_has_sse2   ) printf( " SSE2"    );
-        if ( algo_has_aes    ) printf( " AES"     );
-        if ( algo_has_sse42  ) printf( " SSE4.2"  );
+        if ( algo_has_sse2   ) printf( " SSE2"   );
+        if ( algo_has_aes    ) printf( " AES"    );
+        if ( algo_has_sse42  ) printf( " SSE4.2" );
        if ( algo_has_avx2   ) printf( " AVX2"   );
        if ( algo_has_avx512 ) printf( " AVX512" );
        if ( algo_has_sha    ) printf( " SHA"    );
+        if ( algo_has_vaes   ) printf( " VAES"   );
     }
     printf(".\n");

@@ -3478,8 +3474,9 @@ bool check_cpu_capability ()
     use_avx2   = cpu_has_avx2   && sw_has_avx2   && algo_has_avx2;
     use_avx512 = cpu_has_avx512 && sw_has_avx512 && algo_has_avx512;
     use_sha    = cpu_has_sha    && sw_has_sha    && algo_has_sha;
+     use_vaes   = cpu_has_vaes   && sw_has_vaes   && algo_has_vaes;
     use_none = !( use_sse2 || use_aes || use_sse42 || use_avx512 || use_avx2 ||
-                   use_sha );
+                   use_sha || use_vaes );
      
     // Display best options
     printf( "Start mining with" );
@@ -3724,9 +3721,6 @@ int main(int argc, char *argv[])
 	thr_hashrates = (double *) calloc(opt_n_threads, sizeof(double));
 	if (!thr_hashrates)
 		return 1;
-        thr_hashcount = (double *) calloc(opt_n_threads, sizeof(double));
-        if (!thr_hashcount)
-                return 1;

 	/* init workio thread info */
 	work_thr_id = opt_n_threads;
--- a/miner.h
+++ b/miner.h
@@ -610,6 +610,8 @@ enum algos {
        ALGO_X16S,
        ALGO_X17,
        ALGO_X21S,
+        ALGO_X22I,
+        ALGO_X25X,
        ALGO_XEVAN,
        ALGO_YESCRYPT,
        ALGO_YESCRYPTR8,
@@ -707,6 +709,8 @@ static const char* const algo_names[] = {
        "x16s",
        "x17",
        "x21s",
+        "x22i",
+        "x25x",
        "xevan",
        "yescrypt",
        "yescryptr8",
@@ -871,6 +875,8 @@ Options:\n\
                          x16s\n\
                          x17\n\
                          x21s\n\
+                          x22i\n\
+                          x25x\n\
                          xevan         Bitsend (BSD)\n\
                          yescrypt      Globalboost-Y (BSTY)\n\
                          yescryptr8    BitZeny (ZNY)\n\
--- a/simd-utils.h
+++ b/simd-utils.h
@@ -105,52 +105,36 @@
 //    Ex: mm256_ror1x64_128 rotates each 128 bit lane of a 256 bit vector
 //        right by 64 bits.
 //
-//   Some random thoughts about macros and inline functions, the pros and
-//   cons, when to use them, etc:
+// Vector constants
 //
-// Macros are very convenient and efficient for statement functions.
-// Macro args are passed by value and modifications are seen by the caller.
-// Macros should not generally call regular functions unless it is for a
-// special purpose such overloading a function name.
-// Statement function macros that return a value should not end in ";"
-// Statement function macros that return a value and don't modify input args
-// may be used in function arguments and expressions.
-// Macro args used in expressions should be protected ex: (x)+1
-// Macros force inlining, function inlining can be overridden by the compiler.
-// Inline functions are preferred when multiple statements or local variables
-// are needed.
-// The compiler can't do any syntax checking or type checking of args making
-// macros difficult to debug.
-// Although it is technically posssible to access the callers data without
-// they being passed as arguments it is good practice to always define
-// arguments even if they have the same name.
+// Vector constants are a big problem because they technically don't exist.
+// All vectors used as constants either reside in memory or must be genererated
+// at run time at significant cost. The cost of generating a constant
+// increases non-linearly with the number of vector elements. A 4 element
+// vector costs between 7 and 11 clocks to generate, an 8 element vector
+// is 15-25 clocks. There are also additional clock due to data dependency
+// stalls.
 //
-// General guidelines for inline functions:
+// Vector constants are often used as control indexes for permute, blend, etc,
+// where generating the index can be over 90% of the operation. This is
+// where the problem occurs. An instruction that only requires one to 3
+// clocks needs may times more just to build the index argument.
 //
-// Inline functions should not have loops, it defeats the purpose of inlining.
-// Inline functions should be short, the benefit is lost and the memory cost
-// increases if the function is referenced often.
-// Inline functions may call other functions, inlined or not. It is convenient
-// for wrapper functions whether or not the wrapped function is itself inlined.
-// Care should be taken when unrolling loops that contain calls to inlined
-// functions that may be large.
-// Large code blocks used only once may use function inlining to
-// improve high level code readability without the penalty of function
-// overhead.
+// There is very little a programmer can do to avoid the worst case scenarios.
+// Smaller integers can be merged to form 64 bit integers, and vectors with
+// repeated elements can be generated more efficiently but they have limited
+// benefit and limited application.
 //
-// These utilities avoid memory accesses and assume data is in a register
-// argument. Vector constants, in particular are generated with opcodes instead
-// of being read from memory.
+// If a vector constant is to be used repeatedly it is better to define a local
+// variable to generate the constant only once.
 //
-// The utilities defined here make use features like register aliasing
-// to optimize operations. Many operations have specialized versions as
-// well as more generic versions. It is preferable to use a specialized
-// version whenever possible as they can take advantage of certain
-// optimizations not available to the generic version. The generic
-// version will often have an additional argument used is some extra
-// calculations.
-// 
-///////////////////////////////////////////////////////
+// If a sequence of constants is to be used it can be more efficient to
+// use arithmetic with already existing constants to generate new ones.
+//
+// ex: const __m512i one = _mm512_const1_64( 1 );
+//     const __m512i two = _mm512_add_epi64( one, one );
+//     
+//////////////////////////////////////////////////////////////////////////

 #include <inttypes.h>
 #include <x86intrin.h>
--- a/simd-utils/intrlv.h
+++ b/simd-utils/intrlv.h
--- a/simd-utils/simd-128.h
+++ b/simd-utils/simd-128.h
@@ -3,183 +3,132 @@

 #if defined(__SSE2__)

-//////////////////////////////////////////////////////////////////
+///////////////////////////////////////////////////////////////////////////
 //
 //                 128 bit SSE vectors
 //
-// SSE2 is generally required for full 128 bit support. Some functions
-// are also optimized with SSSE3 or SSE4.1.
-//
-// Do not call intrinsic _mm_extract directly, it isn't supported in SSE2.
-// Use mm128_extr macro instead, it will select the appropriate implementation.
-//
-// 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 but is sometimes the only way for certain operations.
+// SSE2 is required for 128 bit integer support. Some functions are also
+// optimized with SSSE3, SSE4.1 or AVX. Some of these more optimized
+// functions don't have SSE2 equivalents and their use would break SSE2
+// compatibility.
 //
 // Constants are an issue with simd. Simply put, immediate constants don't
 // exist. All simd constants either reside in memory or a register and
-// must be loaded from memory or generated using instructions at run time.
+// must be loaded from memory or generated at run time.
 //
-// Due to the cost of generating constants it is often more efficient to
+// Due to the cost of generating constants it is more efficient to
 // define a local const for repeated references to the same constant.
 //
-// Some constant values can be generated using shortcuts. Zero for example
-// is as simple as XORing any register with itself, and is implemented
-// iby the setzero instrinsic. These shortcuts must be implemented using 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. These pseudo-constants
-// do not perform any memory accesses
+// One common use for simd constants is as a control index for vector
+// instructions like blend and shuffle. Alhough the ultimate instruction
+// may execute in a single clock cycle, generating the control index adds
+// several more cycles to the entire operation. 
 //
-// 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.
+// All of the utilities here assume all data is in registers except
+// in rare cases where arguments are pointers.
+//
+// Intrinsics automatically promote from REX to VEX when AVX is available
+// but ASM needs to be done manually.
+//
+///////////////////////////////////////////////////////////////////////////
+
+
+// Efficient and convenient moving bwtween GP & low bits of XMM.
+// Use VEX when available to give access to xmm8-15 and zero extend for
+// larger vectors.
+
+static inline __m128i mm128_mov64_128( const uint64_t n )
+{
+  __m128i a;
+#if defined(__AVX__)
+  asm( "vmovq %1, %0\n\t" : "=x"(a) : "r"(n) );
+#else
+  asm( "movq %1, %0\n\t" : "=x"(a) : "r"(n) );
+#endif
+  return  a;
+}
+
+static inline __m128i mm128_mov32_128( const uint32_t n )
+{
+  __m128i a;
+#if defined(__AVX__)
+  asm( "vmovd %1, %0\n\t" : "=x"(a) : "r"(n) );
+#else  
+  asm( "movd %1, %0\n\t" : "=x"(a) : "r"(n) );
+#endif
+  return  a;
+}
+
+static inline uint64_t mm128_mov128_64( const __m128i a )
+{
+  uint64_t n;
+#if defined(__AVX__)
+  asm( "vmovq %1, %0\n\t" : "=r"(n) : "x"(a) );
+#else  
+  asm( "movq %1, %0\n\t" : "=r"(n) : "x"(a) );
+#endif
+  return  n;
+}
+
+static inline uint32_t mm128_mov128_32( const __m128i a )
+{
+  uint32_t n;
+#if defined(__AVX__)
+  asm( "vmovd %1, %0\n\t" : "=r"(n) : "x"(a) );
+#else  
+  asm( "movd %1, %0\n\t" : "=r"(n) : "x"(a) );
+#endif
+  return  n;
+}
+
+// Pseudo constants

 #define m128_zero      _mm_setzero_si128()
+#define m128_one_128   mm128_mov64_128( 1 )
+#define m128_one_64    _mm_shuffle_epi32( mm128_mov64_128( 1 ), 0x44 )
+#define m128_one_32    _mm_shuffle_epi32( mm128_mov32_128( 1 ), 0x00 )
+#define m128_one_16    _mm_shuffle_epi32( \
+                                 mm128_mov32_128( 0x00010001 ), 0x00 )
+#define m128_one_8     _mm_shuffle_epi32( \
+                                 mm128_mov32_128( 0x01010101 ), 0x00 )

-static inline __m128i mm128_one_128_fn()
-{
-  __m128i a;
-   const uint64_t one = 1;
-   asm( "movq %1, %0\n\t"
-        : "=x"(a)
-        : "r" (one) );
-   return a;
-}
-#define m128_one_128    mm128_one_128_fn()
-
-static inline __m128i mm128_one_64_fn()
-{
-  __m128i a;
-  const uint64_t one = 1;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r"  (one) );
-  return _mm_shuffle_epi32( a, 0x44 );
-}
-#define m128_one_64    mm128_one_64_fn()
-
-static inline __m128i mm128_one_32_fn()
-{
-  __m128i a;
-  const uint32_t one = 1;
-  asm( "movd %1, %0\n\t"
-       : "=x" (a)
-       : "r"  (one) );
-  return _mm_shuffle_epi32( a, 0x00 );
-}
-#define m128_one_32    mm128_one_32_fn()
-
-static inline __m128i mm128_one_16_fn()
-{
-  __m128i a;
-  const uint32_t one = 0x00010001;
-  asm( "movd %1, %0\n\t"
-       : "=x" (a)
-       : "r"  (one) );
-  return _mm_shuffle_epi32( a, 0x00 );
-}
-#define m128_one_16    mm128_one_16_fn()
-
-static inline __m128i mm128_one_8_fn()
-{
-  __m128i a;
-  const uint32_t one = 0x01010101;
-  asm( "movd %1, %0\n\t"
-       : "=x" (a)
-       : "r"  (one) );
-  return _mm_shuffle_epi32( a, 0x00 );
-}
-#define m128_one_8    mm128_one_8_fn()
+// ASM avoids the need to initialize return variable to avoid compiler warning.
+// Macro abstracts function parentheses to look like an identifier.

 static inline __m128i mm128_neg1_fn()
 {
   __m128i a;
-   asm( "pcmpeqd %0, %0\n\t"
-        : "=x" (a) );
+#if defined(__AVX__) 
+   asm( "vpcmpeqq %0, %0, %0\n\t" : "=x"(a) );
+#else
+   asm( "pcmpeqq %0, %0\n\t" : "=x"(a) );
+#endif
   return a;
 }
 #define m128_neg1    mm128_neg1_fn()

-// move uint64_t to low bits of __m128i, zeros the rest
-static inline __m128i mm128_mov64_128( uint64_t n )
-{
-  __m128i a;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r"  (n) );
-  return  a;
-}

-static inline __m128i mm128_mov32_128( uint32_t n )
-{
-  __m128i a;
-  asm( "movd %1, %0\n\t"
-       : "=x" (a)
-       : "r"  (n) );
-  return  a;
-}
+// const functions work best when arguments are immediate constants or
+// are known to be in registers. If data needs to loaded from memory or cache
+// use set.

-static inline uint64_t mm128_mov128_64( __m128i a )
-{
-  uint64_t n;
-  asm( "movq %1, %0\n\t"
-       : "=x" (n)
-       : "r"  (a) );
-  return  n;
-}
+// Equivalent of set1, broadcast 64 bit integer to all elements.
+#define m128_const1_64( i ) _mm_shuffle_epi32( mm128_mov64_128( i ), 0x44 )
+#define m128_const1_32( i ) _mm_shuffle_epi32( mm128_mov32_128( i ), 0x00 )

-static inline uint32_t mm128_mov128_32( __m128i a )
-{
-  uint32_t n;
-  asm( "movd %1, %0\n\t"
-       : "=x" (n)
-       : "r"  (a) );
-  return  n;
-}
+#if defined(__SSE4_1__)

-static inline __m128i m128_const1_64( const uint64_t n )
-{
-  __m128i a;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r"  (n) );
-  return _mm_shuffle_epi32( a, 0x44 );
-}
+// Assign 64 bit integers to respective elements: {hi, lo}
+#define m128_const_64( hi, lo ) \
+   _mm_insert_epi64( mm128_mov64_128( lo ), hi, 1 )

-static inline __m128i m128_const1_32( const uint32_t n )
-{
-  __m128i a;
-  asm( "movd %1, %0\n\t"
-       : "=x" (a)
-       : "r"  (n) );
-  return _mm_shuffle_epi32( a, 0x00 );
-}
-
-#if defined(__SSE41__)
-
-// alternative to _mm_set_epi64x, doesn't use mem,
-
-static inline __m128i m128_const_64( const uint64_t hi, const uint64_t lo )
-{
-   __m128i a;
-   asm( "movq %2, %0\n\t"
-        "pinsrq $1, %1, %0\n\t"
-        : "=x" (a)
-        : "r"  (hi), "r" (lo) );
-   return a;
-}
-
-#else
+#else  // No insert in SSE2

 #define m128_const_64  _mm_set_epi64x

 #endif

+
 //
 // Basic operations without equivalent SIMD intrinsic

@@ -207,18 +156,9 @@ static inline __m128i m128_const_64( const uint64_t hi, const uint64_t lo )
 #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 ) \
-do { \
-  a0 = _mm_extract_epi32( src, 0 ); \
-  a1 = _mm_extract_epi32( src, 1 ); \
-  a1 = _mm_extract_epi32( src, 2 ); \
-  a3 = _mm_extract_epi32( src, 3 ); \
-} while(0)
-
 // Horizontal vector testing

-#if defined(__SSE41__)
+#if defined(__SSE4_1__)

 #define mm128_allbits0( a )    _mm_testz_si128(   a, a )
 #define mm128_allbits1( a )    _mm_testc_si128(   a, m128_neg1 )
@@ -235,7 +175,7 @@ do { \
 #define mm128_allbits0( a ) ( !mm128_anybits1(a) )
 #define mm128_allbits1( a ) ( !mm128_anybits0(a) )

-#endif // SSE41 else SSE2
+#endif // SSE4.1 else SSE2

 //
 // Vector pointer cast
@@ -256,20 +196,6 @@ do { \
 // returns pointer p+o
 #define casto_m128i(p,o) (((__m128i*)(p))+(o))

-// SSE2 doesn't implement extract
-#if defined(__SSE4_1)
-
-#define mm128_extr_64(a,n)   _mm_extract_epi64( a, n )
-#define mm128_extr_32(a,n)   _mm_extract_epi32( a, n )
-
-#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])
-
-#endif
-

 // Memory functions
 // Mostly for convenience, avoids calculating bytes.
@@ -294,13 +220,14 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
 //
 // Bit rotations

-// AVX512 has implemented bit rotation for 128 bit vectors with
+// AVX512VL has implemented bit rotation for 128 bit vectors with
 // 64 and 32 bit elements.

 // compiler doesn't like when a variable is used for the last arg of
-// _mm_rol_epi32, must be "8 bit immediate". Therefore use rol_var where
+// _mm_rol_epi32, must be "8 bit immediate". Oddly _mm_slli has the same
+// specification but works with a variable. Therefore use rol_var where
 // necessary.
-// sm3-hash-4way.c fails to compile.
+// sm3-hash-4way.c has one instance where mm128_rol_var_32 is required.

 #define mm128_ror_var_64( v, c ) \
   _mm_or_si128( _mm_srli_epi64( v, c ), _mm_slli_epi64( v, 64-(c) ) )
@@ -315,7 +242,7 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
   _mm_or_si128( _mm_slli_epi32( v, c ), _mm_srli_epi32( v, 32-(c) ) )


-/*
+
 #if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

 #define mm128_ror_64    _mm_ror_epi64
@@ -324,14 +251,14 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
 #define mm128_rol_32    _mm_rol_epi32

 #else
-*/
+

 #define mm128_ror_64   mm128_ror_var_64
 #define mm128_rol_64   mm128_rol_var_64
 #define mm128_ror_32   mm128_ror_var_32
 #define mm128_rol_32   mm128_rol_var_32

-//#endif   // AVX512 else
+#endif   // AVX512 else

 #define mm128_ror_16( v, c ) \
   _mm_or_si128( _mm_srli_epi16( v, c ), _mm_slli_epi16( v, 16-(c) ) )
@@ -392,18 +319,19 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
 //
 // Rotate elements within lanes.

-// Equivalent to mm128_ror_64( v, 32 )
 #define mm128_swap32_64( v )  _mm_shuffle_epi32( v, 0xb1 )

-// Equivalent to mm128_ror_64( v, 16 )
-#define mm128_ror16_64( v )   _mm_shuffle_epi8( v, \
-                   m128_const_64( 0x09080f0e0d0c0b0a, 0x0100070605040302 )
-#define mm128_rol16_64( v )   _mm_shuffle_epi8( v, \
-                   m128_const_64( 0x0dc0b0a09080f0e, 0x0504030201000706 )
+#define mm128_ror16_64( v ) \
+   _mm_shuffle_epi8( v, m128_const_64( 0x09080f0e0d0c0b0a, \
+                                       0x0100070605040302 )

-// Equivalent to mm128_ror_32( v, 16 )
-#define mm128_swap16_32( v )  _mm_shuffle_epi8( v, \
-                   m128_const_64( 0x0d0c0f0e09080b0a, 0x0504070601000302 )
+#define mm128_rol16_64( v ) \
+   _mm_shuffle_epi8( v, m128_const_64( 0x0d0c0b0a09080f0e, \
+                                       0x0504030201000706 )
+
+#define mm128_swap16_32( v ) \
+   _mm_shuffle_epi8( v, m128_const_64( 0x0d0c0f0e09080b0a, \
+                                       0x0504070601000302 )

 //
 // Endian byte swap.
@@ -418,8 +346,9 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
   _mm_shuffle_epi8( v, m128_const_64( 0x0c0d0e0f08090a0b, \
                                       0x0405060700010203 ) )

-#define mm128_bswap_16( v ) _mm_shuffle_epi8( \
-                   m128_const_64( 0x0e0f0c0d0a0b0809, 0x0607040502030001 )
+#define mm128_bswap_16( v ) \
+   _mm_shuffle_epi8( v, m128_const_64( 0x0e0f0c0d0a0b0809, \
+                                       0x0607040502030001 )

 // 8 byte qword * 8 qwords * 2 lanes = 128 bytes
 #define mm128_block_bswap_64( d, s ) do \
--- a/simd-utils/simd-256.h
+++ b/simd-utils/simd-256.h
@@ -1,7 +1,7 @@
 #if !defined(SIMD_256_H__)
 #define SIMD_256_H__ 1

-#if defined(__AVX__)
+#if defined(__AVX2__)

 /////////////////////////////////////////////////////////////////////
 //
@@ -14,176 +14,68 @@
 // is limited because 256 bit vectors are less likely to be used when 512
 // is available.

-// set instructions load memory resident constants, this avoids mem.
-// cost 4 pinsert + 1 vinsert, estimate 8 clocks latency.
+// Move integer to low element of vector, other elements are set to zero.

-#if defined(__AVX2__)
+#define mm256_mov64_256( n ) _mm256_castsi128_si256( mm128_mov64_128( n ) )
+#define mm256_mov32_256( n ) _mm256_castsi128_si256( mm128_mov32_128( n ) )

-#define m256_const_128( hi, lo ) \
+#define mm256_mov256_64( a ) mm128_mov128_64( _mm256_castsi256_si128( a ) )
+#define mm256_mov256_32( a ) mm128_mov128_32( _mm256_castsi256_si128( a ) )
+
+// concatenate two 128 bit vectors into one 256 bit vector: { hi, lo }
+#define mm256_concat_128( hi, lo ) \
   _mm256_inserti128_si256( _mm256_castsi128_si256( lo ), hi, 1 )

-#define m256_const_64( i3, i2, i1, i0 ) \
-   m256_const_128( m128_const_64( i3, i2 ), m128_const_64( i1, i0 ) )
+#define m256_const1_128( v ) \
+         _mm256_broadcastsi128_si256( v )

-/*
-#define m256_const_64( i3, i2, i1, i0 ) \
-  _mm256_inserti128_si256( _mm256_castsi128_si256( m128_const_64( i1, i0 ) ), \
-                                                   m128_const_64( i3, i2 ), 1 )
-*/
-
-#else    // AVX
-
-#define m256_const_64( i3, i2, i1, i0 )  _mm256_set_epi64x( i3, i2, i1, i0 )
-
-#endif
-
-static inline __m256i m256_const1_64( uint64_t i )
+// Equavalent of set, move 64 bit integer constants to respective 64 bit
+// elements.
+static inline __m256i m256_const_64( const uint64_t i3, const uint64_t i2,
+                                     const uint64_t i1, const uint64_t i0 )
 {
-   __m128i a;
-   asm( "movq %1, %0\n\t"
-      : "=x" (a)
-      : "r"  (i) );
-   return _mm256_broadcastq_epi64( a );
+    __m128i hi, lo;
+   lo = mm128_mov64_128( i0 );
+   hi = mm128_mov64_128( i2 );
+   lo = _mm_insert_epi64( lo, i1, 1 );
+   hi = _mm_insert_epi64( hi, i3, 1 );
+   return mm256_concat_128( hi, lo );
 }

-static inline __m256i m256_const1_32( uint32_t i )
-{
-   __m128i a;
-   asm( "movd %1, %0\n\t"
-      : "=x" (a)
-      : "r"  (i) );
-   return _mm256_broadcastd_epi32( a );
-}
+// Broadcast 128 bits in pairs of 64 bit integer constants {i1. i0} to all
+// 128 bit lanes.
+#define m256_const2_64( i1, i0 ) \
+    _mm256_permute4x64_epi64( _mm256_castsi128_si256( \
+                              m128_const_64( i1, i0 ) ), 0x44 )

-static inline __m256i m256_const1_16( uint16_t i )
-{
-   __m128i a;
-   asm( "movw %1, %0\n\t"
-      : "=x" (a)
-      : "r"  (i) );
-   return _mm256_broadcastw_epi16( a );
-}
+// Equivalent of set1, broadcast integer constant to all elements.
+#define m256_const1_64( i ) _mm256_broadcastq_epi64( mm128_mov64_128( i ) )
+#define m256_const1_32( i ) _mm256_broadcastd_epi32( mm128_mov32_128( i ) )
+#define m256_const1_16( i ) _mm256_broadcastw_epi16( mm128_mov32_128( i ) )
+#define m256_const1_8 ( i ) _mm256_broadcastb_epi8 ( mm128_mov32_128( i ) )

-static inline __m256i m256_const1_8( uint8_t i )
-{
-   __m128i a;
-   asm( "movb %1, %0\n\t"
-      : "=x" (a)
-      : "r"  (i) );
-   return _mm256_broadcastb_epi8( a );
-}

 //
 // All SIMD constant macros are actually functions containing executable
 // code and therefore can't be used as compile time initializers.

-#define m256_zero         _mm256_setzero_si256()
-
-#if defined(__AVX2__)
-
-// Don't call the frunction directly, use the macro to make appear like
-// a constant identifier instead of a function.
-// __m256i foo = m256_one_64; 
-
-static inline __m256i mm256_one_256_fn()
-{
-  __m256i a;
-  const uint64_t one = 1;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r" (one) );
-  return a;
-}
-#define m256_one_256    mm256_one_256_fn()
-
-static inline __m256i mm256_one_128_fn()
-{
-  __m128i a;
-  const uint64_t one = 1;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r" (one) );
-  return _mm256_broadcastsi128_si256( a );
-}
-#define m256_one_128    mm256_one_128_fn()
-
-static inline __m256i mm256_one_64_fn()
-{
-  __m128i a;
-  const uint64_t one = 1;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r" (one) );
-  return _mm256_broadcastq_epi64( a );
-}
-#define m256_one_64    mm256_one_64_fn()
-
-static inline __m256i mm256_one_32_fn()
-{
-  __m128i a;
-  const uint64_t one = 0x0000000100000001;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r" (one) );
-  return _mm256_broadcastq_epi64( a );
-}
-#define m256_one_32    mm256_one_32_fn()
-
-static inline __m256i mm256_one_16_fn()
-{
-  __m128i a;
-  const uint64_t one = 0x0001000100010001;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r" (one) );
-  return _mm256_broadcastq_epi64( a );
-}
-#define m256_one_16    mm256_one_16_fn()
-
-static inline __m256i mm256_one_8_fn()
-{
-  __m128i a;
-  const uint64_t one = 0x0101010101010101;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r" (one) );
-  return _mm256_broadcastq_epi64( a );
-}
-#define m256_one_8    mm256_one_8_fn()
+#define m256_zero       _mm256_setzero_si256()
+#define m256_one_256    mm256_mov64_256( 1 )
+#define m256_one_128 \
+    _mm256_permute4x64_epi64( _mm256_castsi128_si256( \
+                               mm128_mov64_128( 1 ) ), 0x44 )
+#define m256_one_64     _mm256_broadcastq_epi64( mm128_mov64_128( 1 ) )
+#define m256_one_32     _mm256_broadcastd_epi32( mm128_mov64_128( 1 ) )
+#define m256_one_16     _mm256_broadcastw_epi16( mm128_mov64_128( 1 ) )
+#define m256_one_8      _mm256_broadcastb_epi8 ( mm128_mov64_128( 1 ) )

 static inline __m256i mm256_neg1_fn()
 {
   __m256i a;
-   asm( "vpcmpeqq %0, %0, %0\n\t"
-        : "=x"(a) );
+   asm( "vpcmpeqq %0, %0, %0\n\t" : "=x"(a) );
   return a;
 }
-#define m256_neg1    mm256_neg1_fn()
-
-#else  // AVX
-
-#define m256_one_256 m256_const_64( m128_zero, m128_one ) \
-   _mm256_inserti128_si256( _mm256_castsi128_si256( m128_one_128 ), \
-                                                    m128_zero, 1 )
-
-#define m256_one_128 \
-   _mm256_inserti128_si256( _mm256_castsi128_si256( m128_one_128 ), \
-                                                    m128_one_128, 1 )
-
-#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. 
-static inline __m256i mm256_neg1_fn()
-{
-   __m128i a = m128_neg1;
-   return _mm256_insertf128_si256( _mm256_castsi128_si256( a ), a, 1 );
-}
-#define m256_neg1   mm256_neg1_fn()
-
-#endif  // AVX2 else AVX
+#define m256_neg1  mm256_neg1_fn()


 //
@@ -202,58 +94,32 @@ static inline __m256i mm256_neg1_fn()
 #define mm128_extr_hi128_256( a ) _mm256_extracti128_si256( a, 1 )

 // Extract integers from 256 bit vector, ineficient, avoid if possible..
-#define mm256_extr_4x64( a0, a1, a2, a3, src ) \
+#define mm256_extr_4x64( a3, a2, a1, a0, src ) \
 do { \
  __m128i hi = _mm256_extracti128_si256( src, 1 ); \
-  a0 = mm256_mov256_64( src ); \
+  a0 = mm128_mov128_64( _mm256_castsi256_si128( src) ); \
  a1 = _mm_extract_epi64( _mm256_castsi256_si128( src ), 1 ); \
  a2 = mm128_mov128_64( hi ); \
  a3 = _mm_extract_epi64( hi, 1 ); \
 } while(0)

-#define mm256_extr_8x32( a0, a1, a2, a3, a4, a5, a6, a7, src ) \
+#define mm256_extr_8x32( a7, a6, a5, a4, a3, a2, a1, a0, src ) \
 do { \
+  uint64_t t = _mm_extract_epi64( _mm256_castsi256_si128( src ), 1 ); \
  __m128i hi = _mm256_extracti128_si256( src, 1 ); \
  a0 = mm256_mov256_32( src ); \
  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 ); \
+  a2 = (uint32_t)( t ); \
+  a3 = (uint32_t)( t<<32 ); \
+  t = _mm_extract_epi64(  hi, 1 ); \
  a4 = mm128_mov128_32( hi ); \
  a5 = _mm_extract_epi32( hi, 1 ); \
-  a6 = _mm_extract_epi32( hi, 2 ); \
-  a7 = _mm_extract_epi32( hi, 3 ); \
+  a6 = (uint32_t)( t ); \
+  a7 = (uint32_t)( t<<32 ); \
 } while(0)

-// concatenate two 128 bit vectors into one 256 bit vector: { hi, lo }
-#define mm256_concat_128( hi, lo ) \
-   _mm256_inserti128_si256( _mm256_castsi128_si256( lo ), hi, 1 )
-
-// Move integer to lower bits of vector, upper bits set to zero.
-static inline __m256i mm256_mov64_256( uint64_t n )
-{
-  __m128i a;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r" (n) );
-  return _mm256_castsi128_si256( a );
-}
-
-static inline __m256i mm256_mov32_256( uint32_t n )
-{
-  __m128i a;
-  asm( "movd %1, %0\n\t"
-       : "=x" (a)
-       : "r" (n) );
-  return _mm256_castsi128_si256( a );
-}
-
-// Return lo bits of vector as integer.
-#define mm256_mov256_64( a ) mm128_mov128_64( _mm256_castsi256_si128( a ) )
-
-#define mm256_mov256_32( a ) mm128_mov128_32( _mm256_castsi256_si128( a ) )

 // 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 )
@@ -261,21 +127,6 @@ static inline __m256i mm256_mov32_256( uint32_t n )
 #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 ) \
-         ( (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) )
-
-#endif   // AVX2 else AVX

 // Parallel AES, for when x is expected to be in a 256 bit register.
 // Use same 128 bit key.
@@ -324,12 +175,6 @@ static inline void memset_256( __m256i *dst, const __m256i a, const int n )
 static inline void memcpy_256( __m256i *dst, const __m256i *src, const 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
@@ -388,7 +233,7 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
   _mm256_or_si256( _mm256_slli_epi32( v, c ), \
                    _mm256_srli_epi32( v, 32-(c) ) )

-/*
+
 #if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

 // AVX512, control must be 8 bit immediate.
@@ -399,7 +244,7 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 #define mm256_rol_32    _mm256_rol_epi32

 #else
-*/
+

 // No AVX512, use fallback.

@@ -408,7 +253,7 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 #define mm256_ror_32    mm256_ror_var_32
 #define mm256_rol_32    mm256_rol_var_32

-// #endif     // AVX512 else
+#endif     // AVX512 else

 #define  mm256_ror_16( v, c ) \
   _mm256_or_si256( _mm256_srli_epi16( v, c ), \
@@ -464,6 +309,20 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 //
 // AVX2 has no full vector permute for elements less than 32 bits.
 // AVX512 has finer granularity full vector permutes.
+// AVX512 has full vector alignr which might be faster, especially for 32 bit
+
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define mm256_swap_128( v )   _mm256_alignr_epi64( v, v, 2 )
+#define mm256_ror_1x64( v )   _mm256_alignr_epi64( v, v, 1 )
+#define mm256_rol_1x64( v )   _mm256_alignr_epi64( v, v, 3 )
+#define mm256_ror_1x32( v )   _mm256_alignr_epi32( v, v, 1 )
+#define mm256_rol_1x32( v )   _mm256_alignr_epi32( v, v, 7 )
+#define mm256_ror_3x32( v )   _mm256_alignr_epi32( v, v, 3 )
+#define mm256_rol_3x32( v )   _mm256_alignr_epi32( v, v, 5 )
+
+#else   // AVX2

 // Swap 128 bit elements in 256 bit vector.
 #define mm256_swap_128( v )     _mm256_permute4x64_epi64( v, 0x4e )
@@ -472,7 +331,6 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 #define mm256_ror_1x64( v )     _mm256_permute4x64_epi64( v, 0x39 )
 #define mm256_rol_1x64( v )     _mm256_permute4x64_epi64( v, 0x93 )

-// A little faster with avx512
 // Rotate 256 bit vector by one 32 bit element.
 #define mm256_ror_1x32( v ) \
    _mm256_permutevar8x32_epi32( v, \
@@ -495,6 +353,9 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
                     m256_const_64( 0x0000000400000003, 0x0000000200000001, \
                                    0x0000000000000007, 0x0000000600000005 )

+#endif    // AVX512 else AVX2
+
+
 // AVX512 can do 16 & 8 bit elements.
 #if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

@@ -537,18 +398,16 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )

 // Invert vector: {7,6,5,4,3,2,1,0} -> {0,1,2,3,4,5,6,7}
 #define mm256_invert_16 ( v ) \
-     _mm256_permutexvar_epi16( m256_const_64( 0x0000000100020003, \
-                                              0x0004000500060007, \
-                                              0x00080009000a000b, \
-                                              0x000c000d000e000f ), v )
+     _mm256_permutexvar_epi16( m256_const_64( \
+                                  0x0000000100020003, 0x0004000500060007, \
+                                  0x00080009000a000b, 0x000c000d000e000f ), v )

 #if defined(__AVX512VBMI__)

 #define mm256_invert_8( v ) \
-     _mm256_permutexvar_epi8( m256_const_64( 0x0001020304050607, \
-                                             0x08090a0b0c0d0e0f, \
-                                             0x1011121314151617, \
-                                             0x18191a1b1c1d1e1f ), v )
+     _mm256_permutexvar_epi8( m256_const_64( \
+                                  0x0001020304050607, 0x08090a0b0c0d0e0f, \
+                                  0x1011121314151617, 0x18191a1b1c1d1e1f ), v )
 #endif  // VBMI
 #endif  // AVX512

@@ -563,29 +422,25 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 #define mm256_ror1x32_128( v )  _mm256_shuffle_epi32( v, 0x39 )
 #define mm256_rol1x32_128( v )  _mm256_shuffle_epi32( v, 0x93 )

-// Rotate each 128 bit lane by one 16 bit element.
 #define mm256_ror1x16_128( v ) \
-        _mm256_shuffle_epi8( v, m256_const_64( 0x01000f0e0d0c0b0a, \
-                                               0x0908070605040302, \
-                                               0x01000f0e0d0c0b0a, \
-                                               0x0908070605040302 ) )
-#define mm256_rol1x16_128( v ) \
-        _mm256_shuffle_epi8( v, m256_const_64( 0x0d0c0b0a09080706, \
-                                               0x0504030201000f0e, \
-                                               0x0d0c0b0a09080706, \
-                                               0x0504030201000f0e ) )
+   _mm256_shuffle_epi8( v, \
+         m256_const_64( 0x11101f1e1d1c1b1a, 0x1918171615141312, \
+                        0x01000f0e0d0c0b0a, 0x0908070605040302 ) )
+
+#define mm256_rol1x16_128( v ) \
+   _mm256_shuffle_epi8( v, \
+         m256_const_64( 0x1d1c1b1a19181716, 0x1514131211101f1e, \
+                        0x0d0c0b0a09080706, 0x0504030201000f0e ) )

-// Rotate each 128 bit lane by one byte
 #define mm256_ror1x8_128( v ) \
-        _mm256_shuffle_epi8( v, m256_const_64( 0x000f0e0d0c0b0a09, \
-                                               0x0807060504030201, \
-                                               0x000f0e0d0c0b0a09, \
-                                               0x0807060504030201 ) )
+   _mm256_shuffle_epi8( v, \
+         m256_const_64( 0x101f1e1d1c1b1a19, 0x1817161514131211, \
+                        0x000f0e0d0c0b0a09, 0x0807060504030201 ) )
+
 #define mm256_rol1x8_128( v ) \
-        _mm256_shuffle_epi8( v, m256_const_64( 0x0c0b0a09080f0e0d, \
-                                               0x0504030201000706, \
-                                               0x0d0c0b0a09080f0e, \
-                                               0x0504030201000706 ) )
+   _mm256_shuffle_epi8( v, \
+         m256_const_64( 0x1d1c1b1a19181f1e, 0x1514131211101716, \
+                        0x0d0c0b0a09080f0e, 0x0504030201000706 ) )

 // Rotate each 128 bit lane by c bytes.
 #define mm256_bror_128( v, c ) \
@@ -599,70 +454,65 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 #define mm256_swap32_64( v )    _mm256_shuffle_epi32( v, 0xb1 )

 #define mm256_ror1x16_64( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x09080f0e0d0c0b0a, \
-                                          0x0100070605040302, \
-                                          0x09080f0e0d0c0b0a, \
-                                          0x0100070605040302 ) )
+   _mm256_shuffle_epi8( v, \
+        m256_const_64( 0x19181f1e1d1c1b1a, 0x1110171615141312, \
+                       0x09080f0e0d0c0b0a, 0x0100070605040302 ) )
+
 #define mm256_rol1x16_64( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x0d0c0b0a09080f0e, \
-                                          0x0504030201000706, \
-                                          0x0d0c0b0a09080f0e, \
-                                          0x0504030201000706 ))
+   _mm256_shuffle_epi8( v, \
+        m256_const_64( 0x1d1c1b1a19181f1e, 0x1514131211101716, \
+                       0x0d0c0b0a09080f0e, 0x0504030201000706 ) )

 #define mm256_ror1x8_64( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x080f0e0d0c0b0a09, \
-                                          0x0007060504030201, \
-                                          0x080f0e0d0c0b0a09, \
-                                          0x0007060504030201 ))
+   _mm256_shuffle_epi8( v, \
+        m256_const_64( 0x181f1e1d1c1b1a19, 0x1017161514131211, \
+                       0x080f0e0d0c0b0a09, 0x0007060504030201 ) )
+
 #define mm256_rol1x8_64( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x0e0d0c0b0a09080f, \
-                                          0x0605040302010007, \
-                                          0x0e0d0c0b0a09080f, \
-                                          0x0605040302010007 ) )
+   _mm256_shuffle_epi8( v, \
+        m256_const_64( 0x1e1d1c1b1a19181f, 0x1615141312111017, \
+                       0x0e0d0c0b0a09080f, 0x0605040302010007 ) )

 #define mm256_ror3x8_64( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x0a09080f0e0d0c0b, \
-                                          0x0201000706050403, \
-                                          0x0a09080f0e0d0c0b, \
-                                          0x0201000706050403 ) )
+   _mm256_shuffle_epi8( v, \
+        m256_const_64( 0x1a19181f1e1d1c1b, 0x1211101716151413, \
+                       0x0a09080f0e0d0c0b, 0x0201000706050403 ) )
+
 #define mm256_rol3x8_64( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x0c0b0a09080f0e0d, \
-                                          0x0403020100070605, \
-                                          0x0c0b0a09080f0e0d, \
-                                          0x0403020100070605 ) )
+   _mm256_shuffle_epi8( v, \
+        m256_const_64( 0x1c1b1a19181f1e1d, 0x1413121110171615, \
+                       0x0c0b0a09080f0e0d, 0x0403020100070605 ) )
+

 // Swap 16 bit elements in each 32 bit lane
 #define mm256_swap16_32( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x0b0a09080f0e0d0c, \
-                                          0x0302010007060504, \
-                                          0x0b0a09080f0e0d0c, \
-                                          0x0302010007060504 )
+   _mm256_shuffle_epi8( v, \
+         m256_const_64( 0x1b1a19181f1e1d1c, 0x1312111017161514, \
+                        0x0b0a09080f0e0d0c, 0x0302010007060504 ) )

 //
 // Swap bytes in vector elements, endian bswap.
 #define mm256_bswap_64( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x08090a0b0c0d0e0f, \
-                                          0x0001020304050607, \
-                                          0x08090a0b0c0d0e0f, \
-                                          0x0001020304050607 ) )
+   _mm256_shuffle_epi8( v, \
+         m256_const_64( 0x18191a1b1c1d1e1f, 0x1011121314151617, \
+                        0x08090a0b0c0d0e0f, 0x0001020304050607 ) )

 #define mm256_bswap_32( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x0c0d0e0f08090a0b, \
-                                          0x0405060700010203, \
-                                          0x0c0d0e0f08090a0b, \
-                                          0x0405060700010203 ) )
+   _mm256_shuffle_epi8( v, \
+         m256_const_64( 0x1c1d1e1f18191a1b, 0x1415161710111213, \
+                        0x0c0d0e0f08090a0b, 0x0405060700010203 ) )

 #define mm256_bswap_16( v ) \
-   _mm256_shuffle_epi8( v, m256_const_64( 0x0e0f0c0d0a0b0809, \
-                                          0x0607040502030001, \
-                                          0x0e0f0c0d0a0b0809, \
-                                          0x0607040502030001 ) )
+   _mm256_shuffle_epi8( v, \
+         m256_const_64( 0x1e1f1c1d1a1b1819, 0x1617141512131011, \
+                        0x0e0f0c0d0a0b0809, 0x0607040502030001, ) )

+// Source and destination are pointers, may point to same memory.
 // 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 ); \
+  __m256i ctl = m256_const_64( 0x18191a1b1c1d1e1f, 0x1011121314151617, \
+                               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 ); \
@@ -676,7 +526,7 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 // 4 byte dword * 8 dwords * 8 lanes = 256 bytes
 #define mm256_block_bswap_32( d, s ) do \
 { \
-  __m256i ctl = m256_const_64( 0x0c0d0e0f08090a0b, 0x0405060700010203, \
+  __m256i ctl = m256_const_64( 0x1c1d1e1f18191a1b, 0x1415161710111213, \
                               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 ); \
@@ -695,6 +545,9 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
 // Some of these can use permute but appears to be slower. Maybe a Ryzen
 // issue

+//  _mm256_alignr_epi 64/32 are only available with AVX512 but AVX512 also
+//  makes these macros unnecessary.
+
 #define mm256_swap256_512 (v1, v2) \
   v1 = _mm256_xor_si256(v1, v2); \
   v2 = _mm256_xor_si256(v1, v2); \
@@ -702,75 +555,18 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )

 #define mm256_ror1x128_512( v1, v2 ) \
 do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 16 ); \
-   v1 = _mm256_alignr_epi8( v2, v1, 16 ); \
+   __m256i t = _mm256_permute2x128( v1, v2, 0x03 ); \
+   v1 = _mm256__mm256_permute2x128( v2, v1, 0x21 ); \
   v2 = t; \
 } while(0)

 #define mm256_rol1x128_512( v1, v2 ) \
 do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 16 ); \
-   v2 = _mm256_alignr_epi8( v2, v1, 16 ); \
-   v1 = t; \
-} while(0)
-
-#define mm256_ror1x64_512( v1, v2 ) \
-do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 8 ); \
-   v1 = _mm256_alignr_epi8( v2, v1, 8 ); \
-   v2 = t; \
-} while(0)
-
-#define mm256_rol1x64_512( v1, v2 ) \
-do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 24 ); \
-   v2 = _mm256_alignr_epi8( v2, v1, 24 ); \
-   v1 = t; \
-} while(0)
-
-#define mm256_ror1x32_512( v1, v2 ) \
-do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 4 ); \
-   v1 = _mm256_alignr_epi8( v2, v1, 4 ); \
-   v2 = t; \
-} while(0)
-
-#define mm256_rol1x32_512( v1, v2 ) \
-do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 28 ); \
-   v2 = _mm256_alignr_epi8( v2, v1, 28 ); \
-   v1 = t; \
-} while(0)
-
-#define mm256_ror1x16_512( v1, v2 ) \
-do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 2 ); \
-   v1 = _mm256_alignr_epi8( v2, v1, 2 ); \
-   v2 = t; \
-} while(0)
-
-#define mm256_rol1x16_512( v1, v2 ) \
-do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 30 ); \
-   v2 = _mm256_alignr_epi8( v2, v1, 30 ); \
-   v1 = t; \
-} while(0)
-
-#define mm256_ror1x8_512( v1, v2 ) \
-do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 1 ); \
-   v1 = _mm256_alignr_epi8( v2, v1, 1 ); \
-   v2 = t; \
-} while(0)
-
-#define mm256_rol1x8_512( v1, v2 ) \
-do { \
-   __m256i t = _mm256_alignr_epi8( v1, v2, 31 ); \
-   v2 = _mm256_alignr_epi8( v2, v1, 31 ); \
+   __m256i t = _mm256_permute2x128( v1, v2, 0x03 ); \
+   v2 = _mm256__mm256_permute2x128( v2, v1, 0x21 ); \
   v1 = t; \
 } while(0)

 #endif // __AVX2__
-#endif // __AVX__
 #endif // SIMD_256_H__

--- a/simd-utils/simd-512.h
+++ b/simd-utils/simd-512.h
@@ -37,74 +37,84 @@
 //
 // Experimental, not fully tested.

-//
-// Pseudo constants.
-//
-// Vector constants are not really constants and can't be used as compile time
-// initializers. They contain executable instructions to generate values at
-// run time. They are very slow. If the same constant will be used repeatedly
-// in a function it's better to define it once in a local register variable
-// and use the variable for references.
-// Tthe simpler the constant, the more efficienct it's generation. Zero is
-// the fastest, then all elements set the same, different 64 bit elements,
-// and different smaller elements is the slowest. Caching multiple uses us
-// always faster.
+// Move integer to/from element 0 of vector.

-#define m512_const_256( hi, lo ) \
+#define mm512_mov64_512( n ) _mm512_castsi128_si512( mm128_mov64_128( n ) )
+#define mm512_mov32_512( n ) _mm512_castsi128_si512( mm128_mov32_128( n ) )
+
+#define mm512_mov256_64( a ) mm128_mov128_64( _mm256_castsi512_si128( a ) )
+#define mm512_mov256_32( a ) mm128_mov128_32( _mm256_castsi512_si128( a ) )
+
+
+// Insert and extract integers is a multistage operation.
+// Insert integer into __m128i, then insert __m128i to __m256i, finally
+// insert __256i into __m512i. Reverse the order for extract.
+// Do not use __m512_insert_epi64 or _mm256_insert_epi64 to perform multiple
+// inserts.
+// Avoid small integers for multiple inserts.
+// Shortcuts:
+// Use castsi to reference the low bits of a vector or sub-vector. (free)
+// Use mov to insert integer into low bits of vector or sub-vector. (cheap)
+// Use _mm_insert only to reference the high bits of __m128i. (expensive)
+// Sequence instructions to minimize data dependencies.
+// Use const or const1 only when integer is either immediate or known to be in 
+// a GP register. Use set/set1 when data needs to be loaded from memory or
+// cache.
+
+// Concatenate two 256 bit vectors into one 512 bit vector {hi, lo}
+#define mm512_concat_256( hi, lo ) \
   _mm512_inserti64x4( _mm512_castsi256_si512( lo ), hi, 1 )

-#define m512_const_128( i3, i2, i1, i0 ) \
-   _mm512_inserti64x4( _mm512_castsi256_si512( m256_const_128( i1, i0 ) ), \
-                                               m256_const_128( i3,i2 ), 1 )
-
-#define m512_const_64( i7, i6, i5, i4, i3, i2, i1, i0 ) \
-   m512_const_256( m256_const_64( i7,i6,i5,i4 ), \
-                    m256_const_64( i3,i2,i1,i0 ) )
-
-static inline __m512i m512_const1_256( __m256i v )
+// Equivalent of set, assign 64 bit integers to respective 64 bit elements.
+static inline __m512i m512_const_64( const uint64_t i7, const uint64_t i6,
+                                     const uint64_t i5, const uint64_t i4,
+                                     const uint64_t i3, const uint64_t i2,
+                                     const uint64_t i1, const uint64_t i0 )
 {
-   return _mm512_broadcast_i64x4( v );
+   __m256i hi, lo;
+   __m128i hi1, lo1;
+   lo  = mm256_mov64_256( i0 );
+   lo1 = mm128_mov64_128( i2 );
+   hi  = mm256_mov64_256( i4 );
+   hi1 = mm128_mov64_128( i6 );
+   lo  = _mm256_castsi128_si256(
+         _mm_insert_epi64( _mm256_castsi256_si128( lo ), i1, 1 ) );
+   lo1 = _mm_insert_epi64( lo1, i3, 1 );
+   hi  = _mm256_castsi128_si256(
+         _mm_insert_epi64( _mm256_castsi256_si128( hi ), i5, 1 ) );
+   hi1 = _mm_insert_epi64( hi1, i7, 1 );
+   lo  = _mm256_inserti128_si256( lo, lo1, 1 );
+   hi  = _mm256_inserti128_si256( hi, hi1, 1 );
+   return mm512_concat_256( hi, lo );
 }

-static inline __m512i m512_const1_128( __m128i v )
+// Equivalent of set4, broadcast 256 bits in groups of four 64 bit constants
+// to all 256 bit lanes: {i3,i2,i1,i0,i3,i2,i1,i0,i3,i2,i1,i0,i3,i2,i1,i0}.
+static inline __m512i m512_const4_64( const uint64_t i3, const uint64_t i2,
+                                       const uint64_t i1, const uint64_t i0 )
 {
-   return _mm512_broadcast_i64x2( v );
+   __m256i lo = mm256_mov64_256( i0 );
+   __m128i hi = mm128_mov64_128( i2 );
+   lo = _mm256_castsi128_si256(
+        _mm_insert_epi64( _mm256_castsi256_si128(
+                          lo ), i1, 1 ) );
+   hi = _mm_insert_epi64( hi,   i3, 1 );
+   return _mm512_permutex_epi64( _mm512_castsi256_si512(
+                               _mm256_inserti128_si256( lo, hi, 1 ) ), 0xe4 );
 }

-static inline __m512i m512_const1_64( uint64_t i )
-{
-   __m128i a;
-   asm( "movq %1, %0\n\t"
-        : "=x"(a)
-        : "r"(i) );
-   return _mm512_broadcastq_epi64( a );
-}
+// Broadcast 128 bits in pairs of 64 bit constants {i1. i0} to all
+// 128 bit lanes.
+#define m512_const2_64( i1, i0 ) \
+   _mm512_permutex_epi64( _mm512_castsi128_si512( \
+                          m128_const_64( i1, i0 ) ), 0x44 )

-static inline __m512i m512_const1_32( uint32_t i )
-{
-   __m128i a;
-   asm( "movd %1, %0\n\t"
-        : "=x"(a)
-        : "r"(i) );
-   return _mm512_broadcastd_epi32( a );
-}
+// Equivalent of set1, broadcast 64 bit constant to all 64 bit elements.
+#define m512_const1_64( i ) _mm512_broadcastq_epi64( mm128_mov64_128( i ) )
+#define m512_const1_32( i ) _mm512_broadcastd_epi32( mm128_mov32_128( i ) )
+#define m512_const1_16( i ) _mm512_broadcastw_epi16( mm128_mov32_128( i ) )
+#define m512_const1_8 ( i ) _mm512_broadcastb_epi8 ( mm128_mov32_128( i ) )

-static inline __m512i m512_const1_16( uint16_t i )
-{
-   __m128i a;
-   asm( "movw %1, %0\n\t"
-        : "=x"(a)
-        : "r"(i) );
-   return _mm512_broadcastw_epi16( a );
-}
-static inline __m512i m512_const1_8( uint8_t i )
-{
-   __m128i a;
-   asm( "movb %1, %0\n\t"
-        : "=x"(a)
-        : "r"(i) );
-   return _mm512_broadcastb_epi8( a );
-}

 //
 // Pseudo constants.
@@ -114,105 +124,26 @@ static inline __m512i m512_const1_8( uint8_t i )
 // initialized to zero.

 #define m512_zero       _mm512_setzero_si512()
+#define m512_one_512    mm512_mov64_512( 1 )
+#define m512_one_256    _mm512_broadcast_i64x4 ( mm256_mov64_256( 1 ) )
+#define m512_one_128    _mm512_broadcast_i64x2 ( mm128_mov64_128( 1 ) )
+#define m512_one_64     _mm512_broadcastq_epi64( mm128_mov64_128( 1 ) )
+#define m512_one_32     _mm512_broadcastd_epi32( mm128_mov64_128( 1 ) )
+#define m512_one_16     _mm512_broadcastw_epi16( mm128_mov64_128( 1 ) )
+#define m512_one_8      _mm512_broadcastb_epi8 ( mm128_mov64_128( 1 ) )

-/*
-#define m512_one_512    _mm512_set_epi64(  0ULL, 0ULL, 0ULL, 0ULL, \
-                                           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 )
-*/
-
-static inline __m512i mm512_one_512_fn()
-{
-  __m512i a;
-  const uint64_t one = 1;
-  asm( "movq %1, %0\n\t"
-       : "=x" (a)
-       : "r" (one) );
-  return a;
-}
-#define m512_one_512    mm512_one_512_fn()
-
-static inline __m512i mm512_one_256_fn()
-{
-   __m256i a;
-   const uint64_t one = 1;
-   asm( "movq %1, %0\n\t"
-        : "=x"(a)
-        : "r" (one) );
-   return _mm512_broadcast_i64x4( a );
-}
-#define m512_one_256    mm512_one_256_fn()
-
-static inline __m512i mm512_one_128_fn()
-{
-   __m128i a;
-   const uint64_t one = 1;
-   asm( "movq %1, %0\n\t"
-        : "=x"(a)
-        : "r" (one) );
-   return _mm512_broadcast_i64x2( a );
-}
-#define m512_one_128    mm512_one_128_fn()
-
-static inline __m512i mm512_one_64_fn()
-{
-   __m128i a;
-   const uint64_t one = 1;
-   asm( "movq %1, %0\n\t"
-        : "=x"(a)
-        : "r" (one) );
-   return _mm512_broadcastq_epi64( a );
-}
-#define m512_one_64    mm512_one_64_fn()
-
-static inline __m512i mm512_one_32_fn()
-{
-   __m128i a;
-   const uint64_t one = 0x0000000100000001;
-   asm( "movd %1, %0\n\t"
-        : "=x"(a)
-        : "r" (one) );
-   return _mm512_broadcastq_epi64( a );
-}
-#define m512_one_32    mm512_one_32_fn()
-
-static inline __m512i mm512_one_16_fn()
-{
-   __m128i a;
-   const uint64_t one = 0x0001000100010001;
-   asm( "movd %1, %0\n\t"
-        : "=x"(a)
-        : "r" (one) );
-   return _mm512_broadcastq_epi64( a );
-}
-#define m512_one_16    mm512_one_16_fn()
-
-static inline __m512i mm512_one_8_fn()
-{
-   __m128i a;
-   const uint64_t one = 0x0101010101010101;
-   asm( "movd %1, %0\n\t"
-        : "=x"(a)
-        : "r" (one) );
-   return _mm512_broadcastq_epi64( a );
-}
-#define m512_one_8    mm512_one_8_fn()
+#define m512_neg1 m512_const1_64( 0xffffffffffffffff )

+/* 
+// EVEX vcmpeqq returns a bit mask instead of a vector
 static inline __m512i mm512_neg1_fn()
 {
   __m512i a;
-   asm( "vpcmpeqq %0, %0, %0\n\t"
-        :"=x"(a) );
+   asm( "vpcmpeqq %0, %0, %0\n\t" : "=x"(a) );
   return a;
 }
 #define m512_neg1    mm512_neg1_fn()
-
+*/

 //
 // Basic operations without SIMD equivalent
@@ -222,12 +153,6 @@ static inline __m512i mm512_neg1_fn()
 #define mm512_negate_32( x ) _mm512_sub_epi32( m512_zero, x )  
 #define mm512_negate_16( x ) _mm512_sub_epi16( m512_zero, x )  

-// More efficient to use cast to extract low lanes, it's free.
-#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
@@ -248,6 +173,19 @@ static inline __m512i mm512_neg1_fn()
 // returns p+o as pointer to vector
 #define casto_m512i(p,o) (((__m512i*)(p))+(o))

+//
+// Memory functions
+// n = number of 512 bit (64 byte) vectors
+
+static inline void memset_zero_512( __m512i *dst, const int n )
+{   for ( int i = 0; i < n; i++ ) dst[i] = m512_zero; }
+
+static inline void memset_512( __m512i *dst, const __m512i a, const int n )
+{   for ( int i = 0; i < n; i++ ) dst[i] = a; }
+
+static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
+{   for ( int i = 0; i < n; i ++ ) dst[i] = src[i]; }
+

 // Sum 4 values, fewer dependencies than sequential addition.

@@ -264,19 +202,12 @@ static inline __m512i mm512_neg1_fn()
   _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 ) )
+   _mm512_xor_si512( _mm512_xor_si512( a, b ), _mm512_xor_si512( c, d ) )


-// Vector size conversion
-
-#define mm256_extr_lo256_512( a ) _mm512_castsi512_si256( a )
-#define mm256_extr_hi256_512( a ) _mm512_extracti64x4_epi64( a, 1 )
-
-#define mm512_concat_256( hi, lo ) \
-   _mm512_inserti164x4( _mm512_castsi256_si512( lo ), hi, 1 )

 // Horizontal vector testing
-
+// Returns bit mask
 #define mm512_allbits0( a )    _mm512_cmpeq_epi64_mask( a, m512_zero )
 #define mm512_allbits1( a )    _mm512_cmpeq_epi64_mask( a, m512_neg1 )
 #define mm512_anybits0( a )    _mm512_cmpneq_epi64_mask( a, m512_neg1 )
@@ -294,6 +225,11 @@ static inline __m512i mm512_neg1_fn()
 // _mm512_rolv_epi64, _mm512_rorv_epi64, _mm512_rolv_epi32, _mm512_rorv_epi32
 //

+#define mm512_ror_64 _mm512_ror_epi64
+#define mm512_rol_64 _mm512_rol_epi64
+#define mm512_ror_32 _mm512_ror_epi32
+#define mm512_rol_32 _mm512_rol_epi32
+
 #define mm512_ror_var_64( v, c ) \
   _mm512_or_si512( _mm512_srli_epi64( v, c ), \
                    _mm512_slli_epi64( v, 64-(c) ) )
@@ -331,25 +267,61 @@ static inline __m512i mm512_neg1_fn()
 // Swap bytes in vector elements, vectorized endian conversion.

 #define mm512_bswap_64( v ) \
-   _mm512_shuffle_epi8( v, m512_const_64( \
-                       0x38393A3B3C3D3E3F, 0x3031323334353637, \
-                       0x28292A2B2C2D2E2F, 0x2021222324252627, \
-                       0x18191A1B1C1D1E1F, 0x1011121314151617, \
-                       0x08090A0B0C0D0E0F, 0x0001020304050607 ) )
+   _mm512_shuffle_epi8( v, \
+               m512_const_64( 0x38393a3b3c3d3e3f, 0x3031323334353637, \
+                              0x28292a2b2c2d2e2f, 0x2021222324252627, \
+                              0x18191a1b1c1d1e1f, 0x1011121314151617, \
+                              0x08090a0b0c0d0e0f, 0x0001020304050607 ))

 #define mm512_bswap_32( v ) \
-   _mm512_shuffle_epi8( v, m512_const_64( \
-                       0x3C3D3E3F38393A3B, 0x3435363730313233, \
-                       0x3C3D3E3F38393A3B, 0x3435363730313233, \
-                       0x3C3D3E3F38393A3B, 0x3435363730313233, \
-                       0x3C3D3E3F38393A3B, 0x3435363730313233 ) )
+   _mm512_shuffle_epi8( v, \
+               m512_const_64( 0x3c3d3e3f38393a3b, 0x3435363730313233, \
+                              0x2c2d2e2f28292a2b, 0x2425262720212223, \
+                              0x0c0d0e0f08090a0b, 0x0405060700010203, \
+                              0x1c1d1e1f18191a1b, 0x1415161710111213 ) )

 #define mm512_bswap_16( v ) \
-   _mm512_shuffle_epi8( v, m512_const_64( \
-                       0x3E3F3C3D3A3B3839, 0x3637343532333031, \
-                       0x2E2F2C2D2A2B2829, 0x2627242522232021, \
-                       0x1E1F1C1D1A1B1819, 0x1617141512131011, \
-                       0x0E0F0C0D0A0B0809, 0x0607040502030001 ) )
+   _mm512_shuffle_epi8( v, \
+               m512_const_64( 0x3e3f3c3d3a3b3839, 0x3637343532333031, \
+                              0x2e2f2c2d2a2b2829, 0x2627242522232021, \
+                              0x1e1f1c1d1a1b1819, 0x1617141512131011, \
+                              0x0e0f0c0d0a0b0809, 0x0607040502030001 ) )
+
+// Source and destination are pointers, may point to same memory.
+// 8 lanes of 64 bytes each
+#define mm512_block_bswap_64( d, s ) do \
+{ \
+  __m512i ctl = m512_const_64( 0x38393a3b3c3d3e3f, 0x3031323334353637, \
+                               0x28292a2b2c2d2e2f, 0x2021222324252627, \
+                               0x18191a1b1c1d1e1f, 0x1011121314151617, \
+                               0x08090a0b0c0d0e0f, 0x0001020304050607  ); \
+  casti_m512i( d, 0 ) = _mm512_shuffle_epi8( casti_m512i( s, 0 ), ctl ); \
+  casti_m512i( d, 1 ) = _mm512_shuffle_epi8( casti_m512i( s, 1 ), ctl ); \
+  casti_m512i( d, 2 ) = _mm512_shuffle_epi8( casti_m512i( s, 2 ), ctl ); \
+  casti_m512i( d, 3 ) = _mm512_shuffle_epi8( casti_m512i( s, 3 ), ctl ); \
+  casti_m512i( d, 4 ) = _mm512_shuffle_epi8( casti_m512i( s, 4 ), ctl ); \
+  casti_m512i( d, 5 ) = _mm512_shuffle_epi8( casti_m512i( s, 5 ), ctl ); \
+  casti_m512i( d, 6 ) = _mm512_shuffle_epi8( casti_m512i( s, 6 ), ctl ); \
+  casti_m512i( d, 7 ) = _mm512_shuffle_epi8( casti_m512i( s, 7 ), ctl ); \
+} while(0)
+
+// 16 lanes of 32 bytes each
+#define mm512_block_bswap_32( d, s ) do \
+{ \
+  __m512i ctl = m512_const_64( 0x3c3d3e3f38393a3b, 0x3435363730313233, \
+                               0x2c2d2e2f28292a2b, 0x2425262720212223, \
+                               0x0c0d0e0f08090a0b, 0x0405060700010203, \
+                               0x1c1d1e1f18191a1b, 0x1415161710111213 ); \
+  casti_m512i( d, 0 ) = _mm512_shuffle_epi8( casti_m512i( s, 0 ), ctl ); \
+  casti_m512i( d, 1 ) = _mm512_shuffle_epi8( casti_m512i( s, 1 ), ctl ); \
+  casti_m512i( d, 2 ) = _mm512_shuffle_epi8( casti_m512i( s, 2 ), ctl ); \
+  casti_m512i( d, 3 ) = _mm512_shuffle_epi8( casti_m512i( s, 3 ), ctl ); \
+  casti_m512i( d, 4 ) = _mm512_shuffle_epi8( casti_m512i( s, 4 ), ctl ); \
+  casti_m512i( d, 5 ) = _mm512_shuffle_epi8( casti_m512i( s, 5 ), ctl ); \
+  casti_m512i( d, 6 ) = _mm512_shuffle_epi8( casti_m512i( s, 6 ), ctl ); \
+  casti_m512i( d, 7 ) = _mm512_shuffle_epi8( casti_m512i( s, 7 ), ctl ); \
+} while(0)
+

 //
 // Rotate elements in 512 bit vector.
@@ -367,8 +339,10 @@ static inline __m512i mm512_neg1_fn()

 // Generic for odd rotations
 #define mm512_ror_x64( v, n )      _mm512_alignr_epi64( v, v, n )
+#define mm512_rol_x64( v, n )      _mm512_alignr_epi64( v, v, 8-n )

 #define mm512_ror_x32( v, n )      _mm512_alignr_epi32( v, v, n )
+#define mm512_rol_x32( v, n )      _mm512_alignr_epi32( v, v, 16-n )

 #define mm512_ror_1x16( v ) \
   _mm512_permutexvar_epi16( m512_const_64( \
@@ -400,7 +374,11 @@ static inline __m512i mm512_neg1_fn()


 // Invert vector: {3,2,1,0} -> {0,1,2,3}
-#define mm512_invert_128( v ) _mm512_shuffle_i64x2( v, v, 0x1b )
+#define mm512_invert_256( v ) \
+   _mm512_permutexvar_epi64( v, m512_const_64( 3,2,1,0,7,6,5,4 ) )
+
+#define mm512_invert_128( v ) \
+   _mm512_permutexvar_epi64( v, m512_const_64( 1,0,3,2,5,4,7,6 ) )

 #define mm512_invert_64( v ) \
   _mm512_permutexvar_epi64( v, m512_const_64( 0,1,2,3,4,5,6,7 ) )
@@ -436,86 +414,66 @@ static inline __m512i mm512_neg1_fn()
 #define mm512_ror1x64_256( v )   _mm512_permutex_epi64( v, 0x39 )
 #define mm512_rol1x64_256( v )   _mm512_permutex_epi64( v, 0x93 )

+
+/*  Need to fix 
 // Rotate 256 bit lanes by one 32 bit element
 #define mm512_ror1x32_256( v ) \
-   _mm512_permutexvar_epi32( m512_const_64( \
-                      0x000000080000000f, 0x0000000e0000000d, \
-                      0x0000000c0000000b, 0x0000000a00000009, \
+   _mm512_permutexvar_epi32( m512_const4_64( \
                      0x0000000000000007, 0x0000000600000005, \
-                      0x0000000400000003, 0x0000000200000001, v ) )
-            
+                      0x0000000400000003, 0x0000000200000001 ), v )
+
 #define mm512_rol1x32_256( v ) \
-   _mm512_permutexvar_epi32( m512_const_64( \
-                      0x0000000e0000000d, 0x0000000c0000000b, \
-                      0x0000000a00000009, 0x000000080000000f, \
+   _mm512_permutexvar_epi32( m512_const4_64( \
                      0x0000000600000005, 0x0000000400000003, \
                      0x0000000200000001, 0x0000000000000007 ), v )

 #define mm512_ror1x16_256( v ) \
-   _mm512_permutexvar_epi16( m512_const_64( \
-                     0x0010001F001E001D, 0x001C001B001A0019, \
-                     0x0018001700160015, 0x0014001300120011, \
-                     0x0000000F000E000D, 0x000C000B000A0009, \
+   _mm512_permutexvar_epi16( m512_const4_64( \
+                     0x0000000f000e000d, 0x000c000b000a0009, \
                     0x0008000700060005, 0x0004000300020001 ), v )

 #define mm512_rol1x16_256( v ) \
-   _mm512_permutexvar_epi16( m512_const_64( \
-                     0x001E001D001C001B, 0x001A001900180017, \
-                     0x0016001500140013, 0x001200110000000F, \
-                     0x000E000D000C000B, 0x000A000900080007, \
-                     0x0006000500040003, 0x000200010000001F ), v )
+   _mm512_permutexvar_epi16( m512_const4_64( \
+                     0x000e000d000c000b, 0x000a000900080007, \
+                     0x0006000500040003, 0x000200010000000f ), v )

 #define mm512_ror1x8_256( v ) \
-   _mm512_shuffle_epi8( v, m512_const_64( \
-                     0x203F3E3D3C3B3A39, 0x3837363534333231, \
-                     0x302F2E2D2C2B2A29, 0x2827262524232221, \
-                     0x001F1E1D1C1B1A19, 0x1817161514131211, \
-                     0x100F0E0D0C0B0A09, 0x0807060504030201 ) )
+   _mm512_shuffle_epi8( v, m512_const4_64( \
+                     0x001f1e1d1c1b1a19, 0x1817161514131211, \
+                     0x100f0e0d0c0b0a09, 0x0807060504030201 ), v )

 #define mm512_rol1x8_256( v ) \
-    _mm512_shuffle_epi8( v, m512_const_64( \
-                     0x3E3D3C3B3A393837, 0x363534333231302F, \
-                     0x2E2D2C2B2A292827, 0x262524232221203F, \
-                     0x1E1D1C1B1A191817, 0x161514131211100F, \
-                     0x0E0D0C0B0A090807, 0x060504030201001F ))
-
+    _mm512_shuffle_epi8( v, m512_const4_64( \
+                     0x1e1d1c1b1a191817, 0x161514131211100f, \
+                     0x0e0d0c0b0a090807, 0x060504030201001f ), v )
+*/
 //
 // Rotate elements within 128 bit lanes of 512 bit vector.

 // Swap hi & lo 64 bits in each 128 bit lane
-#define mm512_swap64_128( v )    _mm512_permutex_epi64( v, 0xb1 )
+#define mm512_swap64_128( v )    _mm512_shuffle_epi32( v, 0x4e )

 // Rotate 128 bit lanes by one 32 bit element
 #define mm512_ror1x32_128( v )   _mm512_shuffle_epi32( v, 0x39 )
 #define mm512_rol1x32_128( v )   _mm512_shuffle_epi32( v, 0x93 )

+/*
 #define mm512_ror1x16_128( v ) \
-    _mm512_permutexvar_epi16( m512_const_64( \
-                     0x0018001F001E001D, 0x001C001B001A0019, \
-                     0x0010001700160015, 0x0014001300120011, \
-                     0x0008000F000E000D, 0x000C000B000A0009, \
-                     0x0000000700060005, 0x0004000300020001 ), v )
+    _mm512_permutexvar_epi16( m512_const2_64( \
+                     0x0000000700060005, 0x0004000300020001 ), v ) 

 #define mm512_rol1x16_128( v ) \
-    _mm512_permutexvar_epi16( m512_const_64( \
-                     0x001E001D001C001B, 0x001A00190018001F, \
-                     0x0016001500140013, 0x0012001100100017, \
-                     0x000E000D000C000B, 0x000A00090008000F, \
-                     0x0006000500040003, 0x0002000100000007, v ) )
+    _mm512_permutexvar_epi16( m512_const2_64( \
+                     0x0006000500040003, 0x0002000100000007 ), v ) 

 #define mm512_ror1x8_128( v ) \
-    _mm512_shuffle_epi8( v, m512_const_64( \
-                     0x303F3E3D3C3B3A39, 0x3837363534333231, \
-                     0x202F2E2D2C2B2A29, 0x2827262524232221, \
-                     0x101F1E1D1C1B1A19, 0x1817161514131211, \
-                     0x000F0E0D0C0B0A09, 0x0807060504030201 ) )
+    _mm512_shuffle_epi8( v, m512_const2_64( \
+                     0x000f0e0d0c0b0a09, 0x0807060504030201 ) )

 #define mm512_rol1x8_128( v ) \
-    _mm512_shuffle_epi8( v, m512_const_64( \
-                     0x3E3D3C3B3A393837, 0x363534333231303F, \
-                     0x2E2D2C2B2A292827, 0x262524232221202F, \
-                     0x1E1D1C1B1A191817, 0x161514131211101F, \
-                     0x0E0D0C0B0A090807, 0x060504030201000F ) )
+    _mm512_shuffle_epi8( v, m512_const2_64( \
+                     0x0e0d0c0b0a090807, 0x060504030201000f ) )
+*/

 // Rotate 128 bit lanes by c bytes.  
 #define mm512_bror_128( v, c ) \
@@ -652,33 +610,5 @@ do { \
   v1 = t; \
 } while(0)

-#define mm512_ror1x16_1024( v1, v2 ) \
-do { \
-   __m512i t = _mm512_alignr_epi8( v1, v2, 2 ); \
-   v1 = _mm512_alignr_epi8( v2, v1, 2 ); \
-   v2 = t; \
-} while(0)
-
-#define mm512_rol1x16_1024( v1, v2 ) \
-do { \
-   __m512i t = _mm512_alignr_epi8( v1, v2, 62 ); \
-   v2 = _mm512_alignr_epi8( v2, v1, 62 ); \
-   v1 = t; \
-} while(0)
-
-#define mm512_ror1x8_1024( v1, v2 ) \
-do { \
-   __m512i t = _mm512_alignr_epi8( v1, v2, 1 ); \
-   v1 = _mm512_alignr_epi8( v2, v1, 1 ); \
-   v2 = t; \
-} while(0)
-
-#define mm512_rol1x8_1024( v1, v2 ) \
-do { \
-   __m512i t = _mm512_alignr_epi8( v1, v2, 63 ); \
-   v2 = _mm512_alignr_epi8( v2, v1, 63 ); \
-   v1 = t; \
-} while(0)
-
 #endif // AVX512
 #endif // SIMD_512_H__
--- a/sysinfos.c
+++ b/sysinfos.c
@@ -18,14 +18,47 @@

 #ifndef WIN32

+// 1035g1: /sys/devices/platform/coretemp.0/hwmon/hwmon3/temp1_input
+// 1035g1: /sys/class/hwmon/hwmon1/temp1_input wrong temp
+// ryzen has no /sys/devices/platform/coretemp.0
+// ryzen: /sys/class/hwmon/hwmon0
+// 2400: /sys/class/hwmon/hwmon0/temp1_input incorrect temp
+// 2400 has no /sys/class/hwmon/hwmon2/temp1_input
+// 2400 /sys/devices/platform/coretemp.0/hwmon/hwmon1/temp1_input ok
+// 6700 /sys/devices/platform/coretemp.0/hwmon/hwmon2/temp1_input
+// 6700 /sys/class/hwmon/hwmon2/temp1_input
+// /sys/devices/platform/coretemp.0/hwmon/hwmon0/temp2_input never exists
+// /sys/class/hwmon/hwmon0/temp2_input doesn't exist or shows wrong temp (sys16)
+// /sys/class/hwmon/hwmon0/device/temp1_input doesn't exist
+
+
+// the first 3 will find i5-2400, i7-6700k, r7-1700, i5-1035g1.
+// The others are left in for legacy, some should probably be removed.
+#define HWMON_PATH1 \
+   "/sys/devices/platform/coretemp.0/hwmon/hwmon3/temp1_input"
+
+#define HWMON_PATH2 \
+   "/sys/devices/platform/coretemp.0/hwmon/hwmon1/temp1_input"
+
+#define HWMON_PATH3 \
+   "/sys/class/hwmon/hwmon0/temp1_input"
+
 #define HWMON_PATH \
 "/sys/class/hwmon/hwmon2/temp1_input"
+
+/*
 #define HWMON_ALT \
 "/sys/class/hwmon/hwmon0/temp1_input"
+
 #define HWMON_ALT1 \
 "/sys/devices/platform/coretemp.0/hwmon/hwmon1/temp1_input"
+*/
+
+// This shows wrong temp on i5-1035g1
 #define HWMON_ALT2 \
 "/sys/class/hwmon/hwmon1/temp1_input"
+
+// None of these work on any of the cpus above.
 #define HWMON_ALT3 \
 "/sys/devices/platform/coretemp.0/hwmon/hwmon0/temp2_input"
 #define HWMON_ALT4 \
@@ -33,16 +66,28 @@
 #define HWMON_ALT5 \
 "/sys/class/hwmon/hwmon0/device/temp1_input"

+
 static inline float linux_cputemp(int core)
 {
 	float tc = 0.0;
-	FILE *fd = fopen(HWMON_PATH, "r");
+	FILE *fd;
 	uint32_t val = 0;

-	if (!fd)
-		fd = fopen(HWMON_ALT, "r");
+   fd = fopen(HWMON_PATH1, "r");
+
+   if (!fd)
+      fd = fopen(HWMON_PATH2, "r");
+
+   if (!fd)
+      fd = fopen(HWMON_PATH3, "r");
+
+   if (!fd)
+      fd = fopen(HWMON_PATH, "r");

 	if (!fd)
+//		fd = fopen(HWMON_ALT1, "r");
+
+//	if (!fd)
 		fd = fopen(HWMON_ALT2, "r");

 	if (!fd)
@@ -52,14 +97,14 @@ static inline float linux_cputemp(int core)
 		fd = fopen(HWMON_ALT4, "r");

 	if (!fd)
-                fd = fopen(HWMON_ALT5, "r");
+      fd = fopen(HWMON_ALT5, "r");

 	if (!fd)
 		return tc;

-	if (fscanf(fd, "%d", &val))
+	if ( fscanf( fd, "%d", &val ) )
 		tc = val / 1000.0;
-	fclose(fd);
+	fclose( fd );
 	return tc;
 }

@@ -296,7 +341,7 @@ static inline void cpu_getmodelid(char *outbuf, size_t maxsz)
 // EXTENDED_FEATURES ECX
 #define AVX512VBMI_Flag  (1<<1) 
 #define AVX512VBMI2_Flag (1<<6)
-#define AVX512VAES_Flag  (1<<9)
+#define VAES_Flag        (1<<9)


 // Use this to detect presence of feature
@@ -418,14 +463,14 @@ static inline bool has_avx512()
 #endif
 }

-static inline bool has_avx512vaes()
+static inline bool has_vaes()
 {
 #ifdef __arm__
    return false;
 #else
    int cpu_info[4] = { 0 };
    cpuid( EXTENDED_FEATURES, cpu_info );
-    return cpu_info[ ECX_Reg ] & AVX512VAES_Flag;
+    return cpu_info[ ECX_Reg ] & VAES_Flag;
 #endif
 }

--- a/winbuild-cross.sh
+++ b/winbuild-cross.sh
@@ -41,27 +41,22 @@ make -j 16
 strip -s cpuminer.exe
 mv cpuminer.exe release/cpuminer-zen.exe

-#make clean || echo clean
-#CFLAGS="-O3 -march=corei7-avx -msha -Wall" ./configure $CONFIGURE_ARGS
-#make
-#strip -s cpuminer.exe
-#mv cpuminer.exe release/cpuminer-avx-sha.exe
+# mingw won't compile avx512 without -fno-asynchronous-unwind-tables
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -march=skylake-avx512 -Wall -fno-asynchronous-unwind-tables" ./configure $CONFIGURE_ARGS
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe release/cpuminer-avx512.exe

 make clean || echo clean
 rm -f config.status
-CFLAGS="-O3 -march=core-avx2 -Wall" ./configure $CONFIGURE_ARGS
+# GCC 9 doesn't include AES in core-avx2
+CFLAGS="-O3 -march=core-avx2 -maes -Wall" ./configure $CONFIGURE_ARGS
 make -j 16
 strip -s cpuminer.exe
 mv cpuminer.exe release/cpuminer-avx2.exe

-#make clean || echo clean
-#rm -f config.status
-#CFLAGS="-O3 -march=znver1 -Wall" ./configure $CONFIGURE_ARGS
-#make -j 
-#strip -s cpuminer.exe
-#mv cpuminer.exe release/cpuminer-aes-sha.exe
-
-
 make clean || echo clean
 rm -f config.status
 CFLAGS="-O3 -march=corei7-avx -Wall" ./configure $CONFIGURE_ARGS 
--- a/winbuild-cross.sh.bak
+++ b/winbuild-cross.sh.bak
@@ -0,0 +1,103 @@
+#!/bin/bash
+#
+# Script for building Windows binaries release package using mingw.
+# Requires a custom mingw environment, not intended for users.
+#
+# Compiles Windows EXE files for selected CPU architectures, copies them
+# as well as some DLLs that aren't available in most Windows environments
+# into a release folder ready to be zipped and uploaded.
+
+# define some local variables
+
+export LOCAL_LIB="$HOME/usr/lib"
+
+export LDFLAGS="-L$LOCAL_LIB/curl/lib/.libs -L$LOCAL_LIB/gmp/.libs -L$LOCAL_LIB/openssl"
+
+export CONFIGURE_ARGS="--with-curl=$LOCAL_LIB/curl --with-crypto=$LOCAL_LIB/openssl --host=x86_64-w64-mingw32"
+
+# make link to local gmp header file.
+ln -s $LOCAL_LIB/gmp/gmp.h ./gmp.h
+
+# edit configure to fix pthread lib name for Windows.
+#sed -i 's/"-lpthread"/"-lpthreadGC2"/g' configure.ac
+
+# make release directory and copy selected DLLs.
+mkdir release
+cp README.txt release/
+cp README.md release/
+cp RELEASE_NOTES release/
+cp /usr/x86_64-w64-mingw32/lib/zlib1.dll release/
+cp /usr/x86_64-w64-mingw32/lib/libwinpthread-1.dll release/
+cp /usr/lib/gcc/x86_64-w64-mingw32/7.3-win32/libstdc++-6.dll release/
+cp /usr/lib/gcc/x86_64-w64-mingw32/7.3-win32/libgcc_s_seh-1.dll release/
+cp $LOCAL_LIB/openssl/libcrypto-1_1-x64.dll release/
+cp $LOCAL_LIB/curl/lib/.libs/libcurl-4.dll release/
+
+make distclean || echo clean
+rm -f config.status
+./autogen.sh || echo done
+CFLAGS="-O3 -march=znver1 -Wall" ./configure $CONFIGURE_ARGS
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe release/cpuminer-zen.exe
+
+#make clean || echo clean
+#CFLAGS="-O3 -march=corei7-avx -msha -Wall" ./configure $CONFIGURE_ARGS
+#make
+#strip -s cpuminer.exe
+#mv cpuminer.exe release/cpuminer-avx-sha.exe
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -march=core-avx2 -Wall" ./configure $CONFIGURE_ARGS
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe release/cpuminer-avx2.exe
+
+#make clean || echo clean
+#rm -f config.status
+#CFLAGS="-O3 -march=znver1 -Wall" ./configure $CONFIGURE_ARGS
+#make -j 
+#strip -s cpuminer.exe
+#mv cpuminer.exe release/cpuminer-aes-sha.exe
+
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -march=corei7-avx -Wall" ./configure $CONFIGURE_ARGS 
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe release/cpuminer-avx.exe
+
+# -march=westmere is supported in gcc5
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -march=westmere -Wall" ./configure $CONFIGURE_ARGS
+#CFLAGS="-O3 -maes -msse4.2 -Wall" ./configure $CONFIGURE_ARGS
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe release/cpuminer-aes-sse42.exe
+
+#make clean || echo clean
+#rm -f config.status
+#CFLAGS="-O3 -march=corei7 -Wall" ./configure $CONFIGURE_ARGS
+#make 
+#strip -s cpuminer.exe
+#mv cpuminer.exe release/cpuminer-sse42.exe
+
+#make clean || echo clean
+#rm -f config.status
+#CFLAGS="-O3 -march=core2 -Wall" ./configure $CONFIGURE_ARGS
+#make 
+#strip -s cpuminer.exe
+#mv cpuminer.exe release/cpuminer-ssse3.exe
+#make clean || echo clean
+
+make clean || echo clean
+rm -f config.status
+CFLAGS="-O3 -msse2 -Wall" ./configure $CONFIGURE_ARGS
+make -j 16
+strip -s cpuminer.exe
+mv cpuminer.exe release/cpuminer-sse2.exe
+make clean || echo clean
+
Author	SHA1	Message	Date
Jay D Dee	40039386a0	v3.10.0	2019-12-03 12:26:11 -05:00
Jay D Dee	91ec6f1771	v3.9.11	2019-11-26 09:22:03 -05:00
Jay D Dee	a52c5eccf7	v3.9.10	2019-11-22 20:29:18 -05:00