v3.10.6

INSTALL_LINUX

@@ -1,12 +1,14 @@
 
 
-Requirements:
+1. Requirements:
+---------------
 
 Intel Core2 or newer, or AMD Steamroller or newer CPU. ARM CPUs are not
 supported.
 64 bit Linux operating system. Apple is not supported.
 
-Building on linux prerequisites:
+2. Building on linux prerequisites:
+-----------------------------------
 
 It is assumed users know how to install packages on their system and
 be able to compile standard source packages. This is basic Linux and
@@ -20,49 +22,74 @@ http://askubuntu.com/questions/457526/how-to-install-cpuminer-in-ubuntu
 
 Install any additional dependencies needed by cpuminer-opt. The list below
 are some of the ones that may not be in the default install and need to
-be installed manually. There may be others, read the error messages they
-will give a clue as to the missing package.
+be installed manually. There may be others, read the compiler 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 automake libssl-dev libcurl4-openssl-dev libjansson-dev libgmp-dev zlib1g-dev git
 
 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
-compiler version, to CFLAGS:
-"-march=native" or "-march=znver1" or "-msha".
+openssl 1.1.0e or higher. Add one of the following to CFLAGS for SHA
+support depending on your CPU and compiler version:
+
+"-march=native" is always the best choice
+
+"-march=znver1" for Ryzen 1000 & 2000 series, znver2 for 3000.
+
+"-msha"  Add SHA to other tuning options
 
 Additional instructions for static compilalation can be found here:
 https://lxadm.com/Static_compilation_of_cpuminer
 Static builds should only considered in a homogeneous HW and SW environment.
 Local builds will always have the best performance and compatibility.
 
-Extract cpuminer source.
+3. Download cpuminer-opt
+------------------------
 
-tar xvzf cpuminer-opt-x.y.z.tar.gz
-cd cpuminer-opt-x.y.z
+Download the source code for the latest realease from the official repository.
 
-Run ./build.sh to build on Linux or execute the following commands.
+https://github.com/JayDDee/cpuminer-opt/releases
 
-./autogen.sh
-CFLAGS="-O3 -march=native -Wall" ./configure --with-curl
-make
+Extract the source code.
 
-Start mining.
+$ tar xvzf cpuminer-opt-x.y.z.tar.gz
+
+
+Alternatively it can be cloned from git.
+
+$ git clone https://github.com/JayDDee/cpuminer-opt.git
+ 
+4. Build cpuminer-opt
+---------------------
+
+It is recomended to Build with default options, this will usuallly
+produce the best results.
+
+$ ./build.sh to build on Linux or execute the following commands.
+
+or 
+
+$ ./autogen.sh
+$ CFLAGS="-O3 -march=native -Wall" ./configure --with-curl
+$ make -j n
+
+n is the number of threads.
+
+5. Start mining.
+----------------
+
+$ ./cpuminer -a algo -o url -u username -p password
 
-./cpuminer -a algo -o url -u username -p password
 
 Windows
+-------
+
+See also INSTAL_WINDOWS
+
+The following procedure is obsolete and uses an old compiler.
 
 Precompiled Windows binaries are built on a Linux host using Mingw
 with a more recent compiler than the following Windows hosted procedure.

INSTALL_WINDOWS

@@ -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.

Makefile.am

@@ -84,6 +84,7 @@ cpuminer_SOURCES = \
   algo/cubehash/cubehash_sse2.c\
   algo/cubehash/cube-hash-2way.c \
   algo/echo/sph_echo.c \
+  algo/echo/echo-hash-4way.c \
   algo/echo/aes_ni/hash.c\
   algo/gost/sph_gost.c \
   algo/groestl/sph_groestl.c \
@@ -117,12 +118,15 @@ 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 \
   algo/luffa/luffa-hash-2way.c \
   algo/lyra2/lyra2.c \
   algo/lyra2/sponge.c \
+  algo/lyra2/sponge-2way.c \
+  algo/lyra2/lyra2-hash-2way.c \
   algo/lyra2/lyra2-gate.c \
   algo/lyra2/lyra2rev2.c \
   algo/lyra2/lyra2rev2-4way.c \
@@ -173,7 +177,6 @@ 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 +200,9 @@ 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 +282,11 @@ 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 \

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)
@@ -142,6 +144,9 @@ Supported Algorithms
 Errata
 ------
 
+Old algorithms that are no longer used frequently will not have the latest
+optimizations.
+
 Cryptonight and variants are no longer supported, use another miner.
 
 Neoscrypt crashes on Windows, use legacy version.

README.txt

@@ -15,20 +15,29 @@ the features listed at cpuminer startup to ensure you are mining at
 optimum speed using the best available features.
 
 Architecture names and compile options used are only provided for Intel
-Core series. Even the newest Pentium and Celeron CPUs are often missing
-features.
+Core series. Budget CPUs like Pentium and Celeron are often missing the
+latest features.
 
 AMD CPUs older than Piledriver, including Athlon x2 and Phenom II x4, are not
 supported by cpuminer-opt due to an incompatible implementation of SSE2 on
 these CPUs. Some algos may crash the miner with an invalid instruction.
 Users are recommended to use an unoptimized miner such as cpuminer-multi.
 
+More information for Intel and AMD CPU architectures and their features
+can be found on Wikipedia.
+
+https://en.wikipedia.org/wiki/List_of_Intel_CPU_microarchitectures
+
+https://en.wikipedia.org/wiki/List_of_AMD_CPU_microarchitectures
+
+
 Exe name                Compile flags            Arch name
 
 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-avx.exe       "-march=corei7-avx"       Sandybridge
+cpuminer-avx2.exe      "-march=core-avx2 -maes"  Haswell, Skylake, 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:

RELEASE_NOTES

@@ -1,13 +1,17 @@
 cpuminer-opt is a console program run from the command line using the
 keyboard, not the mouse.
 
+See also README.md for list of supported algorithms,
+
 Security warning
 ----------------
 
 Miner programs are often flagged as malware by antivirus programs. This is
-a false positive, they are flagged simply because they are cryptocurrency 
-miners. The source code is open for anyone to inspect. If you don't trust 
-the software, don't use it.
+usually a false positive, they are flagged simply because they are
+cryptocurrency miners. However, some malware has been spread using the
+cover that miners are known to be subject to false positives. Always be on
+alert. The source code of cpuminer-opt is open for anyone to inspect.
+If you don't trust the software don't download it.
 
 The cryptographic hashing code has been taken from trusted sources but has been
 modified for speed at the expense of accepted security practices. This
@@ -25,12 +29,81 @@ Requirements
 Intel Core2 or newer, or AMD Steamroller or newer CPU. ARM CPUs are not
 supported.
 
-64 bit Linux or Windows operating system. Apple and Android are not supported.
-FreeBSD YMMV.
+64 bit Linux or Windows operating system. Apple, Android and Rpi are
+not supported. FreeBSD YMMV.
 
 Change Log
 ----------
 
+v3.10.6
+
+Added support for SSL stratum: stratum+tcps://
+
+Added job id reporting again, but leaner, suppressed with --quiet.
+
+AVX512 for x21s, x22i, lyra2z, allium
+
+Fixed share overflow warnings mining lbry with Ryzen (SHA).
+
+v3.10.5
+
+AVX512 for x17, sonoa, xevan, hmq1725, lyra2rev3, lyra2rev2. 
+Faster hmq1725 AVX2.
+
+v3.10.4
+
+AVX512 for x16r, x16rv2, x16rt, x16s, x16rt-veil (veil).
+
+v3.10.3
+
+AVX512 for x12, x13, x14, x15.
+Fixed x12 AVX2 invalid shares.
+
+v.10.2
+
+AVX512 added for bmw512, c11, phi1612 (phi), qubit, skunk, x11, x11gost (sib).
+Fixed c11 AVX2 invalid shares.
+
+v3.10.1
+
+AVX512 for blake2b, nist5, quark, tribus.
+
+More broken lane fixes, fixed buffer overflow in skein AVX512, fixed
+quark invalid shares AVX2.
+
+Only the highest ranking feature in a class is listed at startup, lower ranking
+features are available but no longer listed.
+
+v3.10.0
+
+AVX512 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 script 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

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;

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;

algo/argon2/argon2d/argon2d/opt.c

@@ -21,7 +21,7 @@
 
 #include "argon2.h"
 #include "core.h"
-
+#include "simd-utils.h"
 #include "../blake2/blake2.h"
 #include "../blake2/blamka-round-opt.h"
 
@@ -37,24 +37,28 @@
 
 #if defined(__AVX512F__)
 
-static void fill_block(__m512i *state, const block *ref_block,
-                       block *next_block, int with_xor) {
+static void fill_block( __m512i *state, const block *ref_block,
+                       block *next_block, int with_xor )
+{
     __m512i block_XY[ARGON2_512BIT_WORDS_IN_BLOCK];
     unsigned int i;
 
-    if (with_xor) {
-        for (i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++) {
-            state[i] = _mm512_xor_si512(
-                state[i], _mm512_loadu_si512((const __m512i *)ref_block->v + i));
-            block_XY[i] = _mm512_xor_si512(
-                state[i], _mm512_loadu_si512((const __m512i *)next_block->v + i));
-        }
-    } else {
-        for (i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++) {
-            block_XY[i] = state[i] = _mm512_xor_si512(
-                state[i], _mm512_loadu_si512((const __m512i *)ref_block->v + i));
+    if ( with_xor )
+    {
+        for ( i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++ )
+        {
+            state[i] = _mm512_xor_si512( state[i],
+                      _mm512_load_si512( (const __m512i*)ref_block->v + i ) );
+            block_XY[i] = _mm512_xor_si512( state[i],
+                      _mm512_load_si512( (const __m512i*)next_block->v + i ) );
         }
     }
+    else
+    {
+        for ( i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++ )
+            block_XY[i] = state[i] = _mm512_xor_si512( state[i],
+                      _mm512_load_si512( (const __m512i*)ref_block->v + i ) );
+    }
 
     BLAKE2_ROUND_1( state[ 0], state[ 1], state[ 2], state[ 3],
                     state[ 4], state[ 5], state[ 6], state[ 7] );
@@ -66,23 +70,10 @@ static void fill_block(__m512i *state, const block *ref_block,
     BLAKE2_ROUND_2( state[ 1], state[ 3], state[ 5], state[ 7],
                     state[ 9], state[11], state[13], state[15] );
 
-/*
-    for (i = 0; i < 2; ++i) {
-        BLAKE2_ROUND_1(
-            state[8 * i + 0], state[8 * i + 1], state[8 * i + 2], state[8 * i + 3],
-            state[8 * i + 4], state[8 * i + 5], state[8 * i + 6], state[8 * i + 7]);
-    }
-
-    for (i = 0; i < 2; ++i) {
-        BLAKE2_ROUND_2(
-            state[2 * 0 + i], state[2 * 1 + i], state[2 * 2 + i], state[2 * 3 + i],
-            state[2 * 4 + i], state[2 * 5 + i], state[2 * 6 + i], state[2 * 7 + i]);
-    }
-*/
-
-    for (i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++) {
-        state[i] = _mm512_xor_si512(state[i], block_XY[i]);
-        _mm512_storeu_si512((__m512i *)next_block->v + i, state[i]);
+    for ( i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++ )
+    {
+        state[i] = _mm512_xor_si512( state[i], block_XY[i] );
+        _mm512_store_si512( (__m512i*)next_block->v + i, state[i] );
     }
 }
 
@@ -125,18 +116,6 @@ static void fill_block(__m256i *state, const block *ref_block,
     BLAKE2_ROUND_2( state[ 3], state[ 7], state[11], state[15],
                     state[19], state[23], state[27], state[31] );
 
-/*
-    for (i = 0; i < 4; ++i) {
-        BLAKE2_ROUND_1(state[8 * i + 0], state[8 * i + 4], state[8 * i + 1], state[8 * i + 5],
-                       state[8 * i + 2], state[8 * i + 6], state[8 * i + 3], state[8 * i + 7]);
-    }
-
-    for (i = 0; i < 4; ++i) {
-        BLAKE2_ROUND_2(state[ 0 + i], state[ 4 + i], state[ 8 + i], state[12 + i],
-                       state[16 + i], state[20 + i], state[24 + i], state[28 + i]);
-    }
-*/
-
     for (i = 0; i < ARGON2_HWORDS_IN_BLOCK; i++) {
         state[i] = _mm256_xor_si256(state[i], block_XY[i]);
         _mm256_store_si256((__m256i *)next_block->v + i, state[i]);
@@ -153,14 +132,14 @@ static void fill_block(__m128i *state, const block *ref_block,
     if (with_xor) {
         for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
             state[i] = _mm_xor_si128(
-                state[i], _mm_loadu_si128((const __m128i *)ref_block->v + i));
+                state[i], _mm_load_si128((const __m128i *)ref_block->v + i));
             block_XY[i] = _mm_xor_si128(
-                state[i], _mm_loadu_si128((const __m128i *)next_block->v + i));
+                state[i], _mm_load_si128((const __m128i *)next_block->v + i));
         }
     } else {
         for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
             block_XY[i] = state[i] = _mm_xor_si128(
-                state[i], _mm_loadu_si128((const __m128i *)ref_block->v + i));
+                state[i], _mm_load_si128((const __m128i *)ref_block->v + i));
         }
     }
 
@@ -198,22 +177,9 @@ static void fill_block(__m128i *state, const block *ref_block,
     BLAKE2_ROUND( state[ 7], state[15], state[23], state[31],  
                   state[39], state[47], state[55], state[63] );
 
-/*
-    for (i = 0; i < 8; ++i) {
-        BLAKE2_ROUND(state[8 * i + 0], state[8 * i + 1], state[8 * i + 2],
-            state[8 * i + 3], state[8 * i + 4], state[8 * i + 5],
-            state[8 * i + 6], state[8 * i + 7]);
-    }
-
-    for (i = 0; i < 8; ++i) {
-        BLAKE2_ROUND(state[8 * 0 + i], state[8 * 1 + i], state[8 * 2 + i],
-            state[8 * 3 + i], state[8 * 4 + i], state[8 * 5 + i],
-            state[8 * 6 + i], state[8 * 7 + i]);
-    }
-*/
     for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
         state[i] = _mm_xor_si128(state[i], block_XY[i]);
-        _mm_storeu_si128((__m128i *)next_block->v + i, state[i]);
+        _mm_store_si128((__m128i *)next_block->v + i, state[i]);
     }
 }
 

algo/argon2/argon2d/blake2/blamka-round-opt.h

@@ -184,10 +184,10 @@ static BLAKE2_INLINE __m128i fBlaMka(__m128i x, __m128i y) {
 
 #include <immintrin.h>
 
-#define  rotr32  mm256_swap32_64
-#define  rotr24  mm256_ror3x8_64
-#define  rotr16  mm256_ror1x16_64
-#define  rotr63( x ) mm256_rol_64( x, 1 )
+#define  rotr32( x )  mm256_ror_64( x, 32 )
+#define  rotr24( x )  mm256_ror_64( x, 24 )
+#define  rotr16( x )  mm256_ror_64( x, 16 )
+#define  rotr63( x )  mm256_rol_64( x,  1 )
 
 //#define rotr32(x)   _mm256_shuffle_epi32(x, _MM_SHUFFLE(2, 3, 0, 1))
 //#define rotr24(x)   _mm256_shuffle_epi8(x, _mm256_setr_epi8(3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10, 3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10))
@@ -427,14 +427,14 @@ static __m512i muladd(__m512i x, __m512i y)
 #define SWAP_QUARTERS(A0, A1) \
     do { \
         SWAP_HALVES(A0, A1); \
-        A0 = _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 1, 4, 5, 2, 3, 6, 7), A0); \
-        A1 = _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 1, 4, 5, 2, 3, 6, 7), A1); \
+        A0 = _mm512_shuffle_i64x2( A0, A0, 0xd8 ); \
+        A1 = _mm512_shuffle_i64x2( A1, A1, 0xd8 ); \
     } while((void)0, 0)
 
 #define UNSWAP_QUARTERS(A0, A1) \
     do { \
-        A0 = _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 1, 4, 5, 2, 3, 6, 7), A0); \
-        A1 = _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 1, 4, 5, 2, 3, 6, 7), A1); \
+        A0 = _mm512_shuffle_i64x2( A0, A0, 0xd8 ); \
+        A1 = _mm512_shuffle_i64x2( A1, A1, 0xd8 ); \
         SWAP_HALVES(A0, A1); \
     } while((void)0, 0)
 

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];    
@@ -71,19 +70,22 @@ typedef struct {
 // Default 14 rounds
 typedef blake_4way_small_context blake256_4way_context;
 void blake256_4way_init(void *ctx);
-void blake256_4way(void *ctx, const void *data, size_t len);
+void blake256_4way_update(void *ctx, const void *data, size_t len);
+#define blake256_4way blake256_4way_update
 void blake256_4way_close(void *ctx, void *dst);
 
 // 14 rounds, blake, decred
 typedef blake_4way_small_context blake256r14_4way_context;
 void blake256r14_4way_init(void *cc);
-void blake256r14_4way(void *cc, const void *data, size_t len);
+void blake256r14_4way_update(void *cc, const void *data, size_t len);
+#define blake256r14_4way blake256r14_4way_update
 void blake256r14_4way_close(void *cc, void *dst);
 
 // 8 rounds, blakecoin, vanilla
 typedef blake_4way_small_context blake256r8_4way_context;
 void blake256r8_4way_init(void *cc);
-void blake256r8_4way(void *cc, const void *data, size_t len);
+void blake256r8_4way_update(void *cc, const void *data, size_t len);
+#define blake256r8_4way blake256r8_4way_update
 void blake256r8_4way_close(void *cc, void *dst);
 
 #ifdef __AVX2__
@@ -93,7 +95,6 @@ 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
@@ -102,38 +103,93 @@ typedef struct {
 // Default 14 rounds
 typedef blake_8way_small_context blake256_8way_context;
 void blake256_8way_init(void *cc);
-void blake256_8way(void *cc, const void *data, size_t len);
+void blake256_8way_update(void *cc, const void *data, size_t len);
+#define blake256_8way blake256_8way_update
 void blake256_8way_close(void *cc, void *dst);
 
 // 14 rounds, blake, decred
 typedef blake_8way_small_context blake256r14_8way_context;
 void blake256r14_8way_init(void *cc);
-void blake256r14_8way(void *cc, const void *data, size_t len);
+void blake256r14_8way_update(void *cc, const void *data, size_t len);
 void blake256r14_8way_close(void *cc, void *dst);
 
 // 8 rounds, blakecoin, vanilla
 typedef blake_8way_small_context blake256r8_8way_context;
 void blake256r8_8way_init(void *cc);
-void blake256r8_8way(void *cc, const void *data, size_t len);
+void blake256r8_8way_update(void *cc, const void *data, size_t len);
+#define blake256r8_8way blake256r8_8way_update
 void blake256r8_8way_close(void *cc, void *dst);
 
 // Blake-512 4 way
 
 typedef struct {
-   __m256i buf[16] __attribute__ ((aligned (64)));
+   __m256i buf[16];
    __m256i H[8];
    __m256i S[4];   
    size_t ptr;
    sph_u64 T0, T1;
-} blake_4way_big_context;
+} blake_4way_big_context __attribute__ ((aligned (128)));
 
 typedef blake_4way_big_context blake512_4way_context;
 
-void blake512_4way_init(void *cc);
-void blake512_4way(void *cc, const void *data, size_t len);
-void blake512_4way_close(void *cc, void *dst);
-void blake512_4way_addbits_and_close(
-	void *cc, unsigned ub, unsigned n, void *dst);
+void blake512_4way_init( blake_4way_big_context *sc );
+void blake512_4way_update( void *cc, const void *data, size_t len );
+#define blake512_4way blake512_4way_update
+void blake512_4way_close( void *cc, void *dst );
+void blake512_4way_addbits_and_close( void *cc, unsigned ub, unsigned n,
+                                      void *dst );
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+//Blake-256 16 way
+
+typedef struct {
+   __m512i buf[16];
+   __m512i H[8];
+   size_t ptr;
+   uint32_t T0, T1;
+   int rounds;   // 14 for blake, 8 for blakecoin & vanilla
+} blake_16way_small_context __attribute__ ((aligned (128)));
+
+// Default 14 rounds
+typedef blake_16way_small_context blake256_16way_context;
+void blake256_16way_init(void *cc);
+void blake256_16way_update(void *cc, const void *data, size_t len);
+void blake256_16way_close(void *cc, void *dst);
+
+// 14 rounds, blake, decred
+typedef blake_16way_small_context blake256r14_16way_context;
+void blake256r14_16way_init(void *cc);
+void blake256r14_16way_update(void *cc, const void *data, size_t len);
+void blake256r14_16way_close(void *cc, void *dst);
+
+// 8 rounds, blakecoin, vanilla
+typedef blake_16way_small_context blake256r8_16way_context;
+void blake256r8_16way_init(void *cc);
+void blake256r8_16way_update(void *cc, const void *data, size_t len);
+void blake256r8_16way_close(void *cc, void *dst);
+
+
+// Blake-512 8 way
+
+typedef struct {
+   __m512i buf[16];
+   __m512i H[8];
+   __m512i S[4];
+   size_t ptr;
+   sph_u64 T0, T1;
+} blake_8way_big_context __attribute__ ((aligned (128)));
+
+typedef blake_8way_big_context blake512_8way_context;
+
+void blake512_8way_init( blake_8way_big_context *sc );
+void blake512_8way_update( void *cc, const void *data, size_t len );
+void blake512_8way_close( void *cc, void *dst );
+void blake512_8way_addbits_and_close( void *cc, unsigned ub, unsigned n,
+                                      void *dst );
+
+#endif  // AVX512
+
 
 #endif  // AVX2
 

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 ),\
@@ -647,7 +634,7 @@ do { \
                               m256_const1_64( 0x082EFA98082EFA98 ) ); \
    VF = _mm256_xor_si256( _mm256_set1_epi32( T1 ), \
                               m256_const1_64( 0xEC4E6C89EC4E6C89 ) ); \
-   shuf_bswap32 = m256_const_64( 0x0c0d0e0f08090a0b, 0x0405060700010203, \
+   shuf_bswap32 = m256_const_64( 0x1c1d1e1f18191a1b, 0x1415161710111213, \
                                  0x0c0d0e0f08090a0b, 0x0405060700010203 ); \
    M0 = _mm256_shuffle_epi8( * buf    , shuf_bswap32 ); \
    M1 = _mm256_shuffle_epi8( *(buf+ 1), shuf_bswap32 ); \
@@ -682,17 +669,155 @@ 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)
 
 
+#endif
+
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// Blaske-256 16 way AVX512
+
+#define GS_16WAY( m0, m1, c0, c1, a, b, c, d ) \
+do { \
+   a = _mm512_add_epi32( _mm512_add_epi32( a, b ), \
+                         _mm512_xor_si512( _mm512_set1_epi32( c1 ), m0 ) ); \
+   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 ), \
+                         _mm512_xor_si512( _mm512_set1_epi32( c0 ), m1 ) ); \
+   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 ROUND_S_16WAY(r)   do { \
+        GS_16WAY(Mx(r, 0), Mx(r, 1), CSx(r, 0), CSx(r, 1), V0, V4, V8, VC); \
+        GS_16WAY(Mx(r, 2), Mx(r, 3), CSx(r, 2), CSx(r, 3), V1, V5, V9, VD); \
+        GS_16WAY(Mx(r, 4), Mx(r, 5), CSx(r, 4), CSx(r, 5), V2, V6, VA, VE); \
+        GS_16WAY(Mx(r, 6), Mx(r, 7), CSx(r, 6), CSx(r, 7), V3, V7, VB, VF); \
+        GS_16WAY(Mx(r, 8), Mx(r, 9), CSx(r, 8), CSx(r, 9), V0, V5, VA, VF); \
+        GS_16WAY(Mx(r, A), Mx(r, B), CSx(r, A), CSx(r, B), V1, V6, VB, VC); \
+        GS_16WAY(Mx(r, C), Mx(r, D), CSx(r, C), CSx(r, D), V2, V7, V8, VD); \
+        GS_16WAY(Mx(r, E), Mx(r, F), CSx(r, E), CSx(r, F), V3, V4, V9, VE); \
+} while (0)
+
+#define DECL_STATE32_16WAY \
+   __m512i H0, H1, H2, H3, H4, H5, H6, H7; \
+   sph_u32 T0, T1;
+
+#define READ_STATE32_16WAY(state) \
+do { \
+   H0 = (state)->H[0]; \
+   H1 = (state)->H[1]; \
+   H2 = (state)->H[2]; \
+   H3 = (state)->H[3]; \
+   H4 = (state)->H[4]; \
+   H5 = (state)->H[5]; \
+   H6 = (state)->H[6]; \
+   H7 = (state)->H[7]; \
+   T0 = (state)->T0; \
+   T1 = (state)->T1; \
+} while (0)
+
+#define WRITE_STATE32_16WAY(state) \
+do { \
+   (state)->H[0] = H0; \
+   (state)->H[1] = H1; \
+   (state)->H[2] = H2; \
+   (state)->H[3] = H3; \
+   (state)->H[4] = H4; \
+   (state)->H[5] = H5; \
+   (state)->H[6] = H6; \
+   (state)->H[7] = H7; \
+   (state)->T0 = T0; \
+   (state)->T1 = T1; \
+} while (0)
+
+#define COMPRESS32_16WAY( rounds ) \
+do { \
+   __m512i M0, M1, M2, M3, M4, M5, M6, M7; \
+   __m512i M8, M9, MA, MB, MC, MD, ME, MF; \
+   __m512i V0, V1, V2, V3, V4, V5, V6, V7; \
+   __m512i V8, V9, VA, VB, VC, VD, VE, VF; \
+   __m512i shuf_bswap32; \
+   V0 = H0; \
+   V1 = H1; \
+   V2 = H2; \
+   V3 = H3; \
+   V4 = H4; \
+   V5 = H5; \
+   V6 = H6; \
+   V7 = H7; \
+   V8 = m512_const1_64( 0x243F6A88243F6A88 ); \
+   V9 = m512_const1_64( 0x85A308D385A308D3 ); \
+   VA = m512_const1_64( 0x13198A2E13198A2E ); \
+   VB = m512_const1_64( 0x0370734403707344 ); \
+   VC = _mm512_xor_si512( _mm512_set1_epi32( T0 ),\
+                              m512_const1_64( 0xA4093822A4093822 ) ); \
+   VD = _mm512_xor_si512( _mm512_set1_epi32( T0 ),\
+                              m512_const1_64( 0x299F31D0299F31D0 ) ); \
+   VE = _mm512_xor_si512( _mm512_set1_epi32( T1 ), \
+                              m512_const1_64( 0x082EFA98082EFA98 ) ); \
+   VF = _mm512_xor_si512( _mm512_set1_epi32( T1 ), \
+                              m512_const1_64( 0xEC4E6C89EC4E6C89 ) ); \
+   shuf_bswap32 = m512_const_64( 0x3c3d3e3f38393a3b, 0x3435363730313233, \
+                                 0x2c2d2e2f28292a2b, 0x2425262720212223, \
+                                 0x1c1d1e1f18191a1b, 0x1415161710111213, \
+                                 0x0c0d0e0f08090a0b, 0x0405060700010203 ); \
+   M0 = _mm512_shuffle_epi8( * buf    , shuf_bswap32 ); \
+   M1 = _mm512_shuffle_epi8( *(buf+ 1), shuf_bswap32 ); \
+   M2 = _mm512_shuffle_epi8( *(buf+ 2), shuf_bswap32 ); \
+   M3 = _mm512_shuffle_epi8( *(buf+ 3), shuf_bswap32 ); \
+   M4 = _mm512_shuffle_epi8( *(buf+ 4), shuf_bswap32 ); \
+   M5 = _mm512_shuffle_epi8( *(buf+ 5), shuf_bswap32 ); \
+   M6 = _mm512_shuffle_epi8( *(buf+ 6), shuf_bswap32 ); \
+   M7 = _mm512_shuffle_epi8( *(buf+ 7), shuf_bswap32 ); \
+   M8 = _mm512_shuffle_epi8( *(buf+ 8), shuf_bswap32 ); \
+   M9 = _mm512_shuffle_epi8( *(buf+ 9), shuf_bswap32 ); \
+   MA = _mm512_shuffle_epi8( *(buf+10), shuf_bswap32 ); \
+   MB = _mm512_shuffle_epi8( *(buf+11), shuf_bswap32 ); \
+   MC = _mm512_shuffle_epi8( *(buf+12), shuf_bswap32 ); \
+   MD = _mm512_shuffle_epi8( *(buf+13), shuf_bswap32 ); \
+   ME = _mm512_shuffle_epi8( *(buf+14), shuf_bswap32 ); \
+   MF = _mm512_shuffle_epi8( *(buf+15), shuf_bswap32 ); \
+   ROUND_S_16WAY(0); \
+   ROUND_S_16WAY(1); \
+   ROUND_S_16WAY(2); \
+   ROUND_S_16WAY(3); \
+   ROUND_S_16WAY(4); \
+   ROUND_S_16WAY(5); \
+   ROUND_S_16WAY(6); \
+   ROUND_S_16WAY(7); \
+   if (rounds == 14) \
+   { \
+      ROUND_S_16WAY(8); \
+      ROUND_S_16WAY(9); \
+      ROUND_S_16WAY(0); \
+      ROUND_S_16WAY(1); \
+      ROUND_S_16WAY(2); \
+      ROUND_S_16WAY(3); \
+   } \
+   H0 = _mm512_xor_si512( _mm512_xor_si512( V8, V0 ), H0 ); \
+   H1 = _mm512_xor_si512( _mm512_xor_si512( V9, V1 ), H1 ); \
+   H2 = _mm512_xor_si512( _mm512_xor_si512( VA, V2 ), H2 ); \
+   H3 = _mm512_xor_si512( _mm512_xor_si512( VB, V3 ), H3 ); \
+   H4 = _mm512_xor_si512( _mm512_xor_si512( VC, V4 ), H4 ); \
+   H5 = _mm512_xor_si512( _mm512_xor_si512( VD, V5 ), H5 ); \
+   H6 = _mm512_xor_si512( _mm512_xor_si512( VE, V6 ), H6 ); \
+   H7 = _mm512_xor_si512( _mm512_xor_si512( VF, V7 ), H7 ); \
+} while (0)
+
 #endif
 
 // Blake-256 4 way
@@ -703,7 +828,6 @@ 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 +836,6 @@ 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 +943,6 @@ 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 +951,6 @@ 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;
@@ -940,6 +1054,179 @@ blake32_8way_close( blake_8way_small_context *sc, unsigned ub, unsigned n,
 
 #endif
 
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+//Blake-256 16 way AVX512
+
+static void
+blake32_16way_init( blake_16way_small_context *sc, const sph_u32 *iv,
+                   const sph_u32 *salt, int rounds )
+{
+   casti_m512i( sc->H, 0 ) = m512_const1_64( 0x6A09E6676A09E667 );
+   casti_m512i( sc->H, 1 ) = m512_const1_64( 0xBB67AE85BB67AE85 );
+   casti_m512i( sc->H, 2 ) = m512_const1_64( 0x3C6EF3723C6EF372 );
+   casti_m512i( sc->H, 3 ) = m512_const1_64( 0xA54FF53AA54FF53A );
+   casti_m512i( sc->H, 4 ) = m512_const1_64( 0x510E527F510E527F );
+   casti_m512i( sc->H, 5 ) = m512_const1_64( 0x9B05688C9B05688C );
+   casti_m512i( sc->H, 6 ) = m512_const1_64( 0x1F83D9AB1F83D9AB );
+   casti_m512i( sc->H, 7 ) = m512_const1_64( 0x5BE0CD195BE0CD19 );
+   sc->T0 = sc->T1 = 0;
+   sc->ptr = 0;
+   sc->rounds = rounds;
+}
+
+static void
+blake32_16way( blake_16way_small_context *sc, const void *data, size_t len )
+{
+   __m512i *vdata = (__m512i*)data;
+   __m512i *buf;
+   size_t ptr;
+   const int buf_size = 64;   // number of elements, sizeof/4
+   DECL_STATE32_16WAY
+   buf = sc->buf;
+   ptr = sc->ptr;
+   if ( len < buf_size - ptr )
+   {
+        memcpy_512( buf + (ptr>>2), vdata, len>>2 );
+        ptr += len;
+        sc->ptr = ptr;
+        return;
+   }
+   READ_STATE32_16WAY(sc);
+   while ( len > 0 )
+   {
+      size_t clen;
+
+      clen = buf_size - ptr;
+      if (clen > len)
+           clen = len;
+      memcpy_512( buf + (ptr>>2), vdata, clen>>2 );
+      ptr += clen;
+      vdata += (clen>>2);
+      len -= clen;
+      if ( ptr == buf_size )
+      {
+          if ( ( T0 = T0 + 512 ) < 512 )
+                T1 = T1 + 1;
+          COMPRESS32_16WAY( sc->rounds );
+          ptr = 0;
+      }
+   }
+   WRITE_STATE32_16WAY(sc);
+   sc->ptr = ptr;
+}
+
+static void
+blake32_16way_close( blake_16way_small_context *sc, unsigned ub, unsigned n,
+                    void *dst, size_t out_size_w32 )
+{
+   __m512i buf[16];
+   size_t ptr;
+   unsigned bit_len;
+   sph_u32 th, tl;
+
+   ptr = sc->ptr;
+   bit_len = ((unsigned)ptr << 3);
+   buf[ptr>>2] = m512_const1_64( 0x0000008000000080ULL );
+   tl = sc->T0 + bit_len;
+   th = sc->T1;
+
+   if ( ptr == 0 )
+   {
+        sc->T0 = 0xFFFFFE00UL;
+        sc->T1 = 0xFFFFFFFFUL;
+   }
+   else if ( sc->T0 == 0 )
+   {
+        sc->T0 = 0xFFFFFE00UL + bit_len;
+        sc->T1 = sc->T1 - 1;
+   }
+   else
+        sc->T0 -= 512 - bit_len;
+
+   if ( ptr <= 52 )
+   {
+       memset_zero_512( buf + (ptr>>2) + 1, (52 - ptr) >> 2 );
+       if ( out_size_w32 == 8 )
+           buf[52>>2] = _mm512_or_si512( buf[52>>2],
+                                m512_const1_64( 0x0100000001000000ULL ) );
+       buf[+56>>2] = mm512_bswap_32( _mm512_set1_epi32( th ) );
+       buf[+60>>2] = mm512_bswap_32( _mm512_set1_epi32( tl ) );
+       blake32_16way( sc, buf + (ptr>>2), 64 - ptr );
+   }
+   else
+   {
+        memset_zero_512( buf + (ptr>>2) + 1, (60-ptr) >> 2 );
+        blake32_16way( sc, buf + (ptr>>2), 64 - ptr );
+        sc->T0 = 0xFFFFFE00UL;
+        sc->T1 = 0xFFFFFFFFUL;
+        memset_zero_512( buf, 56>>2 );
+       if ( out_size_w32 == 8 )
+           buf[52>>2] = m512_const1_64( 0x0100000001000000ULL );
+        buf[56>>2] = mm512_bswap_32( _mm512_set1_epi32( th ) );
+        buf[60>>2] = mm512_bswap_32( _mm512_set1_epi32( tl ) );
+        blake32_16way( sc, buf, 64 );
+   }
+   mm512_block_bswap_32( (__m512i*)dst, (__m512i*)sc->H );
+}
+
+void
+blake256_16way_init(void *cc)
+{
+   blake32_16way_init( cc, IV256, salt_zero_8way_small, 14 );
+}
+
+void
+blake256_16way_update(void *cc, const void *data, size_t len)
+{
+        blake32_16way(cc, data, len);
+}
+
+void
+blake256_16way_close(void *cc, void *dst)
+{
+        blake32_16way_close(cc, 0, 0, dst, 8);
+}
+
+void blake256r14_16way_init(void *cc)
+{
+   blake32_16way_init( cc, IV256, salt_zero_8way_small, 14 );
+}
+
+void
+blake256r14_16way_update(void *cc, const void *data, size_t len)
+{
+   blake32_16way(cc, data, len);
+}
+
+void
+blake256r14_16way_close(void *cc, void *dst)
+{
+   blake32_16way_close(cc, 0, 0, dst, 8);
+}
+
+void blake256r8_16way_init(void *cc)
+{
+   blake32_16way_init( cc, IV256, salt_zero_8way_small, 8 );
+}
+
+void
+blake256r8_16way_update(void *cc, const void *data, size_t len)
+{
+   blake32_16way(cc, data, len);
+}
+
+void
+blake256r8_16way_close(void *cc, void *dst)
+{
+   blake32_16way_close(cc, 0, 0, dst, 8);
+}
+
+#endif // AVX512
+
+
+
 // Blake-256 4 way
 
 // default 14 rounds, backward copatibility
@@ -972,7 +1259,7 @@ blake256_8way_init(void *cc)
 }
 
 void
-blake256_8way(void *cc, const void *data, size_t len)
+blake256_8way_update(void *cc, const void *data, size_t len)
 {
         blake32_8way(cc, data, len);
 }
@@ -992,7 +1279,7 @@ void blake256r14_4way_init(void *cc)
 }
 
 void
-blake256r14_4way(void *cc, const void *data, size_t len)
+blake256r14_4way_update(void *cc, const void *data, size_t len)
 {
    blake32_4way(cc, data, len);
 }
@@ -1011,7 +1298,7 @@ void blake256r14_8way_init(void *cc)
 }
 
 void
-blake256r14_8way(void *cc, const void *data, size_t len)
+blake256r14_8way_update(void *cc, const void *data, size_t len)
 {
    blake32_8way(cc, data, len);
 }
@@ -1031,7 +1318,7 @@ void blake256r8_4way_init(void *cc)
 }
 
 void
-blake256r8_4way(void *cc, const void *data, size_t len)
+blake256r8_4way_update(void *cc, const void *data, size_t len)
 {
    blake32_4way(cc, data, len);
 }
@@ -1050,7 +1337,7 @@ void blake256r8_8way_init(void *cc)
 }
 
 void
-blake256r8_8way(void *cc, const void *data, size_t len)
+blake256r8_8way_update(void *cc, const void *data, size_t len)
 {
    blake32_8way(cc, data, len);
 }

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[49]);   // 3*16+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)
 {

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;
 };
 

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,

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] ] );

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,

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;

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;
 };
 

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,

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 );
@@ -351,9 +463,203 @@ int blake2s_8way_final( blake2s_8way_state *S, void *out, uint8_t outlen )
    return 0;
 }
 
+// Update and final when inlen is a multiple of 64 bytes
+int blake2s_8way_full_blocks( blake2s_8way_state *S, void *out,
+                              const void *input, uint64_t inlen )
+{
+    __m256i *in = (__m256i*)input;
+    __m256i *buf = (__m256i*)S->buf;
+
+    while( inlen > BLAKE2S_BLOCKBYTES )
+    {
+       memcpy_256( 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_8way_compress( S, buf );
+       S->buflen = 0;
+       in += ( BLAKE2S_BLOCKBYTES >> 2 );
+    }
+
+    // last block
+    memcpy_256( 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_8way_compress( S, buf );
+
+    for ( int i = 0; i < 8; ++i )
+      casti_m256i( out, i ) = S->h[ i ];
+    return 0;
+}
 
 #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 )
 {

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
 

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

@@ -42,21 +42,13 @@
 extern "C"{
 #endif
 
-#if SPH_SMALL_FOOTPRINT && !defined SPH_SMALL_FOOTPRINT_BLAKE
-#define SPH_SMALL_FOOTPRINT_BLAKE   1
-#endif
-
-#if SPH_64 && (SPH_SMALL_FOOTPRINT_BLAKE || !SPH_64_TRUE)
-#define SPH_COMPACT_BLAKE_64   1
-#endif
-
 #ifdef _MSC_VER
 #pragma warning (disable: 4146)
 #endif
 
-
-// Blake-512
-
+// Blake-512 common
+   
+/*
 static const sph_u64 IV512[8] = {
 	SPH_C64(0x6A09E667F3BCC908), SPH_C64(0xBB67AE8584CAA73B),
 	SPH_C64(0x3C6EF372FE94F82B), SPH_C64(0xA54FF53A5F1D36F1),
@@ -64,10 +56,7 @@ static const sph_u64 IV512[8] = {
 	SPH_C64(0x1F83D9ABFB41BD6B), SPH_C64(0x5BE0CD19137E2179)
 };
 
-
-#if SPH_COMPACT_BLAKE_32 || SPH_COMPACT_BLAKE_64
-
-// Blake-256 4 & 8 way, Blake-512 4 way
+static const sph_u64 salt_zero_big[4] = { 0, 0, 0, 0 };
 
 static const unsigned sigma[16][16] = {
 	{  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15 },
@@ -88,7 +77,17 @@ static const unsigned sigma[16][16] = {
 	{  2, 12,  6, 10,  0, 11,  8,  3,  4, 13,  7,  5, 15, 14,  1,  9 }
 };
 
-#endif
+static const sph_u64 CB[16] = {
+   SPH_C64(0x243F6A8885A308D3), SPH_C64(0x13198A2E03707344),
+   SPH_C64(0xA4093822299F31D0), SPH_C64(0x082EFA98EC4E6C89),
+   SPH_C64(0x452821E638D01377), SPH_C64(0xBE5466CF34E90C6C),
+   SPH_C64(0xC0AC29B7C97C50DD), SPH_C64(0x3F84D5B5B5470917),
+   SPH_C64(0x9216D5D98979FB1B), SPH_C64(0xD1310BA698DFB5AC),
+   SPH_C64(0x2FFD72DBD01ADFB7), SPH_C64(0xB8E1AFED6A267E96),
+   SPH_C64(0xBA7C9045F12C7F99), SPH_C64(0x24A19947B3916CF7),
+   SPH_C64(0x0801F2E2858EFC16), SPH_C64(0x636920D871574E69)
+
+*/
 
 #define Z00   0
 #define Z01   1
@@ -264,8 +263,6 @@ static const unsigned sigma[16][16] = {
 #define Mx_(n)      Mx__(n)
 #define Mx__(n)     M ## n
 
-// Blake-512 4 way
-
 #define CBx(r, i)   CBx_(Z ## r ## i)
 #define CBx_(n)     CBx__(n)
 #define CBx__(n)    CB ## n
@@ -287,21 +284,288 @@ static const unsigned sigma[16][16] = {
 #define CBE   SPH_C64(0x0801F2E2858EFC16)
 #define CBF   SPH_C64(0x636920D871574E69)
 
-#if SPH_COMPACT_BLAKE_64
-// not used
-static const sph_u64 CB[16] = {
-	SPH_C64(0x243F6A8885A308D3), SPH_C64(0x13198A2E03707344),
-	SPH_C64(0xA4093822299F31D0), SPH_C64(0x082EFA98EC4E6C89),
-	SPH_C64(0x452821E638D01377), SPH_C64(0xBE5466CF34E90C6C),
-	SPH_C64(0xC0AC29B7C97C50DD), SPH_C64(0x3F84D5B5B5470917),
-	SPH_C64(0x9216D5D98979FB1B), SPH_C64(0xD1310BA698DFB5AC),
-	SPH_C64(0x2FFD72DBD01ADFB7), SPH_C64(0xB8E1AFED6A267E96),
-	SPH_C64(0xBA7C9045F12C7F99), SPH_C64(0x24A19947B3916CF7),
-	SPH_C64(0x0801F2E2858EFC16), SPH_C64(0x636920D871574E69)
-};
+#define READ_STATE64(state)   do { \
+      H0 = (state)->H[0]; \
+      H1 = (state)->H[1]; \
+      H2 = (state)->H[2]; \
+      H3 = (state)->H[3]; \
+      H4 = (state)->H[4]; \
+      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)
 
-#endif
+#define WRITE_STATE64(state)   do { \
+      (state)->H[0] = H0; \
+      (state)->H[1] = H1; \
+      (state)->H[2] = H2; \
+      (state)->H[3] = H3; \
+      (state)->H[4] = H4; \
+      (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)
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// Blake-512 8 way AVX512
+
+#define GB_8WAY(m0, m1, c0, c1, a, b, c, d)   do { \
+   a = _mm512_add_epi64( _mm512_add_epi64( _mm512_xor_si512( \
+                 _mm512_set1_epi64( c1 ), m0 ), b ), a ); \
+   d = mm512_ror_64( _mm512_xor_si512( d, a ), 32 ); \
+   c = _mm512_add_epi64( c, d ); \
+   b = mm512_ror_64( _mm512_xor_si512( b, c ), 25 ); \
+   a = _mm512_add_epi64( _mm512_add_epi64( _mm512_xor_si512( \
+                 _mm512_set1_epi64( c0 ), m1 ), b ), a ); \
+   d = mm512_ror_64( _mm512_xor_si512( d, a ), 16 ); \
+   c = _mm512_add_epi64( c, d ); \
+   b = mm512_ror_64( _mm512_xor_si512( b, c ), 11 ); \
+} while (0)
+
+#define ROUND_B_8WAY(r)   do { \
+   GB_8WAY(Mx(r, 0), Mx(r, 1), CBx(r, 0), CBx(r, 1), V0, V4, V8, VC); \
+   GB_8WAY(Mx(r, 2), Mx(r, 3), CBx(r, 2), CBx(r, 3), V1, V5, V9, VD); \
+   GB_8WAY(Mx(r, 4), Mx(r, 5), CBx(r, 4), CBx(r, 5), V2, V6, VA, VE); \
+   GB_8WAY(Mx(r, 6), Mx(r, 7), CBx(r, 6), CBx(r, 7), V3, V7, VB, VF); \
+   GB_8WAY(Mx(r, 8), Mx(r, 9), CBx(r, 8), CBx(r, 9), V0, V5, VA, VF); \
+   GB_8WAY(Mx(r, A), Mx(r, B), CBx(r, A), CBx(r, B), V1, V6, VB, VC); \
+   GB_8WAY(Mx(r, C), Mx(r, D), CBx(r, C), CBx(r, D), V2, V7, V8, VD); \
+   GB_8WAY(Mx(r, E), Mx(r, F), CBx(r, E), CBx(r, F), V3, V4, V9, VE); \
+   } while (0)
+
+#define DECL_STATE64_8WAY \
+   __m512i H0, H1, H2, H3, H4, H5, H6, H7; \
+        __m512i S0, S1, S2, S3; \
+   sph_u64 T0, T1;
+
+#define COMPRESS64_8WAY   do \
+{ \
+  __m512i M0, M1, M2, M3, M4, M5, M6, M7; \
+  __m512i M8, M9, MA, MB, MC, MD, ME, MF; \
+  __m512i V0, V1, V2, V3, V4, V5, V6, V7; \
+  __m512i V8, V9, VA, VB, VC, VD, VE, VF; \
+  __m512i shuf_bswap64; \
+  V0 = H0; \
+  V1 = H1; \
+  V2 = H2; \
+  V3 = H3; \
+  V4 = H4; \
+  V5 = H5; \
+  V6 = H6; \
+  V7 = H7; \
+  V8 = _mm512_xor_si512( S0, m512_const1_64( CB0 ) );  \
+  V9 = _mm512_xor_si512( S1, m512_const1_64( CB1 ) );  \
+  VA = _mm512_xor_si512( S2, m512_const1_64( CB2 ) );  \
+  VB = _mm512_xor_si512( S3, m512_const1_64( CB3 ) );  \
+  VC = _mm512_xor_si512( _mm512_set1_epi64( T0 ), \
+                         m512_const1_64( CB4 ) );  \
+  VD = _mm512_xor_si512( _mm512_set1_epi64( T0 ), \
+                         m512_const1_64( CB5 ) );  \
+  VE = _mm512_xor_si512( _mm512_set1_epi64( T1 ), \
+                         m512_const1_64( CB6 ) );  \
+  VF = _mm512_xor_si512( _mm512_set1_epi64( T1 ), \
+                         m512_const1_64( CB7 ) );  \
+  shuf_bswap64 = m512_const_64( 0x38393a3b3c3d3e3f, 0x3031323334353637, \
+                                0x28292a2b2c2d2e2f, 0x2021222324252627, \
+                                0x18191a1b1c1d1e1f, 0x1011121314151617, \
+                                0x08090a0b0c0d0e0f, 0x0001020304050607 ); \
+  M0 = _mm512_shuffle_epi8( *(buf+ 0), shuf_bswap64 ); \
+  M1 = _mm512_shuffle_epi8( *(buf+ 1), shuf_bswap64 ); \
+  M2 = _mm512_shuffle_epi8( *(buf+ 2), shuf_bswap64 ); \
+  M3 = _mm512_shuffle_epi8( *(buf+ 3), shuf_bswap64 ); \
+  M4 = _mm512_shuffle_epi8( *(buf+ 4), shuf_bswap64 ); \
+  M5 = _mm512_shuffle_epi8( *(buf+ 5), shuf_bswap64 ); \
+  M6 = _mm512_shuffle_epi8( *(buf+ 6), shuf_bswap64 ); \
+  M7 = _mm512_shuffle_epi8( *(buf+ 7), shuf_bswap64 ); \
+  M8 = _mm512_shuffle_epi8( *(buf+ 8), shuf_bswap64 ); \
+  M9 = _mm512_shuffle_epi8( *(buf+ 9), shuf_bswap64 ); \
+  MA = _mm512_shuffle_epi8( *(buf+10), shuf_bswap64 ); \
+  MB = _mm512_shuffle_epi8( *(buf+11), shuf_bswap64 ); \
+  MC = _mm512_shuffle_epi8( *(buf+12), shuf_bswap64 ); \
+  MD = _mm512_shuffle_epi8( *(buf+13), shuf_bswap64 ); \
+  ME = _mm512_shuffle_epi8( *(buf+14), shuf_bswap64 ); \
+  MF = _mm512_shuffle_epi8( *(buf+15), shuf_bswap64 ); \
+  ROUND_B_8WAY(0); \
+  ROUND_B_8WAY(1); \
+  ROUND_B_8WAY(2); \
+  ROUND_B_8WAY(3); \
+  ROUND_B_8WAY(4); \
+  ROUND_B_8WAY(5); \
+  ROUND_B_8WAY(6); \
+  ROUND_B_8WAY(7); \
+  ROUND_B_8WAY(8); \
+  ROUND_B_8WAY(9); \
+  ROUND_B_8WAY(0); \
+  ROUND_B_8WAY(1); \
+  ROUND_B_8WAY(2); \
+  ROUND_B_8WAY(3); \
+  ROUND_B_8WAY(4); \
+  ROUND_B_8WAY(5); \
+  H0 = mm512_xor4( V8, V0, S0, H0 ); \
+  H1 = mm512_xor4( V9, V1, S1, H1 ); \
+  H2 = mm512_xor4( VA, V2, S2, H2 ); \
+  H3 = mm512_xor4( VB, V3, S3, H3 ); \
+  H4 = mm512_xor4( VC, V4, S0, H4 ); \
+  H5 = mm512_xor4( VD, V5, S1, H5 ); \
+  H6 = mm512_xor4( VE, V6, S2, H6 ); \
+  H7 = mm512_xor4( VF, V7, S3, H7 ); \
+} while (0)
+
+void blake512_8way_init( blake_8way_big_context *sc )
+{
+   __m512i zero = m512_zero;
+   casti_m512i( sc->H, 0 ) = m512_const1_64( 0x6A09E667F3BCC908 );
+   casti_m512i( sc->H, 1 ) = m512_const1_64( 0xBB67AE8584CAA73B );
+   casti_m512i( sc->H, 2 ) = m512_const1_64( 0x3C6EF372FE94F82B );
+   casti_m512i( sc->H, 3 ) = m512_const1_64( 0xA54FF53A5F1D36F1 );
+   casti_m512i( sc->H, 4 ) = m512_const1_64( 0x510E527FADE682D1 );
+   casti_m512i( sc->H, 5 ) = m512_const1_64( 0x9B05688C2B3E6C1F );
+   casti_m512i( sc->H, 6 ) = m512_const1_64( 0x1F83D9ABFB41BD6B );
+   casti_m512i( sc->H, 7 ) = m512_const1_64( 0x5BE0CD19137E2179 );
+
+   casti_m512i( sc->S, 0 ) = zero;
+   casti_m512i( sc->S, 1 ) = zero;
+   casti_m512i( sc->S, 2 ) = zero;
+   casti_m512i( sc->S, 3 ) = zero;
+
+   sc->T0 = sc->T1 = 0;
+   sc->ptr = 0;
+}
+
+static void
+blake64_8way( blake_8way_big_context *sc, const void *data, size_t len )
+{
+   __m512i *vdata = (__m512i*)data;
+   __m512i *buf;
+   size_t ptr;
+   DECL_STATE64_8WAY
+
+   const int buf_size = 128;  //  sizeof/8
+
+   buf = sc->buf;
+   ptr = sc->ptr;
+   if ( len < (buf_size - ptr) )
+   {
+   memcpy_512( buf + (ptr>>3), vdata, len>>3 );
+   ptr += len;
+   sc->ptr = ptr;
+   return;
+   }
+
+   READ_STATE64(sc);
+   while ( len > 0 )
+   {
+   size_t clen;
+
+   clen = buf_size - ptr;
+   if ( clen > len )
+      clen = len;
+   memcpy_512( buf + (ptr>>3), vdata, clen>>3 );
+   ptr += clen;
+   vdata = vdata + (clen>>3);
+   len -= clen;
+   if ( ptr == buf_size )
+        {
+      if ( ( T0 = SPH_T64(T0 + 1024) ) < 1024 )
+         T1 = SPH_T64(T1 + 1);
+      COMPRESS64_8WAY;
+      ptr = 0;
+   }
+   }
+   WRITE_STATE64(sc);
+   sc->ptr = ptr;
+}
+
+static void
+blake64_8way_close( blake_8way_big_context *sc, void *dst )
+{
+   __m512i buf[16];
+   size_t ptr;
+   unsigned bit_len;
+//   uint64_t z, zz;
+   sph_u64 th, tl;
+
+   ptr = sc->ptr;
+   bit_len = ((unsigned)ptr << 3);
+//   z = 0x80 >> n;
+//   zz = ((ub & -z) | z) & 0xFF;
+//   buf[ptr>>3] = _mm512_set1_epi64( zz );
+   buf[ptr>>3] = m512_const1_64( 0x80 );
+   tl = sc->T0 + bit_len;
+   th = sc->T1;
+   if (ptr == 0 )
+   {
+   sc->T0 = SPH_C64(0xFFFFFFFFFFFFFC00ULL);
+   sc->T1 = SPH_C64(0xFFFFFFFFFFFFFFFFULL);
+   }
+   else if ( sc->T0 == 0 )
+   {
+   sc->T0 = SPH_C64(0xFFFFFFFFFFFFFC00ULL) + bit_len;
+   sc->T1 = SPH_T64(sc->T1 - 1);
+   }
+   else
+   {
+        sc->T0 -= 1024 - bit_len;
+   }
+   if ( ptr <= 104 )
+   {
+       memset_zero_512( buf + (ptr>>3) + 1, (104-ptr) >> 3 );
+       buf[104>>3] = _mm512_or_si512( buf[104>>3],
+                                 m512_const1_64( 0x0100000000000000ULL ) );
+       buf[112>>3] = m512_const1_64( bswap_64( th ) );
+       buf[120>>3] = m512_const1_64( bswap_64( tl ) );
+
+       blake64_8way( sc, buf + (ptr>>3), 128 - ptr );
+   }
+   else
+  {
+       memset_zero_512( buf + (ptr>>3) + 1, (120 - ptr) >> 3 );
+
+       blake64_8way( sc, buf + (ptr>>3), 128 - ptr );
+       sc->T0 = SPH_C64(0xFFFFFFFFFFFFFC00ULL);
+       sc->T1 = SPH_C64(0xFFFFFFFFFFFFFFFFULL);
+       memset_zero_512( buf, 112>>3 );
+       buf[104>>3] = m512_const1_64( 0x0100000000000000ULL );
+       buf[112>>3] = m512_const1_64( bswap_64( th ) );
+       buf[120>>3] = m512_const1_64( bswap_64( tl ) );
+
+       blake64_8way( sc, buf, 128 );
+   }
+   mm512_block_bswap_64( (__m512i*)dst, sc->H );
+}
+
+void
+blake512_8way_update(void *cc, const void *data, size_t len)
+{
+   blake64_8way(cc, data, len);
+}
+
+void
+blake512_8way_close(void *cc, void *dst)
+{
+   blake512_8way_addbits_and_close(cc, 0, 0, dst);
+}
+
+void
+blake512_8way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+   blake64_8way_close(cc, dst);
+}
+
+#endif  // AVX512
 
 // Blake-512 4 way
 
@@ -318,29 +582,6 @@ static const sph_u64 CB[16] = {
    b = mm256_ror_64( _mm256_xor_si256( b, c ), 11 ); \
 } while (0)
 
-#if SPH_COMPACT_BLAKE_64
-// not used
-#define ROUND_B_4WAY(r)   do { \
-	GB_4WAY(M[sigma[r][0x0]], M[sigma[r][0x1]], \
-		CB[sigma[r][0x0]], CB[sigma[r][0x1]], V0, V4, V8, VC); \
-	GB_4WAY(M[sigma[r][0x2]], M[sigma[r][0x3]], \
-		CB[sigma[r][0x2]], CB[sigma[r][0x3]], V1, V5, V9, VD); \
-	GB_4WAY(M[sigma[r][0x4]], M[sigma[r][0x5]], \
-		CB[sigma[r][0x4]], CB[sigma[r][0x5]], V2, V6, VA, VE); \
-	GB_4WAY(M[sigma[r][0x6]], M[sigma[r][0x7]], \
-		CB[sigma[r][0x6]], CB[sigma[r][0x7]], V3, V7, VB, VF); \
-	GB_4WAY(M[sigma[r][0x8]], M[sigma[r][0x9]], \
-		CB[sigma[r][0x8]], CB[sigma[r][0x9]], V0, V5, VA, VF); \
-	GB_4WAY(M[sigma[r][0xA]], M[sigma[r][0xB]], \
-		CB[sigma[r][0xA]], CB[sigma[r][0xB]], V1, V6, VB, VC); \
-	GB_4WAY(M[sigma[r][0xC]], M[sigma[r][0xD]], \
-		CB[sigma[r][0xC]], CB[sigma[r][0xD]], V2, V7, V8, VD); \
-	GB_4WAY(M[sigma[r][0xE]], M[sigma[r][0xF]], \
-		CB[sigma[r][0xE]], CB[sigma[r][0xF]], V3, V4, V9, VE); \
-} while (0)
-
-#else
-//current_impl
 #define ROUND_B_4WAY(r)   do { \
 	GB_4WAY(Mx(r, 0), Mx(r, 1), CBx(r, 0), CBx(r, 1), V0, V4, V8, VC); \
 	GB_4WAY(Mx(r, 2), Mx(r, 3), CBx(r, 2), CBx(r, 3), V1, V5, V9, VD); \
@@ -352,118 +593,11 @@ static const sph_u64 CB[16] = {
 	GB_4WAY(Mx(r, E), Mx(r, F), CBx(r, E), CBx(r, F), V3, V4, V9, VE); \
 	} while (0)
 
-#endif
-
-
-// Blake-512 4 way
-
 #define DECL_STATE64_4WAY \
 	__m256i H0, H1, H2, H3, H4, H5, H6, H7; \
         __m256i S0, S1, S2, S3; \
 	sph_u64 T0, T1;
 
-#define READ_STATE64_4WAY(state)   do { \
-		H0 = (state)->H[0]; \
-		H1 = (state)->H[1]; \
-		H2 = (state)->H[2]; \
-		H3 = (state)->H[3]; \
-		H4 = (state)->H[4]; \
-		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)
-
-#define WRITE_STATE64_4WAY(state)   do { \
-		(state)->H[0] = H0; \
-		(state)->H[1] = H1; \
-		(state)->H[2] = H2; \
-		(state)->H[3] = H3; \
-		(state)->H[4] = H4; \
-		(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)
-
-#if SPH_COMPACT_BLAKE_64
-
-// not used
-#define COMPRESS64_4WAY   do { \
-	__m256i M[16]; \
-	__m256i V0, V1, V2, V3, V4, V5, V6, V7; \
-	__m256i V8, V9, VA, VB, VC, VD, VE, VF; \
-	unsigned r; \
-	V0 = H0; \
-	V1 = H1; \
-	V2 = H2; \
-	V3 = H3; \
-	V4 = H4; \
-	V5 = H5; \
-	V6 = H6; \
-	V7 = H7; \
-   V8 = _mm256_xor_si256( S0, _mm256_set_epi64x( CB0, CB0, CB0, CB0 ) ); \
-   V9 = _mm256_xor_si256( S1, _mm256_set_epi64x( CB1, CB1, CB1, CB1 ) ); \
-   VA = _mm256_xor_si256( S2, _mm256_set_epi64x( CB2, CB2, CB2, CB2 ) ); \
-   VB = _mm256_xor_si256( S3, _mm256_set_epi64x( CB3, CB3, CB3, CB3 ) ); \
-   VC = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
-                          _mm256_set_epi64x( CB4, CB4, CB4, CB4 ) ); \
-   VD = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
-                          _mm256_set_epi64x( CB5, CB5, CB5, CB5 ) ); \
-   VE = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
-                          _mm256_set_epi64x( CB6, CB6, CB6, CB6 ) ); \
-   VF = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
-                          _mm256_set_epi64x( CB7, CB7, CB7, CB7 ) ); \
-	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) ); \
-	for (r = 0; r < 16; r ++) \
-		ROUND_B_4WAY(r); \
-        H0 = _mm256_xor_si256( _mm256_xor_si256( \
-                    _mm256_xor_si256( S0, V0 ), V8 ), H0 ); \
-        H1 = _mm256_xor_si256( _mm256_xor_si256( \
-                    _mm256_xor_si256( S1, V1 ), V9 ), H1 ); \
-        H2 = _mm256_xor_si256( _mm256_xor_si256( \
-                    _mm256_xor_si256( S2, V2 ), VA ), H2 ); \
-        H3 = _mm256_xor_si256( _mm256_xor_si256( \
-                    _mm256_xor_si256( S3, V3 ), VB ), H3 ); \
-        H4 = _mm256_xor_si256( _mm256_xor_si256( \
-                    _mm256_xor_si256( S0, V4 ), VC ), H4 ); \
-        H5 = _mm256_xor_si256( _mm256_xor_si256( \
-                    _mm256_xor_si256( S1, V5 ), VD ), H5 ); \
-        H6 = _mm256_xor_si256( _mm256_xor_si256( \
-                    _mm256_xor_si256( S2, V6 ), VE ), H6 ); \
-        H7 = _mm256_xor_si256( _mm256_xor_si256( \
-                    _mm256_xor_si256( S3, V7 ), VF ), H7 ); \
-	} while (0)
-
-#else
-
-//current impl
-
 #define COMPRESS64_4WAY   do \
 { \
   __m256i M0, M1, M2, M3, M4, M5, M6, M7; \
@@ -491,7 +625,7 @@ 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, \
+  shuf_bswap64 = m256_const_64( 0x18191a1b1c1d1e1f, 0x1011121314151617, \
                                 0x08090a0b0c0d0e0f, 0x0001020304050607 ); \
   M0 = _mm256_shuffle_epi8( *(buf+ 0), shuf_bswap64 ); \
   M1 = _mm256_shuffle_epi8( *(buf+ 1), shuf_bswap64 ); \
@@ -535,13 +669,8 @@ static const sph_u64 CB[16] = {
   H7 = mm256_xor4( VF, V7, S3, H7 ); \
 } while (0)
 
-#endif
 
-static const sph_u64 salt_zero_big[4] = { 0, 0, 0, 0 };
-
-static void
-blake64_4way_init( blake_4way_big_context *sc, const sph_u64 *iv,
-              const sph_u64 *salt )
+void blake512_4way_init( blake_4way_big_context *sc )
 {
    __m256i zero = m256_zero;
    casti_m256i( sc->H, 0 ) = m256_const1_64( 0x6A09E667F3BCC908 );
@@ -552,12 +681,10 @@ blake64_4way_init( blake_4way_big_context *sc, const sph_u64 *iv,
    casti_m256i( sc->H, 5 ) = m256_const1_64( 0x9B05688C2B3E6C1F );
    casti_m256i( sc->H, 6 ) = m256_const1_64( 0x1F83D9ABFB41BD6B );
    casti_m256i( sc->H, 7 ) = m256_const1_64( 0x5BE0CD19137E2179 );
-
    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;
 }
@@ -582,7 +709,7 @@ blake64_4way( blake_4way_big_context *sc, const void *data, size_t len)
 	return;
    }
 
-   READ_STATE64_4WAY(sc);
+   READ_STATE64(sc);
    while ( len > 0 )
    {
 	size_t clen;
@@ -602,25 +729,21 @@ blake64_4way( blake_4way_big_context *sc, const void *data, size_t len)
 		ptr = 0;
 	}
    }
-   WRITE_STATE64_4WAY(sc);
+   WRITE_STATE64(sc);
    sc->ptr = ptr;
 }
 
 static void
-blake64_4way_close( blake_4way_big_context *sc,
-	unsigned ub, unsigned n, void *dst, size_t out_size_w64)
+blake64_4way_close( blake_4way_big_context *sc, void *dst )
 {
    __m256i buf[16];
    size_t ptr;
    unsigned bit_len;
-   uint64_t z, zz;
    sph_u64 th, tl;
 
    ptr = sc->ptr;
    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] = m256_const1_64( 0x80 );
    tl = sc->T0 + bit_len;
    th = sc->T1;
    if (ptr == 0 )
@@ -637,43 +760,44 @@ blake64_4way_close( blake_4way_big_context *sc,
    {
         sc->T0 -= 1024 - bit_len;
    }
+
    if ( ptr <= 104 )
    {
        memset_zero_256( buf + (ptr>>3) + 1, (104-ptr) >> 3 );
-       if ( out_size_w64 == 8 )
-          buf[(104>>3)] = _mm256_or_si256( buf[(104>>3)],
+       buf[104>>3] = _mm256_or_si256( buf[104>>3],
                                  m256_const1_64( 0x0100000000000000ULL ) );
-       *(buf+(112>>3)) = _mm256_set1_epi64x( bswap_64( th ) );
-       *(buf+(120>>3)) = _mm256_set1_epi64x( bswap_64( tl ) );
+       buf[112>>3] = m256_const1_64( bswap_64( th ) );
+       buf[120>>3] = m256_const1_64( bswap_64( tl ) );
 
        blake64_4way( sc, buf + (ptr>>3), 128 - ptr );
    }
    else
-  {
+   {
        memset_zero_256( buf + (ptr>>3) + 1, (120 - ptr) >> 3 );
 
        blake64_4way( sc, buf + (ptr>>3), 128 - ptr );
        sc->T0 = SPH_C64(0xFFFFFFFFFFFFFC00ULL);
        sc->T1 = SPH_C64(0xFFFFFFFFFFFFFFFFULL);
        memset_zero_256( buf, 112>>3 ); 
-       if ( out_size_w64 == 8 )
-           buf[104>>3] = m256_const1_64( 0x0100000000000000ULL );
-       *(buf+(112>>3)) = _mm256_set1_epi64x( bswap_64( th ) );
-       *(buf+(120>>3)) = _mm256_set1_epi64x( bswap_64( tl ) );
+       buf[104>>3] = m256_const1_64( 0x0100000000000000ULL );
+       buf[112>>3] = m256_const1_64( bswap_64( th ) );
+       buf[120>>3] = m256_const1_64( bswap_64( tl ) );
 
        blake64_4way( sc, buf, 128 );
    }
    mm256_block_bswap_64( (__m256i*)dst, sc->H );
 }
 
+/*
 void
 blake512_4way_init(void *cc)
 {
 	blake64_4way_init(cc, IV512, salt_zero_big);
 }
+*/
 
 void
-blake512_4way(void *cc, const void *data, size_t len)
+blake512_4way_update(void *cc, const void *data, size_t len)
 {
 	blake64_4way(cc, data, len);
 }
@@ -681,15 +805,18 @@ blake512_4way(void *cc, const void *data, size_t len)
 void
 blake512_4way_close(void *cc, void *dst)
 {
-	blake512_4way_addbits_and_close(cc, 0, 0, dst);
+   blake64_4way_close( cc, dst );
+
+//   blake512_4way_addbits_and_close(cc, dst);
 }
 
+/*
 void
 blake512_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
 {
 	blake64_4way_close(cc, ub, n, dst, 8);
 }
-
+*/
 #ifdef __cplusplus
 }
 #endif

algo/bmw/bmw-hash-4way.h

@@ -64,7 +64,8 @@ typedef bmw_4way_small_context bmw256_4way_context;
 
 void bmw256_4way_init( bmw256_4way_context *ctx );
 
-void bmw256_4way(void *cc, const void *data, size_t len);
+void bmw256_4way_update(void *cc, const void *data, size_t len);
+#define bmw256_4way bmw256_4way_update
 
 void bmw256_4way_close(void *cc, void *dst);
 
@@ -78,7 +79,7 @@ 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
@@ -87,11 +88,33 @@ typedef struct {
 typedef bmw_8way_small_context bmw256_8way_context;
 
 void bmw256_8way_init( bmw256_8way_context *ctx );
-void bmw256_8way( bmw256_8way_context *ctx, const void *data, size_t len );
+void bmw256_8way_update( bmw256_8way_context *ctx, const void *data,
+                         size_t len );
+#define bmw256_8way bmw256_8way_update
 void bmw256_8way_close( bmw256_8way_context *ctx, void *dst );
 
 #endif
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// BMW-256 16 way 32
+
+typedef struct {
+   __m512i buf[16];
+   __m512i H[16];
+   size_t ptr;
+   uint32_t bit_count;  // assume bit_count fits in 32 bits
+} bmw_16way_small_context __attribute__ ((aligned (128)));
+
+typedef bmw_16way_small_context bmw256_16way_context;
+
+void bmw256_16way_init( bmw256_16way_context *ctx );
+void bmw256_16way_update( bmw256_16way_context *ctx, const void *data,
+                          size_t len );
+void bmw256_16way_close( bmw256_16way_context *ctx, void *dst );
+
+#endif
+
 
 #if defined(__SSE2__)
 
@@ -107,7 +130,8 @@ typedef struct {
 typedef bmw_2way_big_context bmw512_2way_context;
 
 void bmw512_2way_init( bmw512_2way_context *ctx );
-void bmw512_2way( bmw512_2way_context *ctx, const void *data, size_t len );
+void bmw512_2way_update( bmw512_2way_context *ctx, const void *data,
+                         size_t len );
 void bmw512_2way_close( bmw512_2way_context *ctx, void *dst );
 
 #endif // __SSE2__
@@ -121,14 +145,15 @@ 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;
 
 
 void bmw512_4way_init(void *cc);
 
-void bmw512_4way(void *cc, const void *data, size_t len);
+void bmw512_4way_update(void *cc, const void *data, size_t len);
+#define bmw512_4way bmw512_4way_update
 
 void bmw512_4way_close(void *cc, void *dst);
 
@@ -137,6 +162,22 @@ 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

algo/bmw/bmw512-4way.c

@@ -1,13 +1,67 @@
 #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 (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];
+   const uint32_t last_nonce = max_nonce - 8;
+   __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 ( unlikely( hash7[ lane<<1 ] < Htarg ) )
+//      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 += 8;
+
+   } while ( likely( ( n < last_nonce ) && !work_restart[thr_id].restart) );
+
+   *hashes_done = n - first_nonce;
+   return 0;
+}
+   
+#elif defined(BMW512_4WAY)
+
+//#ifdef BMW512_4WAY
+
 void bmw512hash_4way(void *state, const void *input)
 {
     bmw512_4way_context ctx;
@@ -19,16 +73,17 @@ 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 )
 {
-   uint32_t vdata[24*4] __attribute__ ((aligned (64)));
-   uint32_t hash[16*4] __attribute__ ((aligned (32)));
-   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t vdata[24*4] __attribute__ ((aligned (128)));
+   uint32_t hash[16*4] __attribute__ ((aligned (64)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
    uint32_t *hash7 = &(hash[25]);   // 3*8+1
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
    uint32_t n = pdata[19];
    const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce -  4;
    __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
-//   const uint32_t Htarg = ptarget[7];
+   const uint32_t Htarg = ptarget[7];
     int thr_id = mythr->id;  // thr_id arg is deprecated
 
    mm256_bswap32_intrlv80_4x64( vdata, pdata );
@@ -39,7 +94,8 @@ int scanhash_bmw512_4way( struct work *work, uint32_t max_nonce,
       bmw512hash_4way( hash, vdata );
 
       for ( int lane = 0; lane < 4; lane++ )
-      if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
+      if ( unlikely( hash7[ lane<<1 ] < Htarg ) )
+//      if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
       {
           extr_lane_4x64( lane_hash, hash, lane, 256 );
           if ( fulltest( lane_hash, ptarget ) )
@@ -50,9 +106,9 @@ int scanhash_bmw512_4way( struct work *work, uint32_t max_nonce,
       }
       n += 4;
 
-   } while ( (n < max_nonce-4) && !work_restart[thr_id].restart);
+   } while ( likely( (n < last_nonce) && !work_restart[thr_id].restart ) );
 
-   *hashes_done = n - first_nonce + 1;
+   *hashes_done = n - first_nonce;
    return 0;
 }
 

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

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

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

@@ -58,8 +58,7 @@ static const sph_u64 IV512[] = {
 
 #if defined(__SSE2__)
 
-// BMW-512 2 way 64
-
+// BMW-512 2 way 64 
 
 #define s2b0(x) \
    _mm_xor_si128( _mm_xor_si128( _mm_srli_epi64( (x), 1), \
@@ -556,18 +555,15 @@ 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__
 
-
-
 #if defined(__AVX2__)
 
 // BMW-512 4 way 64
 
-
 #define sb0(x) \
    mm256_xor4( _mm256_srli_epi64( (x), 1), _mm256_slli_epi64( (x), 3), \
                 mm256_rol_64(     (x), 4),  mm256_rol_64(     (x),37) )
@@ -636,165 +632,152 @@ 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] )
 {
@@ -840,87 +823,57 @@ void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
            mm256_xor4( qt[24], qt[25], qt[26], qt[27] ),
            mm256_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
 
-   dH[ 0] = _mm256_add_epi64(
-               _mm256_xor_si256( M[0],
-                  _mm256_xor_si256( _mm256_slli_epi64( xh, 5 ),
-                                    _mm256_srli_epi64( qt[16], 5 ) ) ),
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[24] ), qt[ 0] ) );
-   dH[ 1] = _mm256_add_epi64(
-               _mm256_xor_si256( M[1],
-                  _mm256_xor_si256( _mm256_srli_epi64( xh, 7 ),
-                                    _mm256_slli_epi64( qt[17], 8 ) ) ),
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[25] ), qt[ 1] ) );
-   dH[ 2] = _mm256_add_epi64(
-               _mm256_xor_si256( M[2],
-                  _mm256_xor_si256( _mm256_srli_epi64( xh, 5 ),
-                                    _mm256_slli_epi64( qt[18], 5 ) ) ),
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[26] ), qt[ 2] ) );
-   dH[ 3] = _mm256_add_epi64(
-               _mm256_xor_si256( M[3],
-                  _mm256_xor_si256( _mm256_srli_epi64( xh, 1 ),
-                                    _mm256_slli_epi64( qt[19], 5 ) ) ),
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[27] ), qt[ 3] ) );
-   dH[ 4] = _mm256_add_epi64(
-               _mm256_xor_si256( M[4],
-                  _mm256_xor_si256( _mm256_srli_epi64( xh, 3 ),
-                                    _mm256_slli_epi64( qt[20], 0 ) ) ),
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[28] ), qt[ 4] ) );
-   dH[ 5] = _mm256_add_epi64(
-               _mm256_xor_si256( M[5],
-                  _mm256_xor_si256( _mm256_slli_epi64( xh, 6 ),
-                                    _mm256_srli_epi64( qt[21], 6 ) ) ),
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[29] ), qt[ 5] ) );
-   dH[ 6] = _mm256_add_epi64(
-               _mm256_xor_si256( M[6],
-                  _mm256_xor_si256( _mm256_srli_epi64( xh, 4 ),
-                                    _mm256_slli_epi64( qt[22], 6 ) ) ),
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[30] ), qt[ 6] ) );
-   dH[ 7] = _mm256_add_epi64(
-               _mm256_xor_si256( M[7],
-                  _mm256_xor_si256( _mm256_srli_epi64( xh, 11 ),
-                                    _mm256_slli_epi64( qt[23], 2 ) ) ),
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[31] ), qt[ 7] ) );
-   dH[ 8] = _mm256_add_epi64( _mm256_add_epi64(
-              mm256_rol_64( dH[4], 9 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[24] ), M[ 8] )),
-                 _mm256_xor_si256( _mm256_slli_epi64( xl, 8 ),
-                                   _mm256_xor_si256( qt[23], qt[ 8] ) ) );
-   dH[ 9] = _mm256_add_epi64( _mm256_add_epi64(
-              mm256_rol_64( dH[5], 10 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[25] ), M[ 9] )),
-                 _mm256_xor_si256( _mm256_srli_epi64( xl, 6 ),
-                                   _mm256_xor_si256( qt[16], qt[ 9] ) ) );
-   dH[10] = _mm256_add_epi64( _mm256_add_epi64(
-              mm256_rol_64( dH[6], 11 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[26] ), M[10] )),
-                 _mm256_xor_si256( _mm256_slli_epi64( xl, 6 ),
-                                   _mm256_xor_si256( qt[17], qt[10] ) ) );
-   dH[11] = _mm256_add_epi64( _mm256_add_epi64(
-              mm256_rol_64( dH[7], 12 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[27] ), M[11] )),
-                 _mm256_xor_si256( _mm256_slli_epi64( xl, 4 ),
-                                   _mm256_xor_si256( qt[18], qt[11] ) ) );
-   dH[12] = _mm256_add_epi64( _mm256_add_epi64(
-              mm256_rol_64( dH[0], 13 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[28] ), M[12] )),
-                 _mm256_xor_si256( _mm256_srli_epi64( xl, 3 ),
-                                   _mm256_xor_si256( qt[19], qt[12] ) ) );
-   dH[13] = _mm256_add_epi64( _mm256_add_epi64(
-              mm256_rol_64( dH[1], 14 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[29] ), M[13] )),
-                 _mm256_xor_si256( _mm256_srli_epi64( xl, 4 ),
-                                   _mm256_xor_si256( qt[20], qt[13] ) ) );
-   dH[14] = _mm256_add_epi64( _mm256_add_epi64(
-              mm256_rol_64( dH[2], 15 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[30] ), M[14] )),
-                 _mm256_xor_si256( _mm256_srli_epi64( xl, 7 ),
-                                   _mm256_xor_si256( qt[21], qt[14] ) ) );
-   dH[15] = _mm256_add_epi64( _mm256_add_epi64(
-              mm256_rol_64( dH[3], 16 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[31] ), M[15] )),
-                 _mm256_xor_si256( _mm256_srli_epi64( xl, 2 ),
-                                   _mm256_xor_si256( qt[22], qt[15] ) ) );
-} 
+
+#define DH1L( m, sl, sr, a, b, c ) \
+   _mm256_add_epi64( \
+               _mm256_xor_si256( M[m], \
+                  _mm256_xor_si256( _mm256_slli_epi64( xh, sl ), \
+                                    _mm256_srli_epi64( qt[a], sr ) ) ), \
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[b] ), qt[c] ) )
+
+#define DH1R( m, sl, sr, a, b, c ) \
+   _mm256_add_epi64( \
+               _mm256_xor_si256( M[m], \
+                  _mm256_xor_si256( _mm256_srli_epi64( xh, sl ), \
+                                    _mm256_slli_epi64( qt[a], sr ) ) ), \
+               _mm256_xor_si256( _mm256_xor_si256( xl, qt[b] ), qt[c] ) )
+
+#define DH2L( m, rl, sl, h, a, b, c ) \
+   _mm256_add_epi64( _mm256_add_epi64( \
+       mm256_rol_64( dH[h], rl ), \
+          _mm256_xor_si256( _mm256_xor_si256( xh, qt[a] ), M[m] )), \
+                 _mm256_xor_si256( _mm256_slli_epi64( xl, sl ), \
+                                   _mm256_xor_si256( qt[b], qt[c] ) ) );
+
+#define DH2R( m, rl, sr, h, a, b, c ) \
+   _mm256_add_epi64( _mm256_add_epi64( \
+       mm256_rol_64( dH[h], rl ), \
+          _mm256_xor_si256( _mm256_xor_si256( xh, qt[a] ), M[m] )), \
+                 _mm256_xor_si256( _mm256_srli_epi64( xl, sr ), \
+                                   _mm256_xor_si256( qt[b], qt[c] ) ) );
+
+   dH[ 0] = DH1L(  0,  5,  5, 16, 24, 0 );
+   dH[ 1] = DH1R(  1,  7,  8, 17, 25, 1 );
+   dH[ 2] = DH1R(  2,  5,  5, 18, 26, 2 );
+   dH[ 3] = DH1R(  3,  1,  5, 19, 27, 3 );
+   dH[ 4] = DH1R(  4,  3,  0, 20, 28, 4 );
+   dH[ 5] = DH1L(  5,  6,  6, 21, 29, 5 );
+   dH[ 6] = DH1R(  6,  4,  6, 22, 30, 6 );
+   dH[ 7] = DH1R(  7, 11,  2, 23, 31, 7 );
+   dH[ 8] = DH2L(  8,  9,  8,  4, 24, 23,  8 );
+   dH[ 9] = DH2R(  9, 10,  6,  5, 25, 16,  9 );
+   dH[10] = DH2L( 10, 11,  6,  6, 26, 17, 10 );
+   dH[11] = DH2L( 11, 12,  4,  7, 27, 18, 11 );
+   dH[12] = DH2R( 12, 13,  3,  0, 28, 19, 12 );
+   dH[13] = DH2R( 13, 14,  4,  1, 29, 20, 13 );
+   dH[14] = DH2R( 14, 15,  7,  2, 30, 21, 14 );
+   dH[15] = DH2R( 15, 16,  2,  3, 31, 22, 15 );
+
+#undef DH1L
+#undef DH1R
+#undef DH2L
+#undef DH2R
+}
 
 static const __m256i final_b[16] =
 {
@@ -1060,7 +1013,7 @@ bmw512_4way_init(void *cc)
 }
 
 void
-bmw512_4way(void *cc, const void *data, size_t len)
+bmw512_4way_update(void *cc, const void *data, size_t len)
 {
 	bmw64_4way(cc, data, len);
 }
@@ -1079,6 +1032,483 @@ 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 DH1L( 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 DH1R( m, sl, sr, a, b, c ) \
+   _mm512_add_epi64( \
+               _mm512_xor_si512( M[m], \
+                  _mm512_xor_si512( _mm512_srli_epi64( xh, sl ), \
+                                    _mm512_slli_epi64( qt[a], sr ) ) ), \
+               _mm512_xor_si512( _mm512_xor_si512( xl, qt[b] ), qt[c] ) )
+
+#define DH2L( 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 DH2R( 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] = DH1L(  0,  5,  5, 16, 24, 0 );
+   dH[ 1] = DH1R(  1,  7,  8, 17, 25, 1 );
+   dH[ 2] = DH1R(  2,  5,  5, 18, 26, 2 );
+   dH[ 3] = DH1R(  3,  1,  5, 19, 27, 3 );
+   dH[ 4] = DH1R(  4,  3,  0, 20, 28, 4 );
+   dH[ 5] = DH1L(  5,  6,  6, 21, 29, 5 );
+   dH[ 6] = DH1R(  6,  4,  6, 22, 30, 6 );
+   dH[ 7] = DH1R(  7, 11,  2, 23, 31, 7 );
+   dH[ 8] = DH2L(  8,  9,  8,  4, 24, 23,  8 );
+   dH[ 9] = DH2R(  9, 10,  6,  5, 25, 16,  9 );
+   dH[10] = DH2L( 10, 11,  6,  6, 26, 17, 10 );
+   dH[11] = DH2L( 11, 12,  4,  7, 27, 18, 11 );
+   dH[12] = DH2R( 12, 13,  3,  0, 28, 19, 12 );
+   dH[13] = DH2R( 13, 14,  4,  1, 29, 20, 13 );
+   dH[14] = DH2R( 14, 15,  7,  2, 30, 21, 14 );
+   dH[15] = DH2R( 15, 16,  2,  3, 31, 22, 15 );
+
+#undef DH1L
+#undef DH1R
+#undef DH2L
+#undef DH2R
+         
+}
+
+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

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 +25,187 @@ static const uint64_t IV512[] =
 0x148FE485FCD398D9, 0xB64445321B017BEF, 0x2FF5781C6A536159, 0x0DBADEA991FA7934,
 0xA5A70E75D65C8A2B, 0xBC796576B1C62456, 0xE7989AF11921C8F7, 0xD43E3B447795D246
 };
-*/
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// 4 way 128 is handy to avoid reinterleaving in many algos.
+// If reinterleaving is necessary it may be more efficient to use
+// 2 way 256. The same transform code should work for both.
+
+static void transform_4way( cube_4way_context *sp )
+{
+    int r;
+    const int rounds = sp->rounds;
+
+    __m512i x0, x1, x2, x3, x4, x5, x6, x7, y0, y1;
+
+    x0 = _mm512_load_si512( (__m512i*)sp->h     );
+    x1 = _mm512_load_si512( (__m512i*)sp->h + 1 );
+    x2 = _mm512_load_si512( (__m512i*)sp->h + 2 );
+    x3 = _mm512_load_si512( (__m512i*)sp->h + 3 );
+    x4 = _mm512_load_si512( (__m512i*)sp->h + 4 );
+    x5 = _mm512_load_si512( (__m512i*)sp->h + 5 );
+    x6 = _mm512_load_si512( (__m512i*)sp->h + 6 );
+    x7 = _mm512_load_si512( (__m512i*)sp->h + 7 );
+
+    for ( r = 0; r < rounds; ++r )
+    {
+        x4 = _mm512_add_epi32( x0, x4 );
+        x5 = _mm512_add_epi32( x1, x5 );
+        x6 = _mm512_add_epi32( x2, x6 );
+        x7 = _mm512_add_epi32( x3, x7 );
+        y0 = x0;
+        y1 = x1;
+        x0 = mm512_rol_32( x2, 7 );
+        x1 = mm512_rol_32( x3, 7 );
+        x2 = mm512_rol_32( y0, 7 );
+        x3 = mm512_rol_32( y1, 7 );
+        x0 = _mm512_xor_si512( x0, x4 );
+        x1 = _mm512_xor_si512( x1, x5 );
+        x2 = _mm512_xor_si512( x2, x6 );
+        x3 = _mm512_xor_si512( x3, x7 );
+        x4 = mm512_swap128_64( x4 );
+        x5 = mm512_swap128_64( x5 );
+        x6 = mm512_swap128_64( x6 );
+        x7 = mm512_swap128_64( x7 );
+        x4 = _mm512_add_epi32( x0, x4 );
+        x5 = _mm512_add_epi32( x1, x5 );
+        x6 = _mm512_add_epi32( x2, x6 );
+        x7 = _mm512_add_epi32( x3, x7 );
+        y0 = x0;
+        y1 = x2;
+        x0 = mm512_rol_32( x1, 11 );
+        x1 = mm512_rol_32( y0, 11 );
+        x2 = mm512_rol_32( x3, 11 );
+        x3 = mm512_rol_32( y1, 11 );
+        x0 = _mm512_xor_si512( x0, x4 );
+        x1 = _mm512_xor_si512( x1, x5 );
+        x2 = _mm512_xor_si512( x2, x6 );
+        x3 = _mm512_xor_si512( x3, x7 );
+        x4 = mm512_swap64_32( x4 );
+        x5 = mm512_swap64_32( x5 );
+        x6 = mm512_swap64_32( x6 );
+        x7 = mm512_swap64_32( x7 );
+    }
+
+    _mm512_store_si512( (__m512i*)sp->h,     x0 );
+    _mm512_store_si512( (__m512i*)sp->h + 1, x1 );
+    _mm512_store_si512( (__m512i*)sp->h + 2, x2 );
+    _mm512_store_si512( (__m512i*)sp->h + 3, x3 );
+    _mm512_store_si512( (__m512i*)sp->h + 4, x4 );
+    _mm512_store_si512( (__m512i*)sp->h + 5, x5 );
+    _mm512_store_si512( (__m512i*)sp->h + 6, x6 );
+    _mm512_store_si512( (__m512i*)sp->h + 7, x7 );
+}
+
+int cube_4way_init( cube_4way_context *sp, int hashbitlen, int rounds,
+                    int blockbytes )
+{
+    __m512i *h = (__m512i*)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;
+
+    h[ 0] = m512_const1_128( iv[0] );
+    h[ 1] = m512_const1_128( iv[1] );
+    h[ 2] = m512_const1_128( iv[2] );
+    h[ 3] = m512_const1_128( iv[3] );
+    h[ 4] = m512_const1_128( iv[4] );
+    h[ 5] = m512_const1_128( iv[5] );
+    h[ 6] = m512_const1_128( iv[6] );
+    h[ 7] = m512_const1_128( iv[7] );
+    h[ 0] = m512_const1_128( iv[0] );
+    h[ 1] = m512_const1_128( iv[1] );
+    h[ 2] = m512_const1_128( iv[2] );
+    h[ 3] = m512_const1_128( iv[3] );
+    h[ 4] = m512_const1_128( iv[4] );
+    h[ 5] = m512_const1_128( iv[5] );
+    h[ 6] = m512_const1_128( iv[6] );
+    h[ 7] = m512_const1_128( iv[7] );
+
+    return 0;
+}
+
+int cube_4way_update( cube_4way_context *sp, const void *data, size_t size )
+{
+    const int len = size >> 4;
+    const __m512i *in = (__m512i*)data;
+    int i;
+
+    for ( i = 0; i < len; i++ )
+    {
+        sp->h[ sp->pos ] = _mm512_xor_si512( sp->h[ sp->pos ], in[i] );
+        sp->pos++;
+        if ( sp->pos == sp->blocksize )
+        {
+           transform_4way( sp );
+           sp->pos = 0;
+        }
+    }
+    return 0;
+}
+
+int cube_4way_close( cube_4way_context *sp, void *output )
+{
+    __m512i *hash = (__m512i*)output;
+    int i;
+
+    // pos is zero for 64 byte data, 1 for 80 byte data.
+    sp->h[ sp->pos ] = _mm512_xor_si512( sp->h[ sp->pos ],
+                                 m512_const2_64( 0, 0x0000000000000080 ) );
+    transform_4way( sp );
+
+    sp->h[7] = _mm512_xor_si512( sp->h[7],
+                                 m512_const2_64( 0x0000000100000000, 0 ) );
+
+    for ( i = 0; i < 10; ++i ) 
+       transform_4way( sp );
+
+    memcpy( hash, sp->h, sp->hashlen<<6 );
+    return 0;
+}
+
+int cube_4way_update_close( cube_4way_context *sp, void *output,
+                               const void *data, size_t size )
+{
+    const int len = size >> 4;
+    const __m512i *in = (__m512i*)data;
+    __m512i *hash = (__m512i*)output;
+    int i;
+
+    for ( i = 0; i < len; i++ )
+    {
+        sp->h[ sp->pos ] = _mm512_xor_si512( sp->h[ sp->pos ], in[i] );
+        sp->pos++;
+        if ( sp->pos == sp->blocksize )
+        {
+           transform_4way( sp );
+           sp->pos = 0;
+        }
+    }
+
+    // pos is zero for 64 byte data, 1 for 80 byte data.
+    sp->h[ sp->pos ] = _mm512_xor_si512( sp->h[ sp->pos ],
+                                    m512_const2_64( 0, 0x0000000000000080 ) );
+    transform_4way( sp );
+
+    sp->h[7] = _mm512_xor_si512( sp->h[7],
+                                    m512_const2_64( 0x0000000100000000, 0 ) );
+
+    for ( i = 0; i < 10; ++i )
+       transform_4way( sp );
+
+    memcpy( hash, sp->h, sp->hashlen<<6);
+    return 0;
+}
+
+
+#endif // AVX512
+
+// 2 way 128 
 
 static void transform_2way( cube_2way_context *sp )
 {
@@ -59,10 +239,10 @@ static void transform_2way( cube_2way_context *sp )
         x1 = _mm256_xor_si256( x1, x5 );
         x2 = _mm256_xor_si256( x2, x6 );
         x3 = _mm256_xor_si256( x3, x7 );
-        x4 = mm256_swap64_128( x4 );
-        x5 = mm256_swap64_128( x5 );
-        x6 = mm256_swap64_128( x6 );
-        x7 = mm256_swap64_128( x7 );
+        x4 = mm256_swap128_64( x4 );
+        x5 = mm256_swap128_64( x5 );
+        x6 = mm256_swap128_64( x6 );
+        x7 = mm256_swap128_64( x7 );
         x4 = _mm256_add_epi32( x0, x4 );
         x5 = _mm256_add_epi32( x1, x5 );
         x6 = _mm256_add_epi32( x2, x6 );
@@ -77,10 +257,10 @@ static void transform_2way( cube_2way_context *sp )
         x1 = _mm256_xor_si256( x1, x5 );
         x2 = _mm256_xor_si256( x2, x6 );
         x3 = _mm256_xor_si256( x3, x7 );
-        x4 = mm256_swap32_64( x4 );
-        x5 = mm256_swap32_64( x5 );
-        x6 = mm256_swap32_64( x6 );
-        x7 = mm256_swap32_64( x7 );
+        x4 = mm256_swap64_32( x4 );
+        x5 = mm256_swap64_32( x5 );
+        x6 = mm256_swap64_32( x6 );
+        x7 = mm256_swap64_32( x7 );
     }
 
     _mm256_store_si256( (__m256i*)sp->h,     x0 );
@@ -91,45 +271,35 @@ static void transform_2way( cube_2way_context *sp )
     _mm256_store_si256( (__m256i*)sp->h + 5, x5 );
     _mm256_store_si256( (__m256i*)sp->h + 6, x6 );
     _mm256_store_si256( (__m256i*)sp->h + 7, x7 );
-
 }
 
 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;
 }
@@ -141,9 +311,6 @@ int cube_2way_update( cube_2way_context *sp, const void *data, size_t size )
     const __m256i *in = (__m256i*)data;
     int i;
 
-    // It is assumed data is aligned to 256 bits and is a multiple of 128 bits.
-    // Current usage sata is either 64 or 80 bytes.
-
     for ( i = 0; i < len; i++ )
     {
         sp->h[ sp->pos ] = _mm256_xor_si256( sp->h[ sp->pos ], in[i] );
@@ -164,11 +331,11 @@ 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 +364,13 @@ 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;

algo/cubehash/cube-hash-2way.h

@@ -1,11 +1,38 @@
 #ifndef CUBE_HASH_2WAY_H__
-#define CUBE_HASH_2WAY_H__
-
-#if defined(__AVX2__)
+#define CUBE_HASH_2WAY_H__ 1
 
 #include <stdint.h>
 #include "simd-utils.h"
 
+#if defined(__AVX2__)
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+struct _cube_4way_context
+{
+    __m512i h[8];
+    int hashlen;
+    int rounds;
+    int blocksize;
+    int pos; 
+} __attribute__ ((aligned (128)));
+
+typedef struct _cube_4way_context cube_4way_context;
+
+int cube_4way_init( cube_4way_context* sp, int hashbitlen, int rounds,
+                       int blockbytes );
+// reinitialize context with same parameters, much faster.
+int cube_4way_reinit( cube_4way_context *sp );
+
+int cube_4way_update( cube_4way_context *sp, const void *data, size_t size );
+
+int cube_4way_close( cube_4way_context *sp, void *output );
+
+int cube_4way_update_close( cube_4way_context *sp, void *output,
+                            const void *data, size_t size );
+
+#endif
+
 // 2x128, 2 way parallel SSE2
 
 struct _cube_2way_context
@@ -15,7 +42,7 @@ struct _cube_2way_context
     int rounds;
     int blocksize;         // __m128i
     int pos;               // number of __m128i read into x from current block
-} __attribute__ ((aligned (64)));
+} __attribute__ ((aligned (128)));
 
 typedef struct _cube_2way_context cube_2way_context;
 

algo/cubehash/cubehash_sse2.c

@@ -39,8 +39,8 @@ static void transform( cubehashParam *sp )
         x1 = mm256_rol_32( y0, 7 );
         x0 = _mm256_xor_si256( x0, x2 );
         x1 = _mm256_xor_si256( x1, x3 );
-        x2 = mm256_swap64_128( x2 );
-        x3 = mm256_swap64_128( x3 );
+        x2 = mm256_swap128_64( x2 );
+        x3 = mm256_swap128_64( x3 );
         x2 = _mm256_add_epi32( x0, x2 );
         x3 = _mm256_add_epi32( x1, x3 );
         y0 = mm256_swap_128( x0 );
@@ -49,8 +49,8 @@ static void transform( cubehashParam *sp )
         x1 = mm256_rol_32( y1, 11 );
         x0 = _mm256_xor_si256( x0, x2 );
         x1 = _mm256_xor_si256( x1, x3 );
-        x2 = mm256_swap32_64( x2 );
-        x3 = mm256_swap32_64( x3 );
+        x2 = mm256_swap64_32( x2 );
+        x3 = mm256_swap64_32( x3 );
     }
 
     _mm256_store_si256( (__m256i*)sp->x,     x0 );

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

@@ -0,0 +1,559 @@
+#if defined(__AVX512VAES__) && defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#include "simd-utils.h"
+#include "echo-hash-4way.h"
+
+/*
+#include <memory.h>
+#include "miner.h"
+#include "hash_api.h"
+//#include "vperm.h"
+#include <immintrin.h>
+*/
+/*
+#ifndef NO_AES_NI
+#include <wmmintrin.h>
+#else
+#include <tmmintrin.h>
+#endif
+*/
+
+// not used
+/*
+const unsigned int _k_s0F[] = {0x0F0F0F0F, 0x0F0F0F0F, 0x0F0F0F0F, 0x0F0F0F0F};
+const unsigned int _k_ipt[] = {0x5A2A7000, 0xC2B2E898, 0x52227808, 0xCABAE090, 0x317C4D00, 0x4C01307D, 0xB0FDCC81, 0xCD80B1FC};
+const unsigned int _k_opt[] = {0xD6B66000, 0xFF9F4929, 0xDEBE6808, 0xF7974121, 0x50BCEC00, 0x01EDBD51, 0xB05C0CE0, 0xE10D5DB1};
+const unsigned int _k_inv[] = {0x0D080180, 0x0E05060F, 0x0A0B0C02, 0x04070309, 0x0F0B0780, 0x01040A06, 0x02050809, 0x030D0E0C};
+const unsigned int _k_sb1[] = {0xCB503E00, 0xB19BE18F, 0x142AF544, 0xA5DF7A6E, 0xFAE22300, 0x3618D415, 0x0D2ED9EF, 0x3BF7CCC1};
+const unsigned int _k_sb2[] = {0x0B712400, 0xE27A93C6, 0xBC982FCD, 0x5EB7E955, 0x0AE12900, 0x69EB8840, 0xAB82234A, 0xC2A163C8};
+const unsigned int _k_sb3[] = {0xC0211A00, 0x53E17249, 0xA8B2DA89, 0xFB68933B, 0xF0030A00, 0x5FF35C55, 0xA6ACFAA5, 0xF956AF09};
+const unsigned int _k_sb4[] = {0x3FD64100, 0xE1E937A0, 0x49087E9F, 0xA876DE97, 0xC393EA00, 0x3D50AED7, 0x876D2914, 0xBA44FE79};
+const unsigned int _k_sb5[] = {0xF4867F00, 0x5072D62F, 0x5D228BDB, 0x0DA9A4F9, 0x3971C900, 0x0B487AC2, 0x8A43F0FB, 0x81B332B8};
+const unsigned int _k_sb7[] = {0xFFF75B00, 0xB20845E9, 0xE1BAA416, 0x531E4DAC, 0x3390E000, 0x62A3F282, 0x21C1D3B1, 0x43125170};
+const unsigned int _k_sbo[] = {0x6FBDC700, 0xD0D26D17, 0xC502A878, 0x15AABF7A, 0x5FBB6A00, 0xCFE474A5, 0x412B35FA, 0x8E1E90D1};
+const unsigned int _k_h63[] = {0x63636363, 0x63636363, 0x63636363, 0x63636363};
+const unsigned int _k_hc6[] = {0xc6c6c6c6, 0xc6c6c6c6, 0xc6c6c6c6, 0xc6c6c6c6};
+const unsigned int _k_h5b[] = {0x5b5b5b5b, 0x5b5b5b5b, 0x5b5b5b5b, 0x5b5b5b5b};
+const unsigned int _k_h4e[] = {0x4e4e4e4e, 0x4e4e4e4e, 0x4e4e4e4e, 0x4e4e4e4e};
+const unsigned int _k_h0e[] = {0x0e0e0e0e, 0x0e0e0e0e, 0x0e0e0e0e, 0x0e0e0e0e};
+const unsigned int _k_h15[] = {0x15151515, 0x15151515, 0x15151515, 0x15151515};
+const unsigned int _k_aesmix1[] = {0x0f0a0500, 0x030e0904, 0x07020d08, 0x0b06010c};
+const unsigned int _k_aesmix2[] = {0x000f0a05, 0x04030e09, 0x0807020d, 0x0c0b0601};
+const unsigned int _k_aesmix3[] = {0x05000f0a, 0x0904030e, 0x0d080702, 0x010c0b06};
+const unsigned int _k_aesmix4[] = {0x0a05000f, 0x0e090403, 0x020d0807, 0x06010c0b};
+*/
+
+/*
+MYALIGN const unsigned int 	const1[]		= {0x00000001, 0x00000000, 0x00000000, 0x00000000};
+MYALIGN const unsigned int	mul2mask[]		= {0x00001b00, 0x00000000, 0x00000000, 0x00000000};
+MYALIGN const unsigned int	lsbmask[]		= {0x01010101, 0x01010101, 0x01010101, 0x01010101};
+MYALIGN const unsigned int	invshiftrows[]	= {0x070a0d00, 0x0b0e0104, 0x0f020508, 0x0306090c};
+MYALIGN const unsigned int	zero[]			= {0x00000000, 0x00000000, 0x00000000, 0x00000000};
+*/
+
+MYALIGN const unsigned int	mul2ipt[]		= {0x728efc00, 0x6894e61a, 0x3fc3b14d, 0x25d9ab57, 0xfd5ba600, 0x2a8c71d7, 0x1eb845e3, 0xc96f9234};
+
+// do these need to be reversed?
+
+#define mul2mask \
+   m512_const4_32( 0x00001b00, 0, 0, 0 ) 
+
+#define lsbmask    m512_const1_32( 0x01010101 ) 
+
+#define ECHO_SUBBYTES( state, i, j ) \
+	state[i][j] = _mm512_aesenc_epi128( state[i][j], k1 ); \
+	state[i][j] = _mm512_aesenc_epi128( state[i][j], m512_zero ); \
+	k1 = _mm512_add_epi32( k1, m512_one_32 )
+
+#define ECHO_MIXBYTES( state1, state2, j, t1, t2, s2 ) do \
+{ \
+   const int j1 = ( j+1 ) & 3; \
+   const int j2 = ( j+2 ) & 3; \
+   const int j3 = ( j+3 ) & 3; \
+   s2 = _mm512_add_epi8( state1[ 0 ] [j ], state1[ 0 ][ j ] ); \
+	t1 = _mm512_srli_epi16( state1[ 0 ][ j ], 7 ); \
+	t1 = _mm512_and_si128( t1, lsbmask );\
+	t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
+	s2 = _mm512_xor_si512( s2, t2 ); \
+	state2[ 0 ] [j ] = s2; \
+	state2[ 1 ] [j ] = state1[ 0 ][ j ]; \
+	state2[ 2 ] [j ] = state1[ 0 ][ j ]; \
+	state2[ 3 ] [j ] = _mm512_xor_si512( s2, state1[ 0 ][ j ] );\
+	s2 = _mm512_add_epi8( state1[ 1 ][ j1 ], state1[ 1 ][ j1 ] ); \
+	t1 = _mm512_srli_epi16( state1[ 1 ][ j1 ], 7 ); \
+	t1 = _mm512_and_si512( t1, lsbmask ); \
+	t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
+	s2 = _mm512_xor_si512( s2, t2 );\
+	state2[ 0 ][ j ] = _mm512_xor_si512( state2[ 0 ][ j ], \
+                            _mm512_xor_si512( s2, state1[ 1 ][ j1 ] ) ); \
+	state2[ 1 ][ j ] = _mm512_xor_si512( state2[ 1 ][ j ], s2 ); \
+	state2[ 2 ][ j ] = _mm512_xor_si512( state2[ 2 ][ j ], state1[ 1 ][ j1 ] ); \
+	state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3 ][ j ], state1[ 1 ][ j1 ] ); \
+	s2 = _mm512_add_epi8( state1[ 2 ][ j2 ], state1[ 2 ][ j2 ] ); \
+	t1 = _mm512_srli_epi16( state1[ 2 ][ j2 ], 7 ); \
+	t1 = _mm512_and_si512( t1, lsbmask ); \
+	t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
+	s2 = _mm512_xor_si512( s2, t2 ); \
+	state2[ 0 ][ j ] = _mm512_xor_si512( state2[ 0 ][ j ], state1[ 2 ][ j2 ] ); \
+	state2[ 1 ][ j ] = _mm512_xor_si512( state2[ 1 ][ j ], \
+                            _mm512_xor_si512( s2, state1[ 2 ][ j2 ] ) ); \
+	state2[ 2 ][ j ] = _mm512_xor_si512128( state2[ 2 ][ j ], s2 ); \
+	state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3][ j ], state1[ 2 ][ j2 ] ); \
+	s2 = _mm512_add_epi8( state1[ 3 ][ j3 ], state1[ 3 ][ j3 ] ); \
+	t1 = _mm512_srli_epi16( state1[ 3 ][ j3 ], 7 ); \
+	t1 = _mm512_and_si512( t1, lsbmask ); \
+	t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
+	s2 = _mm512_xor_si512( s2, t2 ); \
+	state2[ 0 ][ j ] = _mm512_xor_si512( state2[ 0 ][ j ], state1[ 3 ][ j3 ] ); \
+	state2[ 1 ][ j ] = _mm512_xor_si512( state2[ 1 ][ j ], state1[ 3 ][ j3 ] ); \
+	state2[ 2 ][ j ] = _mm512_xor_si512( state2[ 2 ][ j ], \
+                            _mm512_xor_si512( s2, state1[ 3 ][ j3] ) ); \
+	state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3 ][ j ], s2 )
+} while(0)
+
+#define ECHO_ROUND_UNROLL2 \
+	ECHO_SUBBYTES(_state, 0, 0);\
+	ECHO_SUBBYTES(_state, 1, 0);\
+	ECHO_SUBBYTES(_state, 2, 0);\
+	ECHO_SUBBYTES(_state, 3, 0);\
+	ECHO_SUBBYTES(_state, 0, 1);\
+	ECHO_SUBBYTES(_state, 1, 1);\
+	ECHO_SUBBYTES(_state, 2, 1);\
+	ECHO_SUBBYTES(_state, 3, 1);\
+	ECHO_SUBBYTES(_state, 0, 2);\
+	ECHO_SUBBYTES(_state, 1, 2);\
+	ECHO_SUBBYTES(_state, 2, 2);\
+	ECHO_SUBBYTES(_state, 3, 2);\
+	ECHO_SUBBYTES(_state, 0, 3);\
+	ECHO_SUBBYTES(_state, 1, 3);\
+	ECHO_SUBBYTES(_state, 2, 3);\
+	ECHO_SUBBYTES(_state, 3, 3);\
+	ECHO_MIXBYTES(_state, _state2, 0, t1, t2, s2);\
+	ECHO_MIXBYTES(_state, _state2, 1, t1, t2, s2);\
+	ECHO_MIXBYTES(_state, _state2, 2, t1, t2, s2);\
+	ECHO_MIXBYTES(_state, _state2, 3, t1, t2, s2);\
+	ECHO_SUBBYTES(_state2, 0, 0);\
+	ECHO_SUBBYTES(_state2, 1, 0);\
+	ECHO_SUBBYTES(_state2, 2, 0);\
+	ECHO_SUBBYTES(_state2, 3, 0);\
+	ECHO_SUBBYTES(_state2, 0, 1);\
+	ECHO_SUBBYTES(_state2, 1, 1);\
+	ECHO_SUBBYTES(_state2, 2, 1);\
+	ECHO_SUBBYTES(_state2, 3, 1);\
+	ECHO_SUBBYTES(_state2, 0, 2);\
+	ECHO_SUBBYTES(_state2, 1, 2);\
+	ECHO_SUBBYTES(_state2, 2, 2);\
+	ECHO_SUBBYTES(_state2, 3, 2);\
+	ECHO_SUBBYTES(_state2, 0, 3);\
+	ECHO_SUBBYTES(_state2, 1, 3);\
+	ECHO_SUBBYTES(_state2, 2, 3);\
+	ECHO_SUBBYTES(_state2, 3, 3);\
+	ECHO_MIXBYTES(_state2, _state, 0, t1, t2, s2);\
+	ECHO_MIXBYTES(_state2, _state, 1, t1, t2, s2);\
+	ECHO_MIXBYTES(_state2, _state, 2, t1, t2, s2);\
+	ECHO_MIXBYTES(_state2, _state, 3, t1, t2, s2)
+
+
+
+#define SAVESTATE(dst, src)\
+	dst[0][0] = src[0][0];\
+	dst[0][1] = src[0][1];\
+	dst[0][2] = src[0][2];\
+	dst[0][3] = src[0][3];\
+	dst[1][0] = src[1][0];\
+	dst[1][1] = src[1][1];\
+	dst[1][2] = src[1][2];\
+	dst[1][3] = src[1][3];\
+	dst[2][0] = src[2][0];\
+	dst[2][1] = src[2][1];\
+	dst[2][2] = src[2][2];\
+	dst[2][3] = src[2][3];\
+	dst[3][0] = src[3][0];\
+	dst[3][1] = src[3][1];\
+	dst[3][2] = src[3][2];\
+	dst[3][3] = src[3][3]
+
+
+void echo_4way_compress( echo_4way_context *ctx, const unsigned char *pmsg,
+               unsigned int uBlockCount )
+{
+  unsigned int r, b, i, j;
+  __m512i t1, t2, s2, k1;
+  __m512i _state[4][4], _state2[4][4], _statebackup[4][4]; 
+
+// unroll   
+  for ( i = 0; i < 4; i++ )
+  for ( j = 0; j < ctx->uHashSize / 256; j++ )
+	 _state[ i ][ j ] = ctx->state[ i ][ j ];
+
+  for ( b = 0; b < uBlockCount; b++ )
+  {
+    ctx->k = _mm512_add_epi64( ctx->k, ctx->const1536 );
+
+    // load message, make aligned, remove loadu
+    for( j = ctx->uHashSize / 256; j < 4; j++ )
+    {
+      for ( i = 0; i < 4; i++ )
+	   {
+        _state[ i ][ j ] = _mm512_loadu_si512( 
+                     (__m512i*)pmsg + 4 * (j - (ctx->uHashSize / 256)) + i );
+	   }
+	 }
+
+    // save state
+	 SAVESTATE( _statebackup, _state );
+
+	 k1 = ctx->k;
+
+	 for ( r = 0; r < ctx->uRounds / 2; r++ )
+	 {
+		ECHO_ROUND_UNROLL2;
+	 }
+		
+	 if ( ctx->uHashSize == 256 )
+	 {
+	   for ( i = 0; i < 4; i++ )
+	   {
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _state[ i ][ 1 ] );
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _state[ i ][ 2 ] );
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _state[ i ][ 3 ] );
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 0 ] );
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 1 ] );
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 2 ] ) ;
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 3 ] );
+	   }
+	 }
+	 else
+	 {
+	   for ( i = 0; i < 4; i++ )
+	   {
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _state[ i ][ 2 ] );
+		   _state[ i ][ 1 ] = _mm512_xor_si512( _state[ i ][ 1 ],
+                                              _state[ i ][ 3 ] );
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 0 ] );
+		   _state[ i ][ 0 ] = _mm512_xor_si512( _state[ i ] [0 ],
+                                              _statebackup[ i ][ 2 ] );
+		   _state[ i ][ 1 ] = _mm512_xor_si512( _state[ i ][ 1 ],
+                                              _statebackup[ i ][ 1 ] );
+		   _state[ i ][ 1 ] = _mm512_xor_si512( _state[ i ][ 1 ],
+                                              _statebackup[ i ][ 3 ] );
+      }
+	 }
+    pmsg += ctx->uBlockLength;
+  }
+  SAVESTATE(ctx->state, _state);
+
+}
+
+
+
+int echo_4way_init( echo_4way_context *ctx, int nHashSize )
+{
+	int i, j;
+
+   ctx->k = m512_zero; 
+	ctx->processed_bits = 0;
+	ctx->uBufferBytes = 0;
+
+	switch( nHashSize )
+	{
+		case 256:
+			ctx->uHashSize = 256;
+			ctx->uBlockLength = 192;
+			ctx->uRounds = 8;
+			ctx->hashsize = _mm512_const4_32( 0, 0, 0, 0x100 );
+			ctx->const1536 = _mm512_const4_32( 0, 0, 0, 0x600 );
+			break;
+
+		case 512:
+			ctx->uHashSize = 512;
+			ctx->uBlockLength = 128;
+			ctx->uRounds = 10;
+			ctx->hashsize = _mm512_const4_32( 0, 0, 0, 0x200 );
+			ctx->const1536 = _mm512_const4_32( 0, 0, 0, 0x400);
+			break;
+
+		default:
+			return BAD_HASHBITLEN;
+	}
+
+
+	for( i = 0; i < 4; i++ )
+		for( j = 0; j < nHashSize / 256; j++ )
+			ctx->state[ i ][ j ] = ctx->hashsize;
+
+	for( i = 0; i < 4; i++ )
+		for( j = nHashSize / 256; j < 4; j++ )
+			ctx->state[ i ][ j ] = m512_zero;
+
+	return SUCCESS;
+}
+
+int echo_4way_update( echo_4way_context *state, const BitSequence *data, DataLength databitlen )
+{
+	unsigned int uByteLength, uBlockCount, uRemainingBytes;
+
+	uByteLength = (unsigned int)(databitlen / 8);
+
+	if ( ( state->uBufferBytes + uByteLength ) >= state->uBlockLength )
+	{
+		if ( state->uBufferBytes != 0 )
+		{
+			// Fill the buffer
+			memcpy( state->buffer + state->uBufferBytes,
+               (void*)data, state->uBlockLength - state->uBufferBytes );
+
+			// Process buffer
+			echo_4way_compress( state, state->buffer, 1 );
+			state->processed_bits += state->uBlockLength * 8;
+
+			data += state->uBlockLength - state->uBufferBytes;
+			uByteLength -= state->uBlockLength - state->uBufferBytes;
+		}
+
+		// buffer now does not contain any unprocessed bytes
+
+		uBlockCount = uByteLength / state->uBlockLength;
+		uRemainingBytes = uByteLength % state->uBlockLength;
+
+		if ( uBlockCount > 0 )
+		{
+			echo_4way_compress( state, data, uBlockCount );
+
+			state->processed_bits += uBlockCount * state->uBlockLength * 8;
+			data += uBlockCount * state->uBlockLength;
+		}
+
+		if ( uRemainingBytes > 0 )
+		{
+			memcpy( state->buffer, (void*)data, uRemainingBytes );
+		}
+
+		state->uBufferBytes = uRemainingBytes;
+	}
+	else
+	{
+		memcpy( state->buffer + state->uBufferBytes, (void*)data, uByteLength );
+		state->uBufferBytes += uByteLength;
+	}
+
+	return 0;
+}
+
+echo_4way_close( echo_4way_context *state, BitSequence *hashval )
+{
+	__m512i remainingbits;
+
+	// Add remaining bytes in the buffer
+	state->processed_bits += state->uBufferBytes * 8;
+
+	remainingbits = _mm512_set4_epi32( 0, 0, 0, state->uBufferBytes * 8 );
+
+	// Pad with 0x80
+	state->buffer[ state->uBufferBytes++ ] = 0x80;
+	
+	// Enough buffer space for padding in this block?
+	if ( ( state->uBlockLength - state->uBufferBytes ) >= 18)
+	{
+		// Pad with zeros
+		memset( state->buffer + state->uBufferBytes, 0,
+                         state->uBlockLength - ( state->uBufferBytes + 18 ) );
+
+		// Hash size
+		*( (unsigned short*)( state->buffer + state->uBlockLength - 18 ) )
+           = state->uHashSize;
+
+		// Processed bits
+		*( ( DataLength*)( state->buffer + state->uBlockLength - 16 ) )
+           = state->processed_bits;
+		*( ( DataLength*)( state->buffer + state->uBlockLength - 8 ) ) = 0;
+
+		// Last block contains message bits?
+		if ( state->uBufferBytes == 1 )
+		{
+			state->k = _mm512_xor_si512( state->k, state->k );
+			state->k = _mm512_sub_epi64( state->k, state->const1536 );
+		}
+		else
+		{
+			state->k = _mm512_add_epi64( state->k, remainingbits );
+			state->k = _mm512_sub_epi64( state->k, state->const1536 );
+		}
+
+		// Compress
+		echo_4way_compress( state, state->buffer, 1 );
+	}
+	else
+	{
+		// Fill with zero and compress
+		memset( state->buffer + state->uBufferBytes, 0,
+                        state->uBlockLength - state->uBufferBytes );
+		state->k = _mm512_add_epi64( state->k, remainingbits );
+		state->k = _mm512_sub_epi64( state->k, state->const1536 );
+		echo_4way_compress( state, state->buffer, 1 );
+
+		// Last block
+		memset( state->buffer, 0, state->uBlockLength - 18 );
+
+		// Hash size
+		*( (unsigned short*)( state->buffer + state->uBlockLength - 18 ) )
+            = state->uHashSize;
+
+		// Processed bits
+		*( (DataLength*)( state->buffer + state->uBlockLength - 16 ) )
+            = state->processed_bits;
+		*( (DataLength*)( state->buffer + state->uBlockLength - 8 ) ) = 0;
+
+		// Compress the last block
+		state->k = _mm512_xor_si512(state->k, state->k);
+		state->k = _mm512_sub_epi64(state->k, state->const1536);
+		echo_4way_compress(state, state->buffer, 1);
+	}
+
+	// Store the hash value
+	_mm512_storeu_si512( (__m512i*)hashval + 0, state->state[ 0][ 0 ]);
+	_mm512_storeu_si512( (__m512i*)hashval + 1, state->state[ 1][ 0 ]);
+
+	if ( state->uHashSize == 512 )
+	{
+		_mm512_storeu_si512((__m512i*)hashval + 2, state->state[ 2 ][ 0 ]);
+		_mm512_storeu_si512((__m512i*)hashval + 3, state->state[ 3 ][ 0 ]);
+	}
+
+	return 0;
+}
+
+int echo_4way_update_close( echo_4way_context *state, BitSequence *hashval,
+                              const BitSequence *data, DataLength databitlen )
+{
+  unsigned int uByteLength, uBlockCount, uRemainingBytes;
+
+  uByteLength = (unsigned int)(databitlen / 8);
+
+  if ( (state->uBufferBytes + uByteLength) >= state->uBlockLength )
+  {
+     if ( state->uBufferBytes != 0 )
+     {
+        // Fill the buffer
+        memcpy( state->buffer + state->uBufferBytes,
+                   (void*)data, state->uBlockLength - state->uBufferBytes );
+
+        // Process buffer
+        echo_4way_compress( state, state->buffer, 1 );
+        state->processed_bits += state->uBlockLength * 8;
+
+        data += state->uBlockLength - state->uBufferBytes;
+        uByteLength -= state->uBlockLength - state->uBufferBytes;
+     }
+
+     // buffer now does not contain any unprocessed bytes
+
+     uBlockCount = uByteLength / state->uBlockLength;
+     uRemainingBytes = uByteLength % state->uBlockLength;
+
+     if ( uBlockCount > 0 )
+     {
+        echo_4way_compress( state, data, uBlockCount );
+        state->processed_bits += uBlockCount * state->uBlockLength * 8;
+        data += uBlockCount * state->uBlockLength;
+     }
+
+     if ( uRemainingBytes > 0 )
+     memcpy(state->buffer, (void*)data, uRemainingBytes);
+     state->uBufferBytes = uRemainingBytes;
+  }
+  else
+  {
+     memcpy( state->buffer + state->uBufferBytes, (void*)data, uByteLength );
+     state->uBufferBytes += uByteLength;
+  } 
+
+  __m512i remainingbits;
+
+  // Add remaining bytes in the buffer
+  state->processed_bits += state->uBufferBytes * 8;
+
+  remainingbits = _mm512_set4_epi32( 0, 0, 0, state->uBufferBytes * 8 );
+
+  // Pad with 0x80
+  state->buffer[ state->uBufferBytes++ ] = 0x80;
+  // Enough buffer space for padding in this block?
+  if ( (state->uBlockLength - state->uBufferBytes) >= 18 )
+   {
+     // Pad with zeros
+     memset( state->buffer + state->uBufferBytes, 0,i
+                        state->uBlockLength - (state->uBufferBytes + 18) );
+
+     // Hash size
+     *( (unsigned short*)(state->buffer + state->uBlockLength - 18) )
+                   = state->uHashSize;
+
+     // Processed bits
+     *( (DataLength*)(state->buffer + state->uBlockLength - 16) ) =
+                   state->processed_bits;
+     *( (DataLength*)(state->buffer + state->uBlockLength - 8) ) = 0;
+
+     // Last block contains message bits?
+     if( state->uBufferBytes == 1 )
+     {
+        state->k = _mm512_xor_si512( state->k, state->k );
+        state->k = _mm512_sub_epi64( state->k, state->const1536 );
+     }
+     else
+     {
+        state->k = _mm_add_epi64( state->k, remainingbits );
+        state->k = _mm_sub_epi64( state->k, state->const1536 );
+     }
+
+     // Compress
+     echo_4way_compress( state, state->buffer, 1 );
+  }
+  else
+  {
+     // Fill with zero and compress
+     memset( state->buffer + state->uBufferBytes, 0,
+                state->uBlockLength - state->uBufferBytes );
+     state->k = _mm512_add_epi64( state->k, remainingbits );
+     state->k = _mm512_sub_epi64( state->k, state->const1536 );
+     echo_4way_compress( state, state->buffer, 1 );
+
+     // Last block
+     memset( state->buffer, 0, state->uBlockLength - 18 );
+
+     // Hash size
+     *( (unsigned short*)(state->buffer + state->uBlockLength - 18) ) =
+                 state->uHashSize;
+
+     // Processed bits
+     *( (DataLength*)(state->buffer + state->uBlockLength - 16) ) =
+                  state->processed_bits;
+     *( (DataLength*)(state->buffer + state->uBlockLength - 8) ) = 0;
+     // Compress the last block
+     state->k = _mm512_xor_si512( state->k, state->k );
+     state->k = _mm512_sub_epi64( state->k, state->const1536 );
+     echo_4way_compress( state, state->buffer, 1) ;
+  }
+
+  // Store the hash value
+  _mm512_storeu_si512( (__m512i*)hashval + 0, state->state[ 0 ][ 0] );
+  _mm512_storeu_si512( (__m512i*)hashval + 1, state->state[ 1 ][ 0] );
+
+  if ( state->uHashSize == 512 )
+  {
+     _mm512_storeu_si512( (__m512i*)hashval + 2, state->state[ 2 ][ 0 ] );
+     _mm512_storeu_si512( (__m512i*)hashval + 3, state->state[ 3 ][ 0 ] );
+
+  }
+  return 0;
+}
+
+#endif

algo/echo/echo-hash-4way.h

@@ -0,0 +1,36 @@
+#if !defined(ECHO_HASH_4WAY_H__)
+#define ECHO_HASH_4WAY_H__ 1
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#include "simd-utils.h"
+
+typedef struct
+{
+   __m512i    state[4][4];
+   __m512i    buffer[ 4 * 192 / 16 ];  // 4x128 interleaved 192 bytes
+   __m512i    k;
+   __m512i    hashsize;
+   __m512i    const1536;
+
+   unsigned int   uRounds;
+   unsigned int   uHashSize;
+   unsigned int   uBlockLength;
+   unsigned int   uBufferBytes;
+   unsigned int   processed_bits;
+
+} echo_4way_context __attribute__ ((aligned (64)));
+
+int echo_4way_init( echo_4way_context *state, int hashbitlen );
+
+
+int echo_4way_update( echo_4way_context *state, const void *data,
+    unsigned int databitlen);
+
+int echo_close( echo_4way_context *state, void *hashval );
+
+int echo_4way_update_close( echo_4way_context *state, void *hashval,
+                              const void *data, int databitlen );
+
+#endif 
+#endif

algo/groestl/aes_ni/groestl-version.h

@@ -9,6 +9,7 @@
 
 //#ifndef NO_AES_NI
 
+// Not to be confused with AVX512VAES
 #define VAES
 // #define VAVX
 // #define VVPERM

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 +528,374 @@ static const sph_u32 T512[64][16] = {
 	  SPH_C32(0xe7e00a94) }
 };
 
+#define s0   m0
+#define s1   c0
+#define s2   m1
+#define s3   c1
+#define s4   c2
+#define s5   m2
+#define s6   c3
+#define s7   m3
+#define s8   m4
+#define s9   c4
+#define sA   m5
+#define sB   c5
+#define sC   c6
+#define sD   m6
+#define sE   c7
+#define sF   m7
+
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// Hamsi 8 way 
+
+#define INPUT_BIG8 \
+do { \
+  __m512i db = *buf; \
+  const uint64_t *tp = (uint64_t*)&T512[0][0];  \
+  m0 = m1 = m2 = m3 = m4 = m5 = m6 = m7 = m512_zero; \
+  for ( int u = 0; u < 64; u++ ) \
+  { \
+     __m512i dm = _mm512_and_si512( db, m512_one_64 ) ; \
+     dm = mm512_negate_32( _mm512_or_si512( dm, \
+                                          _mm512_slli_epi64( dm, 32 ) ) ); \
+     m0 = _mm512_xor_si512( m0, _mm512_and_si512( dm, \
+                                          m512_const1_64( tp[0] ) ) ); \
+     m1 = _mm512_xor_si512( m1, _mm512_and_si512( dm, \
+                                          m512_const1_64( tp[1] ) ) ); \
+     m2 = _mm512_xor_si512( m2, _mm512_and_si512( dm, \
+                                          m512_const1_64( tp[2] ) ) ); \
+     m3 = _mm512_xor_si512( m3, _mm512_and_si512( dm, \
+                                          m512_const1_64( tp[3] ) ) ); \
+     m4 = _mm512_xor_si512( m4, _mm512_and_si512( dm, \
+                                          m512_const1_64( tp[4] ) ) ); \
+     m5 = _mm512_xor_si512( m5, _mm512_and_si512( dm, \
+                                          m512_const1_64( tp[5] ) ) ); \
+     m6 = _mm512_xor_si512( m6, _mm512_and_si512( dm, \
+                                          m512_const1_64( tp[6] ) ) ); \
+     m7 = _mm512_xor_si512( m7, _mm512_and_si512( dm, \
+                                          m512_const1_64( tp[7] ) ) ); \
+     tp += 8; \
+     db = _mm512_srli_epi64( db, 1 ); \
+  } \
+} while (0)
+
+#define SBOX8( a, b, c, d ) \
+do { \
+  __m512i t; \
+  t = a; \
+  a = _mm512_and_si512( a, c ); \
+  a = _mm512_xor_si512( a, d ); \
+  c = _mm512_xor_si512( c, b ); \
+  c = _mm512_xor_si512( c, a ); \
+  d = _mm512_or_si512( d, t ); \
+  d = _mm512_xor_si512( d, b ); \
+  t = _mm512_xor_si512( t, c ); \
+  b = d; \
+  d = _mm512_or_si512( d, t ); \
+  d = _mm512_xor_si512( d, a ); \
+  a = _mm512_and_si512( a, b ); \
+  t = _mm512_xor_si512( t, a ); \
+  b = _mm512_xor_si512( b, d ); \
+  b = _mm512_xor_si512( b, t ); \
+  a = c; \
+  c = b; \
+  b = d; \
+  d = mm512_not( t ); \
+} while (0)
+
+#define L8( a, b, c, d ) \
+do { \
+   a = mm512_rol_32( a, 13 ); \
+   c = mm512_rol_32( c,  3 ); \
+   b = _mm512_xor_si512( b, _mm512_xor_si512( a, c ) ); \
+   d = _mm512_xor_si512( d, _mm512_xor_si512( c, \
+                                              _mm512_slli_epi32( a, 3 ) ) ); \
+   b = mm512_rol_32( b, 1 ); \
+   d = mm512_rol_32( d, 7 ); \
+   a = _mm512_xor_si512( a, _mm512_xor_si512( b, d ) ); \
+   c = _mm512_xor_si512( c, _mm512_xor_si512( d, \
+                                              _mm512_slli_epi32( b, 7 ) ) ); \
+   a = mm512_rol_32( a,  5 ); \
+   c = mm512_rol_32( c, 22 ); \
+} while (0)
+
+#define DECL_STATE_BIG8 \
+   __m512i c0, c1, c2, c3, c4, c5, c6, c7; \
+
+#define READ_STATE_BIG8(sc) \
+do { \
+   c0 = sc->h[0x0]; \
+   c1 = sc->h[0x1]; \
+   c2 = sc->h[0x2]; \
+   c3 = sc->h[0x3]; \
+   c4 = sc->h[0x4]; \
+   c5 = sc->h[0x5]; \
+   c6 = sc->h[0x6]; \
+   c7 = sc->h[0x7]; \
+} while (0)
+
+#define WRITE_STATE_BIG8(sc) \
+do { \
+   sc->h[0x0] = c0; \
+   sc->h[0x1] = c1; \
+   sc->h[0x2] = c2; \
+   sc->h[0x3] = c3; \
+   sc->h[0x4] = c4; \
+   sc->h[0x5] = c5; \
+   sc->h[0x6] = c6; \
+   sc->h[0x7] = c7; \
+} while (0)
+
+
+#define ROUND_BIG8(rc, alpha) \
+do { \
+   __m512i t0, t1, t2, t3; \
+   s0 = _mm512_xor_si512( s0, m512_const1_64( \
+                   ( (uint64_t)(rc) << 32 ) ^ ( (uint64_t*)(alpha) )[ 0] ) ); \
+   s1 = _mm512_xor_si512( s1, m512_const1_64( ( (uint64_t*)(alpha) )[ 1] ) ); \
+   s2 = _mm512_xor_si512( s2, m512_const1_64( ( (uint64_t*)(alpha) )[ 2] ) ); \
+   s3 = _mm512_xor_si512( s3, m512_const1_64( ( (uint64_t*)(alpha) )[ 3] ) ); \
+   s4 = _mm512_xor_si512( s4, m512_const1_64( ( (uint64_t*)(alpha) )[ 4] ) ); \
+   s5 = _mm512_xor_si512( s5, m512_const1_64( ( (uint64_t*)(alpha) )[ 5] ) ); \
+   s6 = _mm512_xor_si512( s6, m512_const1_64( ( (uint64_t*)(alpha) )[ 6] ) ); \
+   s7 = _mm512_xor_si512( s7, m512_const1_64( ( (uint64_t*)(alpha) )[ 7] ) ); \
+   s8 = _mm512_xor_si512( s8, m512_const1_64( ( (uint64_t*)(alpha) )[ 8] ) ); \
+   s9 = _mm512_xor_si512( s9, m512_const1_64( ( (uint64_t*)(alpha) )[ 9] ) ); \
+   sA = _mm512_xor_si512( sA, m512_const1_64( ( (uint64_t*)(alpha) )[10] ) ); \
+   sB = _mm512_xor_si512( sB, m512_const1_64( ( (uint64_t*)(alpha) )[11] ) ); \
+   sC = _mm512_xor_si512( sC, m512_const1_64( ( (uint64_t*)(alpha) )[12] ) ); \
+   sD = _mm512_xor_si512( sD, m512_const1_64( ( (uint64_t*)(alpha) )[13] ) ); \
+   sE = _mm512_xor_si512( sE, m512_const1_64( ( (uint64_t*)(alpha) )[14] ) ); \
+   sF = _mm512_xor_si512( sF, m512_const1_64( ( (uint64_t*)(alpha) )[15] ) ); \
+\
+  SBOX8( s0, s4, s8, sC ); \
+  SBOX8( s1, s5, s9, sD ); \
+  SBOX8( s2, s6, sA, sE ); \
+  SBOX8( s3, s7, sB, sF ); \
+\
+  t1 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( s4, 4 ), \
+                                        _mm512_bslli_epi128( s5, 4 ) ); \
+  t3 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( sD, 4 ), \
+                                        _mm512_bslli_epi128( sE, 4 ) ); \
+  L8( s0, t1, s9, t3 ); \
+  s4 = _mm512_mask_blend_epi32( 0xaaaa, s4, _mm512_bslli_epi128( t1, 4 ) ); \
+  s5 = _mm512_mask_blend_epi32( 0x5555, s5, _mm512_bsrli_epi128( t1, 4 ) ); \
+  sD = _mm512_mask_blend_epi32( 0xaaaa, sD, _mm512_bslli_epi128( t3, 4 ) ); \
+  sE = _mm512_mask_blend_epi32( 0x5555, sE, _mm512_bsrli_epi128( t3, 4 ) ); \
+\
+  t1 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( s5, 4 ), \
+                                        _mm512_bslli_epi128( s6, 4 ) ); \
+  t3 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( sE, 4 ), \
+                                        _mm512_bslli_epi128( sF, 4 ) ); \
+  L8( s1, t1, sA, t3 ); \
+  s5 = _mm512_mask_blend_epi32( 0xaaaa, s5, _mm512_bslli_epi128( t1, 4 ) ); \
+  s6 = _mm512_mask_blend_epi32( 0x5555, s6, _mm512_bsrli_epi128( t1, 4 ) ); \
+  sE = _mm512_mask_blend_epi32( 0xaaaa, sE, _mm512_bslli_epi128( t3, 4 ) ); \
+  sF = _mm512_mask_blend_epi32( 0x5555, sF, _mm512_bsrli_epi128( t3, 4 ) ); \
+\
+  t1 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( s6, 4 ), \
+                                        _mm512_bslli_epi128( s7, 4 ) ); \
+  t3 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( sF, 4 ), \
+                                        _mm512_bslli_epi128( sC, 4 ) ); \
+  L8( s2, t1, sB, t3 ); \
+  s6 = _mm512_mask_blend_epi32( 0xaaaa, s6, _mm512_bslli_epi128( t1, 4 ) ); \
+  s7 = _mm512_mask_blend_epi32( 0x5555, s7, _mm512_bsrli_epi128( t1, 4 ) ); \
+  sF = _mm512_mask_blend_epi32( 0xaaaa, sF, _mm512_bslli_epi128( t3, 4 ) ); \
+  sC = _mm512_mask_blend_epi32( 0x5555, sC, _mm512_bsrli_epi128( t3, 4 ) ); \
+\
+  t1 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( s7, 4 ), \
+                                        _mm512_bslli_epi128( s4, 4 ) ); \
+  t3 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( sC, 4 ), \
+                                        _mm512_bslli_epi128( sD, 4 ) ); \
+  L8( s3, t1, s8, t3 ); \
+  s7 = _mm512_mask_blend_epi32( 0xaaaa, s7, _mm512_bslli_epi128( t1, 4 ) ); \
+  s4 = _mm512_mask_blend_epi32( 0x5555, s4, _mm512_bsrli_epi128( t1, 4 ) ); \
+  sC = _mm512_mask_blend_epi32( 0xaaaa, sC, _mm512_bslli_epi128( t3, 4 ) ); \
+  sD = _mm512_mask_blend_epi32( 0x5555, sD, _mm512_bsrli_epi128( t3, 4 ) ); \
+\
+  t0 = _mm512_mask_blend_epi32( 0xaaaa, s0, _mm512_bslli_epi128( s8, 4 ) ); \
+  t1 = _mm512_mask_blend_epi32( 0xaaaa, s1, s9 ); \
+  t2 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( s2, 4 ), sA ); \
+  t3 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( s3, 4 ), \
+                                        _mm512_bslli_epi128( sB, 4 ) ); \
+  L8( t0, t1, t2, t3 ); \
+  s0 = _mm512_mask_blend_epi32( 0x5555, s0, t0 ); \
+  s8 = _mm512_mask_blend_epi32( 0x5555, s8, _mm512_bsrli_epi128( t0, 4 ) ); \
+  s1 = _mm512_mask_blend_epi32( 0x5555, s1, t1 ); \
+  s9 = _mm512_mask_blend_epi32( 0xaaaa, s9, t1 ); \
+  s2 = _mm512_mask_blend_epi32( 0xaaaa, s2, _mm512_bslli_epi128( t2, 4 ) ); \
+  sA = _mm512_mask_blend_epi32( 0xaaaa, sA, t2 ); \
+  s3 = _mm512_mask_blend_epi32( 0xaaaa, s3, _mm512_bslli_epi128( t3, 4 ) ); \
+  sB = _mm512_mask_blend_epi32( 0x5555, sB, _mm512_bsrli_epi128( t3, 4 ) ); \
+\
+  t0 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( s4, 4 ), sC ); \
+  t1 = _mm512_mask_blend_epi32( 0xaaaa, _mm512_bsrli_epi128( s5, 4 ), \
+                                        _mm512_bslli_epi128( sD, 4 ) ); \
+  t2 = _mm512_mask_blend_epi32( 0xaaaa, s6, _mm512_bslli_epi128( sE, 4 ) ); \
+  t3 = _mm512_mask_blend_epi32( 0xaaaa, s7, sF ); \
+  L8( t0, t1, t2, t3 ); \
+  s4 = _mm512_mask_blend_epi32( 0xaaaa, s4, _mm512_bslli_epi128( t0, 4 ) ); \
+  sC = _mm512_mask_blend_epi32( 0xaaaa, sC, t0 ); \
+  s5 = _mm512_mask_blend_epi32( 0xaaaa, s5, _mm512_bslli_epi128( t1, 4 ) ); \
+  sD = _mm512_mask_blend_epi32( 0x5555, sD, _mm512_bsrli_epi128( t1, 4 ) ); \
+  s6 = _mm512_mask_blend_epi32( 0x5555, s6, t2 ); \
+  sE = _mm512_mask_blend_epi32( 0x5555, sE, _mm512_bsrli_epi128( t2, 4 ) ); \
+  s7 = _mm512_mask_blend_epi32( 0x5555, s7, t3 ); \
+  sF = _mm512_mask_blend_epi32( 0xaaaa, sF, t3 ); \
+} while (0)
+
+#define P_BIG8 \
+do { \
+   ROUND_BIG8(0, alpha_n); \
+   ROUND_BIG8(1, alpha_n); \
+   ROUND_BIG8(2, alpha_n); \
+   ROUND_BIG8(3, alpha_n); \
+   ROUND_BIG8(4, alpha_n); \
+   ROUND_BIG8(5, alpha_n); \
+} while (0)
+
+#define PF_BIG8 \
+do { \
+   ROUND_BIG8( 0, alpha_f); \
+   ROUND_BIG8( 1, alpha_f); \
+   ROUND_BIG8( 2, alpha_f); \
+   ROUND_BIG8( 3, alpha_f); \
+   ROUND_BIG8( 4, alpha_f); \
+   ROUND_BIG8( 5, alpha_f); \
+   ROUND_BIG8( 6, alpha_f); \
+   ROUND_BIG8( 7, alpha_f); \
+   ROUND_BIG8( 8, alpha_f); \
+   ROUND_BIG8( 9, alpha_f); \
+   ROUND_BIG8(10, alpha_f); \
+   ROUND_BIG8(11, alpha_f); \
+} while (0)
+
+#define T_BIG8 \
+do { /* order is important */ \
+   c7 = sc->h[ 0x7 ] = _mm512_xor_si512( sc->h[ 0x7 ], sB ); \
+   c6 = sc->h[ 0x6 ] = _mm512_xor_si512( sc->h[ 0x6 ], sA ); \
+   c5 = sc->h[ 0x5 ] = _mm512_xor_si512( sc->h[ 0x5 ], s9 ); \
+   c4 = sc->h[ 0x4 ] = _mm512_xor_si512( sc->h[ 0x4 ], s8 ); \
+   c3 = sc->h[ 0x3 ] = _mm512_xor_si512( sc->h[ 0x3 ], s3 ); \
+   c2 = sc->h[ 0x2 ] = _mm512_xor_si512( sc->h[ 0x2 ], s2 ); \
+   c1 = sc->h[ 0x1 ] = _mm512_xor_si512( sc->h[ 0x1 ], s1 ); \
+   c0 = sc->h[ 0x0 ] = _mm512_xor_si512( sc->h[ 0x0 ], s0 ); \
+} while (0)
+
+void hamsi_8way_big( hamsi_8way_big_context *sc, __m512i *buf, size_t num )
+{
+   DECL_STATE_BIG8
+   uint32_t tmp = num << 6;
+
+   sc->count_low = SPH_T32( sc->count_low + tmp );
+   sc->count_high += (sph_u32)( (num >> 13) >> 13 );
+   if ( sc->count_low < tmp )
+      sc->count_high++;
+
+   READ_STATE_BIG8( sc );
+   while ( num-- > 0 )
+   {
+      __m512i m0, m1, m2, m3, m4, m5, m6, m7;
+
+      INPUT_BIG8;
+      P_BIG8;
+      T_BIG8;
+      buf++;
+   }
+   WRITE_STATE_BIG8( sc );
+}
+
+void hamsi_8way_big_final( hamsi_8way_big_context *sc, __m512i *buf )
+{
+   __m512i m0, m1, m2, m3, m4, m5, m6, m7;
+   DECL_STATE_BIG8
+   READ_STATE_BIG8( sc );
+   INPUT_BIG8;
+   PF_BIG8;
+   T_BIG8;
+   WRITE_STATE_BIG8( sc );
+}
+
+
+void hamsi512_8way_init( hamsi_8way_big_context *sc )
+{
+   sc->partial_len = 0;
+   sc->count_high = sc->count_low = 0;
+
+   sc->h[0] = m512_const1_64( 0x6c70617273746565 );
+   sc->h[1] = m512_const1_64( 0x656e62656b204172 );
+   sc->h[2] = m512_const1_64( 0x302c206272672031 );
+   sc->h[3] = m512_const1_64( 0x3434362c75732032 );
+   sc->h[4] = m512_const1_64( 0x3030312020422d33 );
+   sc->h[5] = m512_const1_64( 0x656e2d484c657576 );
+   sc->h[6] = m512_const1_64( 0x6c65652c65766572 );
+   sc->h[7] = m512_const1_64( 0x6769756d2042656c );
+}
+
+void hamsi512_8way_update( hamsi_8way_big_context *sc, const void *data,
+                           size_t len )
+{
+   __m512i *vdata = (__m512i*)data;
+
+   hamsi_8way_big( sc, vdata, len>>3 );
+   vdata += ( (len& ~(size_t)7) >> 3 );
+   len &= (size_t)7;
+   memcpy_512( sc->buf, vdata, len>>3 );
+   sc->partial_len = len;
+}
+
+void hamsi512_8way_close( hamsi_8way_big_context *sc, void *dst )
+{
+   __m512i pad[1];
+   int ch, cl;
+
+   sph_enc32be( &ch, sc->count_high );
+   sph_enc32be( &cl, sc->count_low + ( sc->partial_len << 3 ) );
+   pad[0] =  _mm512_set_epi32( cl, ch, cl, ch, cl, ch, cl, ch,
+                               cl, ch, cl, ch, cl, ch, cl, ch );
+//   pad[0] =  m512_const2_32( cl, ch );
+   sc->buf[0] = m512_const1_64( 0x80 );
+   hamsi_8way_big( sc, sc->buf, 1 );
+   hamsi_8way_big_final( sc, pad );
+
+   mm512_block_bswap_32( (__m512i*)dst, sc->h );
+}
+
+
+#endif // AVX512
+
+
+// Hamsi 4 way
+
 #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)
@@ -643,6 +967,7 @@ do { \
    sc->h[0x7] = c7; \
 } while (0)
 
+/*
 #define s0   m0
 #define s1   c0
 #define s2   m1
@@ -659,58 +984,28 @@ do { \
 #define sD   m6
 #define sE   c7
 #define sF   m7
+*/
 
 #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) << 32 ) ^ ( (uint64_t*)(alpha) )[ 0] ) ); \
+   s1 = _mm256_xor_si256( s1, m256_const1_64( ( (uint64_t*)(alpha) )[ 1] ) ); \
+   s2 = _mm256_xor_si256( s2, m256_const1_64( ( (uint64_t*)(alpha) )[ 2] ) ); \
+   s3 = _mm256_xor_si256( s3, m256_const1_64( ( (uint64_t*)(alpha) )[ 3] ) ); \
+   s4 = _mm256_xor_si256( s4, m256_const1_64( ( (uint64_t*)(alpha) )[ 4] ) ); \
+   s5 = _mm256_xor_si256( s5, m256_const1_64( ( (uint64_t*)(alpha) )[ 5] ) ); \
+   s6 = _mm256_xor_si256( s6, m256_const1_64( ( (uint64_t*)(alpha) )[ 6] ) ); \
+   s7 = _mm256_xor_si256( s7, m256_const1_64( ( (uint64_t*)(alpha) )[ 7] ) ); \
+   s8 = _mm256_xor_si256( s8, m256_const1_64( ( (uint64_t*)(alpha) )[ 8] ) ); \
+   s9 = _mm256_xor_si256( s9, m256_const1_64( ( (uint64_t*)(alpha) )[ 9] ) ); \
+   sA = _mm256_xor_si256( sA, m256_const1_64( ( (uint64_t*)(alpha) )[10] ) ); \
+   sB = _mm256_xor_si256( sB, m256_const1_64( ( (uint64_t*)(alpha) )[11] ) ); \
+   sC = _mm256_xor_si256( sC, m256_const1_64( ( (uint64_t*)(alpha) )[12] ) ); \
+   sD = _mm256_xor_si256( sD, m256_const1_64( ( (uint64_t*)(alpha) )[13] ) ); \
+   sE = _mm256_xor_si256( sE, m256_const1_64( ( (uint64_t*)(alpha) )[14] ) ); \
+   sF = _mm256_xor_si256( sF, m256_const1_64( ( (uint64_t*)(alpha) )[15] ) ); \
 \
   SBOX( s0, s4, s8, sC ); \
   SBOX( s1, s5, s9, sD ); \
@@ -864,47 +1159,22 @@ 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 +1190,9 @@ 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 );
 

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

@@ -60,9 +60,32 @@ typedef struct {
 typedef hamsi_4way_big_context hamsi512_4way_context;
 
 void hamsi512_4way_init( hamsi512_4way_context *sc );
-void hamsi512_4way( hamsi512_4way_context *sc, const void *data, size_t len );
+void hamsi512_4way_update( hamsi512_4way_context *sc, const void *data,
+      size_t len );
+#define hamsi512_4way hamsi512_4way_update
 void hamsi512_4way_close( hamsi512_4way_context *sc, void *dst );
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+typedef struct {
+   __m512i h[8];
+   __m512i buf[1];
+   size_t partial_len;
+   sph_u32 count_high, count_low;
+} hamsi_8way_big_context;
+
+typedef hamsi_8way_big_context hamsi512_8way_context;
+
+void hamsi512_8way_init( hamsi512_8way_context *sc );
+void hamsi512_8way_update( hamsi512_8way_context *sc, const void *data,
+                           size_t len );
+void hamsi512_8way_close( hamsi512_8way_context *sc, void *dst );
+
+
+
+#endif
+
+
 #ifdef __cplusplus
 }
 #endif

algo/haval/haval-8way-helper.c

@@ -0,0 +1,115 @@
+/* $Id: haval_helper.c 218 2010-06-08 17:06:34Z tp $ */
+/*
+ * Helper code, included (three times !) by HAVAL implementation.
+ *
+ * TODO: try to merge this with md_helper.c.
+ *
+ * ==========================(LICENSE BEGIN)============================
+ *
+ * Copyright (c) 2007-2010  Projet RNRT SAPHIR
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ===========================(LICENSE END)=============================
+ *
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#undef SPH_XCAT
+#define SPH_XCAT(a, b)    SPH_XCAT_(a, b)
+#undef SPH_XCAT_
+#define SPH_XCAT_(a, b)   a ## b
+
+static void
+SPH_XCAT(SPH_XCAT(haval, PASSES), _8way_update)
+( haval_8way_context *sc, const void *data, size_t len )
+{
+   __m256i *vdata = (__m256i*)data;
+   unsigned current;
+
+   current = (unsigned)sc->count_low & 127U;
+   while ( len > 0 )
+   {
+      unsigned clen;
+      uint32_t clow, clow2;
+
+      clen = 128U - current;
+      if ( clen > len )
+         clen = len;
+      memcpy_256( sc->buf + (current>>2), vdata, clen>>2 );
+      vdata += clen>>2;
+      current += clen;
+      len -= clen;
+      if ( current == 128U )
+      {
+         DSTATE_8W;
+         IN_PREPARE_8W(sc->buf);
+         RSTATE_8W;
+         SPH_XCAT(CORE_8W, PASSES)(INW_8W);
+         WSTATE_8W;
+         current = 0;
+      }
+      clow = sc->count_low;
+      clow2 = clow + clen;
+      sc->count_low = clow2;
+      if ( clow2 < clow )
+         sc->count_high ++;
+   }
+}
+
+static void
+SPH_XCAT(SPH_XCAT(haval, PASSES), _8way_close)( haval_8way_context *sc,
+                                                void *dst)
+{
+   unsigned current;
+   DSTATE_8W;
+
+   current = (unsigned)sc->count_low & 127UL;
+
+   sc->buf[ current>>2 ] = m256_one_32;
+   current += 4;   
+   RSTATE_8W;
+   if ( current > 116UL )
+   {
+      memset_zero_256( sc->buf + ( current>>2 ), (128UL-current) >> 2 );
+      do
+      {
+         IN_PREPARE_8W(sc->buf);
+         SPH_XCAT(CORE_8W, PASSES)(INW_8W);
+      } while (0);
+      current = 0;
+   }
+
+   uint32_t t1, t2;
+   memset_zero_256( sc->buf + ( current>>2 ), (116UL-current) >> 2 );
+   t1 = 0x01 | (PASSES << 3);
+   t2 = sc->olen << 3;
+   sc->buf[ 116>>2 ] = _mm256_set1_epi32( ( t1 << 16 ) | ( t2 << 24 ) );
+   sc->buf[ 120>>2 ] = _mm256_set1_epi32( sc->count_low << 3 );
+   sc->buf[ 124>>2 ] = _mm256_set1_epi32( (sc->count_high << 3)
+                                     | (sc->count_low >> 29) );
+   do
+   {
+      IN_PREPARE_8W(sc->buf);
+      SPH_XCAT(CORE_8W, PASSES)(INW_8W);
+   } while (0);
+   WSTATE_8W;
+   haval_8way_out( sc, dst );
+}

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

@@ -40,7 +40,7 @@
 #include <string.h>
 #include "haval-hash-4way.h"
 
-// won't compile with sse4.2
+// won't compile with sse4.2, not a problem, it's only used with AVX2 4 way.
 //#if defined (__SSE4_2__)
 #if defined(__AVX__)
 
@@ -518,6 +518,301 @@ do { \
 
 #define INMSG(i)   msg[i]
 
+#if defined(__AVX2__)
+
+// Haval-256 8 way 32 bit avx2
+
+#define F1_8W(x6, x5, x4, x3, x2, x1, x0) \
+   _mm256_xor_si256( x0, \
+       _mm256_xor_si256( _mm256_and_si256(_mm256_xor_si256( x0, x4 ), x1 ), \
+                      _mm256_xor_si256( _mm256_and_si256( x2, x5 ), \
+                                     _mm256_and_si256( x3, x6 ) ) ) ) \
+
+#define F2_8W(x6, x5, x4, x3, x2, x1, x0) \
+   _mm256_xor_si256( \
+      _mm256_and_si256( x2, \
+         _mm256_xor_si256( _mm256_andnot_si256( x3, x1 ), \
+                        _mm256_xor_si256( _mm256_and_si256( x4, x5 ), \
+                                       _mm256_xor_si256( x6, x0 ) ) ) ), \
+         _mm256_xor_si256( \
+             _mm256_and_si256( x4, _mm256_xor_si256( x1, x5 ) ), \
+             _mm256_xor_si256( _mm256_and_si256( x3, x5 ), x0 ) ) ) \
+
+#define F3_8W(x6, x5, x4, x3, x2, x1, x0) \
+  _mm256_xor_si256( \
+    _mm256_and_si256( x3, \
+      _mm256_xor_si256( _mm256_and_si256( x1, x2 ), \
+                     _mm256_xor_si256( x6, x0 ) ) ), \
+      _mm256_xor_si256( _mm256_xor_si256(_mm256_and_si256( x1, x4 ), \
+                                   _mm256_and_si256( x2, x5 ) ), x0 ) )
+
+#define F4_8W(x6, x5, x4, x3, x2, x1, x0) \
+  _mm256_xor_si256( \
+     _mm256_xor_si256( \
+        _mm256_and_si256( x3, \
+           _mm256_xor_si256( _mm256_xor_si256( _mm256_and_si256( x1, x2 ), \
+                                         _mm256_or_si256( x4, x6 ) ), x5 ) ), \
+        _mm256_and_si256( x4, \
+           _mm256_xor_si256( _mm256_xor_si256( _mm256_and_si256( mm256_not(x2), x5 ), \
+                          _mm256_xor_si256( x1, x6 ) ), x0 ) ) ), \
+     _mm256_xor_si256( _mm256_and_si256( x2, x6 ), x0 ) )
+
+
+#define F5_8W(x6, x5, x4, x3, x2, x1, x0) \
+   _mm256_xor_si256( \
+       _mm256_and_si256( x0, \
+            mm256_not( _mm256_xor_si256( \
+                    _mm256_and_si256( _mm256_and_si256( x1, x2 ), x3 ), x5 ) ) ), \
+      _mm256_xor_si256( _mm256_xor_si256( _mm256_and_si256( x1, x4 ), \
+                                    _mm256_and_si256( x2, x5 ) ), \
+                                    _mm256_and_si256( x3, x6 ) ) )
+
+#define FP3_1_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F1_8W(x1, x0, x3, x5, x6, x2, x4)
+#define FP3_2_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F2_8W(x4, x2, x1, x0, x5, x3, x6)
+#define FP3_3_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F3_8W(x6, x1, x2, x3, x4, x5, x0)
+
+#define FP4_1_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F1_8W(x2, x6, x1, x4, x5, x3, x0)
+#define FP4_2_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F2_8W(x3, x5, x2, x0, x1, x6, x4)
+#define FP4_3_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F3_8W(x1, x4, x3, x6, x0, x2, x5)
+#define FP4_4_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F4_8W(x6, x4, x0, x5, x2, x1, x3)
+
+#define FP5_1_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F1_8W(x3, x4, x1, x0, x5, x2, x6)
+#define FP5_2_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F2_8W(x6, x2, x1, x0, x3, x4, x5)
+#define FP5_3_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F3_8W(x2, x6, x0, x4, x3, x1, x5)
+#define FP5_4_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F4_8W(x1, x5, x3, x2, x0, x4, x6)
+#define FP5_5_8W(x6, x5, x4, x3, x2, x1, x0) \
+   F5_8W(x2, x5, x0, x6, x4, x3, x1)
+
+#define STEP_8W(n, p, x7, x6, x5, x4, x3, x2, x1, x0, w, c) \
+do { \
+   __m256i t = FP ## n ## _ ## p ## _8W(x6, x5, x4, x3, x2, x1, x0); \
+   x7 = _mm256_add_epi32( _mm256_add_epi32( mm256_ror_32( t, 7 ), \
+                                      mm256_ror_32( x7, 11 ) ), \
+                       _mm256_add_epi32( w, _mm256_set1_epi32( c ) ) ); \
+} while (0)
+
+#define PASS1_8W(n, in)   do { \
+      unsigned pass_count; \
+      for (pass_count = 0; pass_count < 32; pass_count += 8) { \
+         STEP_8W(n, 1, s7, s6, s5, s4, s3, s2, s1, s0, \
+            in(pass_count + 0), SPH_C32(0x00000000)); \
+         STEP_8W(n, 1, s6, s5, s4, s3, s2, s1, s0, s7, \
+            in(pass_count + 1), SPH_C32(0x00000000)); \
+         STEP_8W(n, 1, s5, s4, s3, s2, s1, s0, s7, s6, \
+            in(pass_count + 2), SPH_C32(0x00000000)); \
+         STEP_8W(n, 1, s4, s3, s2, s1, s0, s7, s6, s5, \
+            in(pass_count + 3), SPH_C32(0x00000000)); \
+         STEP_8W(n, 1, s3, s2, s1, s0, s7, s6, s5, s4, \
+            in(pass_count + 4), SPH_C32(0x00000000)); \
+         STEP_8W(n, 1, s2, s1, s0, s7, s6, s5, s4, s3, \
+            in(pass_count + 5), SPH_C32(0x00000000)); \
+         STEP_8W(n, 1, s1, s0, s7, s6, s5, s4, s3, s2, \
+            in(pass_count + 6), SPH_C32(0x00000000)); \
+         STEP_8W(n, 1, s0, s7, s6, s5, s4, s3, s2, s1, \
+            in(pass_count + 7), SPH_C32(0x00000000)); \
+         } \
+   } while (0)
+
+#define PASSG_8W(p, n, in)   do { \
+      unsigned pass_count; \
+      for (pass_count = 0; pass_count < 32; pass_count += 8) { \
+         STEP_8W(n, p, s7, s6, s5, s4, s3, s2, s1, s0, \
+            in(MP ## p[pass_count + 0]), \
+            RK ## p[pass_count + 0]); \
+         STEP_8W(n, p, s6, s5, s4, s3, s2, s1, s0, s7, \
+            in(MP ## p[pass_count + 1]), \
+            RK ## p[pass_count + 1]); \
+         STEP_8W(n, p, s5, s4, s3, s2, s1, s0, s7, s6, \
+            in(MP ## p[pass_count + 2]), \
+            RK ## p[pass_count + 2]); \
+         STEP_8W(n, p, s4, s3, s2, s1, s0, s7, s6, s5, \
+            in(MP ## p[pass_count + 3]), \
+            RK ## p[pass_count + 3]); \
+         STEP_8W(n, p, s3, s2, s1, s0, s7, s6, s5, s4, \
+            in(MP ## p[pass_count + 4]), \
+            RK ## p[pass_count + 4]); \
+         STEP_8W(n, p, s2, s1, s0, s7, s6, s5, s4, s3, \
+            in(MP ## p[pass_count + 5]), \
+            RK ## p[pass_count + 5]); \
+         STEP_8W(n, p, s1, s0, s7, s6, s5, s4, s3, s2, \
+            in(MP ## p[pass_count + 6]), \
+            RK ## p[pass_count + 6]); \
+         STEP_8W(n, p, s0, s7, s6, s5, s4, s3, s2, s1, \
+            in(MP ## p[pass_count + 7]), \
+            RK ## p[pass_count + 7]); \
+         } \
+   } while (0)
+
+#define PASS2_8W(n, in)    PASSG_8W(2, n, in)
+#define PASS3_8W(n, in)    PASSG_8W(3, n, in)
+#define PASS4_8W(n, in)    PASSG_8W(4, n, in)
+#define PASS5_8W(n, in)    PASSG_8W(5, n, in)
+
+#define SAVE_STATE_8W \
+   __m256i u0, u1, u2, u3, u4, u5, u6, u7; \
+   do { \
+      u0 = s0; \
+      u1 = s1; \
+      u2 = s2; \
+      u3 = s3; \
+      u4 = s4; \
+      u5 = s5; \
+      u6 = s6; \
+      u7 = s7; \
+   } while (0)
+
+#define UPDATE_STATE_8W \
+do { \
+   s0 = _mm256_add_epi32( s0, u0 ); \
+   s1 = _mm256_add_epi32( s1, u1 ); \
+   s2 = _mm256_add_epi32( s2, u2 ); \
+   s3 = _mm256_add_epi32( s3, u3 ); \
+   s4 = _mm256_add_epi32( s4, u4 ); \
+   s5 = _mm256_add_epi32( s5, u5 ); \
+   s6 = _mm256_add_epi32( s6, u6 ); \
+   s7 = _mm256_add_epi32( s7, u7 ); \
+} while (0)
+
+#define CORE_8W5(in)  do { \
+      SAVE_STATE_8W; \
+      PASS1_8W(5, in); \
+      PASS2_8W(5, in); \
+      PASS3_8W(5, in); \
+      PASS4_8W(5, in); \
+      PASS5_8W(5, in); \
+      UPDATE_STATE_8W; \
+   } while (0)
+
+#define DSTATE_8W   __m256i s0, s1, s2, s3, s4, s5, s6, s7
+
+#define RSTATE_8W \
+do { \
+   s0 = sc->s0; \
+   s1 = sc->s1; \
+   s2 = sc->s2; \
+   s3 = sc->s3; \
+   s4 = sc->s4; \
+   s5 = sc->s5; \
+   s6 = sc->s6; \
+   s7 = sc->s7; \
+} while (0)
+
+#define WSTATE_8W \
+do { \
+   sc->s0 = s0; \
+   sc->s1 = s1; \
+   sc->s2 = s2; \
+   sc->s3 = s3; \
+   sc->s4 = s4; \
+   sc->s5 = s5; \
+   sc->s6 = s6; \
+   sc->s7 = s7; \
+} while (0)
+
+static void
+haval_8way_init( haval_8way_context *sc, unsigned olen, unsigned passes )
+{
+   sc->s0 = m256_const1_32( 0x243F6A88UL );
+   sc->s1 = m256_const1_32( 0x85A308D3UL );
+   sc->s2 = m256_const1_32( 0x13198A2EUL );
+   sc->s3 = m256_const1_32( 0x03707344UL );
+   sc->s4 = m256_const1_32( 0xA4093822UL );
+   sc->s5 = m256_const1_32( 0x299F31D0UL );
+   sc->s6 = m256_const1_32( 0x082EFA98UL );
+   sc->s7 = m256_const1_32( 0xEC4E6C89UL );
+   sc->olen = olen;
+   sc->passes = passes;
+   sc->count_high = 0;
+   sc->count_low = 0;
+
+}
+#define IN_PREPARE_8W(indata) const __m256i *const load_ptr_8w = (indata)
+
+#define INW_8W(i)   load_ptr_8w[ i ] 
+
+static void
+haval_8way_out( haval_8way_context *sc, void *dst )
+{
+   __m256i *buf = (__m256i*)dst;
+   DSTATE_8W;
+   RSTATE_8W;
+
+   buf[0] = s0;
+   buf[1] = s1;
+   buf[2] = s2;
+   buf[3] = s3;
+   buf[4] = s4;
+   buf[5] = s5;
+   buf[6] = s6;
+   buf[7] = s7;
+}
+
+#undef PASSES
+#define PASSES   5
+#include "haval-8way-helper.c"
+
+#define API_8W(xxx, y) \
+void \
+haval ## xxx ## _ ## y ## _8way_init(void *cc) \
+{ \
+   haval_8way_init(cc, xxx >> 5, y); \
+} \
+ \
+void \
+haval ## xxx ## _ ## y ## _8way_update (void *cc, const void *data, size_t len) \
+{ \
+   haval ## y ## _8way_update(cc, data, len); \
+} \
+ \
+void \
+haval ## xxx ## _ ## y ## _8way_close(void *cc, void *dst) \
+{ \
+   haval ## y ## _8way_close(cc, dst); \
+} \
+
+API_8W(256, 5)
+
+#define RVAL_8W \
+do { \
+   s0 = val[0]; \
+   s1 = val[1]; \
+   s2 = val[2]; \
+   s3 = val[3]; \
+   s4 = val[4]; \
+   s5 = val[5]; \
+   s6 = val[6]; \
+   s7 = val[7]; \
+} while (0)
+
+#define WVAL_8W \
+do { \
+   val[0] = s0; \
+   val[1] = s1; \
+   val[2] = s2; \
+   val[3] = s3; \
+   val[4] = s4; \
+   val[5] = s5; \
+   val[6] = s6; \
+   val[7] = s7; \
+} while (0)
+
+#define INMSG_8W(i)   msg[i]
+
+
+
+#endif // AVX2
+
 #ifdef __cplusplus
 }
 #endif	

algo/haval/haval-hash-4way.h

@@ -59,7 +59,7 @@
  */
 
 #ifndef HAVAL_HASH_4WAY_H__
-#define HAVAL_HASH_4WAY_H__
+#define HAVAL_HASH_4WAY_H__ 1
 
 #if defined(__AVX__)
 
@@ -84,10 +84,30 @@ typedef haval_4way_context haval256_5_4way_context;
 
 void haval256_5_4way_init( void *cc );
 
-void haval256_5_4way( void *cc, const void *data, size_t len );
+void haval256_5_4way_update( void *cc, const void *data, size_t len );
+#define haval256_5_4way haval256_5_4way_update
 
 void haval256_5_4way_close( void *cc, void *dst );
 
+#if defined(__AVX2__)
+
+typedef struct {
+   __m256i buf[32];
+   __m256i s0, s1, s2, s3, s4, s5, s6, s7;
+   unsigned olen, passes;
+   uint32_t count_high, count_low;
+} haval_8way_context __attribute__ ((aligned (64)));
+
+typedef haval_8way_context haval256_5_8way_context;
+
+void haval256_5_8way_init( void *cc );
+
+void haval256_5_8way_update( void *cc, const void *data, size_t len );
+
+void haval256_5_8way_close( void *cc, void *dst );
+
+#endif // AVX2
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -92,9 +92,41 @@ extern "C"{
 
 #endif
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define Sb_8W(x0, x1, x2, x3, c) \
+do { \
+   __m512i cc = _mm512_set1_epi64( c ); \
+    x3 = mm512_not( x3 ); \
+    x0 = _mm512_xor_si512( x0, _mm512_andnot_si512( x2, cc ) ); \
+    tmp = _mm512_xor_si512( cc, _mm512_and_si512( x0, x1 ) ); \
+    x0 = _mm512_xor_si512( x0, _mm512_and_si512( x2, x3 ) ); \
+    x3 = _mm512_xor_si512( x3, _mm512_andnot_si512( x1, x2 ) ); \
+    x1 = _mm512_xor_si512( x1, _mm512_and_si512( x0, x2 ) ); \
+    x2 = _mm512_xor_si512( x2, _mm512_andnot_si512( x3, x0 ) ); \
+    x0 = _mm512_xor_si512( x0, _mm512_or_si512( x1, x3 ) ); \
+    x3 = _mm512_xor_si512( x3, _mm512_and_si512( x1, x2 ) ); \
+    x1 = _mm512_xor_si512( x1, _mm512_and_si512( tmp, x0 ) ); \
+    x2 = _mm512_xor_si512( x2, tmp ); \
+} while (0)
+
+#define Lb_8W(x0, x1, x2, x3, x4, x5, x6, x7) \
+do { \
+    x4 = _mm512_xor_si512( x4, x1 ); \
+    x5 = _mm512_xor_si512( x5, x2 ); \
+    x6 = _mm512_xor_si512( x6, _mm512_xor_si512( x3, x0 ) ); \
+    x7 = _mm512_xor_si512( x7, x0 ); \
+    x0 = _mm512_xor_si512( x0, x5 ); \
+    x1 = _mm512_xor_si512( x1, x6 ); \
+    x2 = _mm512_xor_si512( x2, _mm512_xor_si512( x7, x4 ) ); \
+    x3 = _mm512_xor_si512( x3, x4 ); \
+} while (0)
+
+#endif
+
 #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 ) ); \
@@ -226,6 +258,48 @@ static const sph_u64 C[] = {
 			x4 ## l, x5 ## l, x6 ## l, x7 ## l); \
 	} while (0)
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define S_8W(x0, x1, x2, x3, cb, r)   do { \
+      Sb_8W(x0 ## h, x1 ## h, x2 ## h, x3 ## h, cb ## hi(r)); \
+      Sb_8W(x0 ## l, x1 ## l, x2 ## l, x3 ## l, cb ## lo(r)); \
+   } while (0)
+
+#define L_8W(x0, x1, x2, x3, x4, x5, x6, x7)   do { \
+      Lb_8W(x0 ## h, x1 ## h, x2 ## h, x3 ## h, \
+         x4 ## h, x5 ## h, x6 ## h, x7 ## h); \
+      Lb_8W(x0 ## l, x1 ## l, x2 ## l, x3 ## l, \
+         x4 ## l, x5 ## l, x6 ## l, x7 ## l); \
+   } while (0)
+
+#define Wz_8W(x, c, n) \
+do { \
+   __m512i t = _mm512_slli_epi64( _mm512_and_si512(x ## h, (c)), (n) ); \
+   x ## h = _mm512_or_si512( _mm512_and_si512( \
+                                _mm512_srli_epi64(x ## h, (n)), (c)), t ); \
+   t = _mm512_slli_epi64( _mm512_and_si512(x ## l, (c)), (n) ); \
+   x ## l = _mm512_or_si512( _mm512_and_si512((x ## l >> (n)), (c)), t ); \
+} while (0)
+
+#define W80(x)   Wz_8W(x, m512_const1_64( 0x5555555555555555 ),  1 )
+#define W81(x)   Wz_8W(x, m512_const1_64( 0x3333333333333333 ),  2 )
+#define W82(x)   Wz_8W(x, m512_const1_64( 0x0F0F0F0F0F0F0F0F ),  4 )
+#define W83(x)   Wz_8W(x, m512_const1_64( 0x00FF00FF00FF00FF ),  8 ) 
+#define W84(x)   Wz_8W(x, m512_const1_64( 0x0000FFFF0000FFFF ), 16 )
+#define W85(x)   Wz_8W(x, m512_const1_64( 0x00000000FFFFFFFF ), 32 )
+#define W86(x) \
+do { \
+   __m512i t = x ## h; \
+   x ## h = x ## l; \
+   x ## l = t; \
+} while (0)
+
+#define DECL_STATE_8W \
+   __m512i h0h, h1h, h2h, h3h, h4h, h5h, h6h, h7h; \
+   __m512i h0l, h1l, h2l, h3l, h4l, h5l, h6l, h7l; \
+   __m512i tmp;
+
+#endif
 
 #define Wz(x, c, n) \
 do { \
@@ -236,16 +310,6 @@ do { \
    x ## l = _mm256_or_si256( _mm256_and_si256((x ## l >> (n)), (c)), t ); \
 } while (0)
 
-
-/*
-#define Wz(x, c, n)   do { \
-		sph_u64 t = (x ## h & (c)) << (n); \
-		x ## h = ((x ## h >> (n)) & (c)) | t; \
-		t = (x ## l & (c)) << (n); \
-		x ## l = ((x ## l >> (n)) & (c)) | t; \
-	} while (0)
-*/
-
 #define W0(x)   Wz(x, m256_const1_64( 0x5555555555555555 ),  1 )
 #define W1(x)   Wz(x, m256_const1_64( 0x3333333333333333 ),  2 )
 #define W2(x)   Wz(x, m256_const1_64( 0x0F0F0F0F0F0F0F0F ),  4 )
@@ -259,25 +323,12 @@ do { \
    x ## l = t; \
 } while (0)
 
-/*
-#define W0(x)   Wz(x, SPH_C64(0x5555555555555555),  1)
-#define W1(x)   Wz(x, SPH_C64(0x3333333333333333),  2)
-#define W2(x)   Wz(x, SPH_C64(0x0F0F0F0F0F0F0F0F),  4)
-#define W3(x)   Wz(x, SPH_C64(0x00FF00FF00FF00FF),  8)
-#define W4(x)   Wz(x, SPH_C64(0x0000FFFF0000FFFF), 16)
-#define W5(x)   Wz(x, SPH_C64(0x00000000FFFFFFFF), 32)
-#define W6(x)   do { \
-		sph_u64 t = x ## h; \
-		x ## h = x ## l; \
-		x ## l = t; \
-	} while (0)
-*/
-
 #define DECL_STATE \
 	__m256i h0h, h1h, h2h, h3h, h4h, h5h, h6h, h7h; \
 	__m256i h0l, h1l, h2l, h3l, h4l, h5l, h6l, h7l; \
 	__m256i tmp;
 
+
 #define READ_STATE(state)   do { \
 		h0h = (state)->H[ 0]; \
 		h0l = (state)->H[ 1]; \
@@ -316,6 +367,38 @@ do { \
 		(state)->H[15] = h7l; \
 	} while (0)
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define INPUT_BUF1_8W \
+   __m512i m0h = buf[0]; \
+   __m512i m0l = buf[1]; \
+   __m512i m1h = buf[2]; \
+   __m512i m1l = buf[3]; \
+   __m512i m2h = buf[4]; \
+   __m512i m2l = buf[5]; \
+   __m512i m3h = buf[6]; \
+   __m512i m3l = buf[7]; \
+   h0h = _mm512_xor_si512( h0h, m0h ); \
+   h0l = _mm512_xor_si512( h0l, m0l ); \
+   h1h = _mm512_xor_si512( h1h, m1h ); \
+   h1l = _mm512_xor_si512( h1l, m1l ); \
+   h2h = _mm512_xor_si512( h2h, m2h ); \
+   h2l = _mm512_xor_si512( h2l, m2l ); \
+   h3h = _mm512_xor_si512( h3h, m3h ); \
+   h3l = _mm512_xor_si512( h3l, m3l ); \
+
+#define INPUT_BUF2_8W \
+   h4h = _mm512_xor_si512( h4h, m0h ); \
+   h4l = _mm512_xor_si512( h4l, m0l ); \
+   h5h = _mm512_xor_si512( h5h, m1h ); \
+   h5l = _mm512_xor_si512( h5l, m1l ); \
+   h6h = _mm512_xor_si512( h6h, m2h ); \
+   h6l = _mm512_xor_si512( h6l, m2l ); \
+   h7h = _mm512_xor_si512( h7h, m3h ); \
+   h7l = _mm512_xor_si512( h7l, m3l ); \
+
+#endif
+
 #define INPUT_BUF1 \
 	__m256i m0h = buf[0]; \
 	__m256i m0l = buf[1]; \
@@ -344,6 +427,7 @@ do { \
    h7h = _mm256_xor_si256( h7h, m3h ); \
    h7l = _mm256_xor_si256( h7l, m3l ); \
 
+
 static const sph_u64 IV256[] = {
 	C64e(0xeb98a3412c20d3eb), C64e(0x92cdbe7b9cb245c1),
 	C64e(0x1c93519160d4c7fa), C64e(0x260082d67e508a03),
@@ -370,6 +454,22 @@ static const sph_u64 IV512[] = {
 #else
 
 
+#endif
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define SL_8W(ro)   SLu_8W(r + ro, ro)
+
+#define SLu_8W(r, ro)   do { \
+      S_8W(h0, h2, h4, h6, Ceven_, r); \
+      S_8W(h1, h3, h5, h7, Codd_, r); \
+      L_8W(h0, h2, h4, h6, h1, h3, h5, h7); \
+      W8 ## ro(h1); \
+      W8 ## ro(h3); \
+      W8 ## ro(h5); \
+      W8 ## ro(h7); \
+   } while (0)
+
 #endif
 
 #define SL(ro)   SLu(r + ro, ro)
@@ -393,6 +493,23 @@ static const sph_u64 IV512[] = {
  * loop.
  */
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define E8_8W   do { \
+      unsigned r; \
+      for (r = 0; r < 42; r += 7) { \
+         SL_8W(0); \
+         SL_8W(1); \
+         SL_8W(2); \
+         SL_8W(3); \
+         SL_8W(4); \
+         SL_8W(5); \
+         SL_8W(6); \
+      } \
+   } while (0)
+
+#endif
+
 #define E8   do { \
 		unsigned r; \
 		for (r = 0; r < 42; r += 7) { \
@@ -419,51 +536,100 @@ static const sph_u64 IV512[] = {
  * On a "true 64-bit" architecture, we can unroll at will.
  */
 
-#define E8   do { \
-		SLu( 0, 0); \
-		SLu( 1, 1); \
-		SLu( 2, 2); \
-		SLu( 3, 3); \
-		SLu( 4, 4); \
-		SLu( 5, 5); \
-		SLu( 6, 6); \
-		SLu( 7, 0); \
-		SLu( 8, 1); \
-		SLu( 9, 2); \
-		SLu(10, 3); \
-		SLu(11, 4); \
-		SLu(12, 5); \
-		SLu(13, 6); \
-		SLu(14, 0); \
-		SLu(15, 1); \
-		SLu(16, 2); \
-		SLu(17, 3); \
-		SLu(18, 4); \
-		SLu(19, 5); \
-		SLu(20, 6); \
-		SLu(21, 0); \
-		SLu(22, 1); \
-		SLu(23, 2); \
-		SLu(24, 3); \
-		SLu(25, 4); \
-		SLu(26, 5); \
-		SLu(27, 6); \
-		SLu(28, 0); \
-		SLu(29, 1); \
-		SLu(30, 2); \
-		SLu(31, 3); \
-		SLu(32, 4); \
-		SLu(33, 5); \
-		SLu(34, 6); \
-		SLu(35, 0); \
-		SLu(36, 1); \
-		SLu(37, 2); \
-		SLu(38, 3); \
-		SLu(39, 4); \
-		SLu(40, 5); \
-		SLu(41, 6); \
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define E8_8W   do { \
+		SLu_8W( 0, 0); \
+		SLu_8W( 1, 1); \
+		SLu_8W( 2, 2); \
+		SLu_8W( 3, 3); \
+		SLu_8W( 4, 4); \
+		SLu_8W( 5, 5); \
+		SLu_8W( 6, 6); \
+		SLu_8W( 7, 0); \
+		SLu_8W( 8, 1); \
+		SLu_8W( 9, 2); \
+		SLu_8W(10, 3); \
+		SLu_8W(11, 4); \
+		SLu_8W(12, 5); \
+		SLu_8W(13, 6); \
+		SLu_8W(14, 0); \
+		SLu_8W(15, 1); \
+		SLu_8W(16, 2); \
+		SLu_8W(17, 3); \
+		SLu_8W(18, 4); \
+		SLu_8W(19, 5); \
+		SLu_8W(20, 6); \
+		SLu_8W(21, 0); \
+		SLu_8W(22, 1); \
+		SLu_8W(23, 2); \
+		SLu_8W(24, 3); \
+		SLu_8W(25, 4); \
+		SLu_8W(26, 5); \
+		SLu_8W(27, 6); \
+		SLu_8W(28, 0); \
+		SLu_8W(29, 1); \
+		SLu_8W(30, 2); \
+		SLu_8W(31, 3); \
+		SLu_8W(32, 4); \
+		SLu_8W(33, 5); \
+		SLu_8W(34, 6); \
+		SLu_8W(35, 0); \
+		SLu_8W(36, 1); \
+		SLu_8W(37, 2); \
+		SLu_8W(38, 3); \
+		SLu_8W(39, 4); \
+		SLu_8W(40, 5); \
+		SLu_8W(41, 6); \
 	} while (0)
 
+#endif  // AVX512
+
+#define E8   do { \
+      SLu( 0, 0); \
+      SLu( 1, 1); \
+      SLu( 2, 2); \
+      SLu( 3, 3); \
+      SLu( 4, 4); \
+      SLu( 5, 5); \
+      SLu( 6, 6); \
+      SLu( 7, 0); \
+      SLu( 8, 1); \
+      SLu( 9, 2); \
+      SLu(10, 3); \
+      SLu(11, 4); \
+      SLu(12, 5); \
+      SLu(13, 6); \
+      SLu(14, 0); \
+      SLu(15, 1); \
+      SLu(16, 2); \
+      SLu(17, 3); \
+      SLu(18, 4); \
+      SLu(19, 5); \
+      SLu(20, 6); \
+      SLu(21, 0); \
+      SLu(22, 1); \
+      SLu(23, 2); \
+      SLu(24, 3); \
+      SLu(25, 4); \
+      SLu(26, 5); \
+      SLu(27, 6); \
+      SLu(28, 0); \
+      SLu(29, 1); \
+      SLu(30, 2); \
+      SLu(31, 3); \
+      SLu(32, 4); \
+      SLu(33, 5); \
+      SLu(34, 6); \
+      SLu(35, 0); \
+      SLu(36, 1); \
+      SLu(37, 2); \
+      SLu(38, 3); \
+      SLu(39, 4); \
+      SLu(40, 5); \
+      SLu(41, 6); \
+   } while (0)
+
 #else
 
 
@@ -471,6 +637,158 @@ static const sph_u64 IV512[] = {
 
 #endif
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+void jh256_8way_init( jh_8way_context *sc )
+{
+    // bswapped IV256
+    sc->H[ 0] = m512_const1_64( 0xebd3202c41a398eb );
+    sc->H[ 1] = m512_const1_64( 0xc145b29c7bbecd92 );
+    sc->H[ 2] = m512_const1_64( 0xfac7d4609151931c );
+    sc->H[ 3] = m512_const1_64( 0x038a507ed6820026 );
+    sc->H[ 4] = m512_const1_64( 0x45b92677269e23a4 );
+    sc->H[ 5] = m512_const1_64( 0x77941ad4481afbe0 );
+    sc->H[ 6] = m512_const1_64( 0x7a176b0226abb5cd );
+    sc->H[ 7] = m512_const1_64( 0xa82fff0f4224f056 );
+    sc->H[ 8] = m512_const1_64( 0x754d2e7f8996a371 );
+    sc->H[ 9] = m512_const1_64( 0x62e27df70849141d );
+    sc->H[10] = m512_const1_64( 0x948f2476f7957627 );
+    sc->H[11] = m512_const1_64( 0x6c29804757b6d587 );
+    sc->H[12] = m512_const1_64( 0x6c0d8eac2d275e5c );
+    sc->H[13] = m512_const1_64( 0x0f7a0557c6508451 );
+    sc->H[14] = m512_const1_64( 0xea12247067d3e47b );
+    sc->H[15] = m512_const1_64( 0x69d71cd313abe389 );
+    sc->ptr = 0;
+    sc->block_count = 0;
+}
+
+void jh512_8way_init( jh_8way_context *sc )
+{
+    // bswapped IV512
+    sc->H[ 0] = m512_const1_64( 0x17aa003e964bd16f );
+    sc->H[ 1] = m512_const1_64( 0x43d5157a052e6a63 );
+    sc->H[ 2] = m512_const1_64( 0x0bef970c8d5e228a );
+    sc->H[ 3] = m512_const1_64( 0x61c3b3f2591234e9 );
+    sc->H[ 4] = m512_const1_64( 0x1e806f53c1a01d89 );
+    sc->H[ 5] = m512_const1_64( 0x806d2bea6b05a92a );
+    sc->H[ 6] = m512_const1_64( 0xa6ba7520dbcc8e58 );
+    sc->H[ 7] = m512_const1_64( 0xf73bf8ba763a0fa9 );
+    sc->H[ 8] = m512_const1_64( 0x694ae34105e66901 );
+    sc->H[ 9] = m512_const1_64( 0x5ae66f2e8e8ab546 );
+    sc->H[10] = m512_const1_64( 0x243c84c1d0a74710 );
+    sc->H[11] = m512_const1_64( 0x99c15a2db1716e3b );
+    sc->H[12] = m512_const1_64( 0x56f8b19decf657cf );
+    sc->H[13] = m512_const1_64( 0x56b116577c8806a7 );
+    sc->H[14] = m512_const1_64( 0xfb1785e6dffcc2e3 );
+    sc->H[15] = m512_const1_64( 0x4bdd8ccc78465a54 );
+    sc->ptr = 0;
+    sc->block_count = 0;
+}
+
+static void
+jh_8way_core( jh_8way_context *sc, const void *data, size_t len )
+{
+    __m512i *buf;
+    __m512i *vdata = (__m512i*)data;
+   const int buf_size = 64;   // 64 * _m512i
+   size_t ptr;
+   DECL_STATE_8W
+
+   buf = sc->buf;
+   ptr = sc->ptr;
+
+   if ( len < (buf_size - ptr) )
+   {
+       memcpy_512( buf + (ptr>>3), vdata, len>>3 );
+       ptr += len;
+       sc->ptr = ptr;
+       return;
+   }
+
+   READ_STATE(sc);
+   while ( len > 0 )
+   {
+       size_t clen;
+       clen = buf_size - ptr;
+       if ( clen > len )
+          clen = len;
+
+       memcpy_512( buf + (ptr>>3), vdata, clen>>3 );
+       ptr += clen;
+       vdata += (clen>>3);
+       len -= clen;
+       if ( ptr == buf_size )
+       {
+          INPUT_BUF1_8W;
+          E8_8W;
+          INPUT_BUF2_8W;
+          sc->block_count ++;
+          ptr = 0;
+       }
+   }
+   WRITE_STATE(sc);
+   sc->ptr = ptr;
+}
+
+static void
+jh_8way_close( jh_8way_context *sc, unsigned ub, unsigned n, void *dst,
+               size_t out_size_w32, const void *iv )
+{
+   __m512i buf[16*4];
+   __m512i *dst512 = (__m512i*)dst;
+   size_t numz, u;
+   sph_u64 l0, l1, l0e, l1e;
+
+   buf[0] = m512_const1_64( 0x80ULL );
+
+   if ( sc->ptr == 0 )
+       numz = 48;
+   else
+       numz = 112 - sc->ptr;
+
+   memset_zero_512( buf+1, (numz>>3) - 1 );
+
+   l0 = SPH_T64(sc->block_count << 9) + (sc->ptr << 3);
+   l1 = SPH_T64(sc->block_count >> 55);
+   sph_enc64be( &l0e, l0 );
+   sph_enc64be( &l1e, l1 );
+   *(buf + (numz>>3)    ) = _mm512_set1_epi64( l1e );
+   *(buf + (numz>>3) + 1) = _mm512_set1_epi64( l0e );
+
+   jh_8way_core( sc, buf, numz + 16 );
+
+   for ( u=0; u < 8; u++ )
+       buf[u] = sc->H[u+8];
+
+    memcpy_512( dst512, buf, 8 );
+}
+
+void
+jh256_8way_update(void *cc, const void *data, size_t len)
+{
+   jh_8way_core(cc, data, len);
+}
+
+void
+jh256_8way_close(void *cc, void *dst)
+{
+   jh_8way_close(cc, 0, 0, dst, 8, IV256);
+}
+
+void
+jh512_8way_update(void *cc, const void *data, size_t len)
+{
+   jh_8way_core(cc, data, len);
+}
+
+void
+jh512_8way_close(void *cc, void *dst)
+{
+   jh_8way_close(cc, 0, 0, dst, 16, IV512);
+}
+
+#endif
+
 void jh256_4way_init( jh_4way_context *sc )
 {
     // bswapped IV256
@@ -595,16 +913,8 @@ jh_4way_close( jh_4way_context *sc, unsigned ub, unsigned n, void *dst,
     memcpy_256( dst256, buf, 8 );
 }
 
-/*
 void
-jh256_4way_init(void *cc)
-{
-	jhs_4way_init(cc, IV256);
-}
-*/
-
-void
-jh256_4way(void *cc, const void *data, size_t len)
+jh256_4way_update(void *cc, const void *data, size_t len)
 {
 	jh_4way_core(cc, data, len);
 }
@@ -615,16 +925,8 @@ jh256_4way_close(void *cc, void *dst)
 	jh_4way_close(cc, 0, 0, dst, 8, IV256);
 }
 
-/*
 void
-jh512_4way_init(void *cc)
-{
-	jhb_4way_init(cc, IV512);
-}
-*/
-
-void
-jh512_4way(void *cc, const void *data, size_t len)
+jh512_4way_update(void *cc, const void *data, size_t len)
 {
 	jh_4way_core(cc, data, len);
 }
@@ -635,6 +937,7 @@ jh512_4way_close(void *cc, void *dst)
 	jh_4way_close(cc, 0, 0, dst, 16, IV512);
 }
 
+
 #ifdef __cplusplus
 }
 #endif

algo/jh/jh-hash-4way.h

@@ -60,20 +60,41 @@ extern "C"{
  * can be cloned by copying the context (e.g. with a simple
  * <code>memcpy()</code>).
  */
+
+ 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 typedef struct {
-    __m256i buf[8] __attribute__ ((aligned (64)));
+    __m512i buf[8];
+    __m512i H[16];
+    size_t ptr;
+    uint64_t block_count;
+} jh_8way_context __attribute__ ((aligned (128)));
+
+typedef jh_8way_context jh256_8way_context;
+
+typedef jh_8way_context jh512_8way_context;
+
+void jh256_8way_init( jh_8way_context *sc);
+
+void jh256_8way_update(void *cc, const void *data, size_t len);
+
+void jh256_8way_close(void *cc, void *dst);
+
+void jh512_8way_init( jh_8way_context *sc );
+
+void jh512_8way_update(void *cc, const void *data, size_t len);
+
+void jh512_8way_close(void *cc, void *dst);
+
+#endif
+
+typedef struct {
+    __m256i buf[8];
     __m256i H[16];
     size_t ptr;
     uint64_t block_count;
-/*
-	unsigned char buf[64]; 
-	size_t ptr;
-	union {
-		sph_u64 wide[16];
-	} H;
-	sph_u64 block_count;
-*/
-} jh_4way_context;
+} jh_4way_context __attribute__ ((aligned (128)));
 
 typedef jh_4way_context jh256_4way_context;
 
@@ -81,13 +102,15 @@ typedef jh_4way_context jh512_4way_context;
 
 void jh256_4way_init( jh_4way_context *sc);
 
-void jh256_4way(void *cc, const void *data, size_t len);
+void jh256_4way_update(void *cc, const void *data, size_t len);
+#define jh256_4way jh256_4way_update
 
 void jh256_4way_close(void *cc, void *dst);
 
 void jh512_4way_init( jh_4way_context *sc );
 
-void jh512_4way(void *cc, const void *data, size_t len);
+void jh512_4way_update(void *cc, const void *data, size_t len);
+#define jh512_4way jh512_4way_update
 
 void jh512_4way_close(void *cc, void *dst);
 
@@ -95,6 +118,6 @@ void jh512_4way_close(void *cc, void *dst);
 }
 #endif
 
-#endif
+#endif // AVX2
 
 #endif

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 )

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;
 };

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

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

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
 

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)
+
+

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 */

algo/luffa/luffa-hash-2way.h

@@ -51,12 +51,30 @@
 #define LIMIT_512 128
 /*********************************/
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 typedef struct {
-    uint32 buffer[8*2] __attribute((aligned(64)));
-    __m256i chainv[10] __attribute((aligned(32)));   /* Chaining values */
+    uint32 buffer[8*4];
+    __m512i chainv[10];   /* Chaining values */
     int hashbitlen;
     int rembytes;
-} luffa_2way_context;
+} luffa_4way_context __attribute((aligned(128)));
+
+int luffa_4way_init( luffa_4way_context *state, int hashbitlen );
+int luffa_4way_update( luffa_4way_context *state, const void *data,
+                       size_t len );
+int luffa_4way_close( luffa_4way_context *state, void *hashval );
+int luffa_4way_update_close( luffa_4way_context *state, void *output,
+                                   const void *data, size_t inlen );
+
+#endif
+
+typedef struct {
+    uint32 buffer[8*2];
+    __m256i chainv[10];   /* Chaining values */
+    int hashbitlen;
+    int rembytes;
+} luffa_2way_context __attribute((aligned(128)));
 
 int luffa_2way_init( luffa_2way_context *state, int hashbitlen );
 int luffa_2way_update( luffa_2way_context *state, const void *data,

algo/luffa/luffa_for_sse2.c

@@ -541,7 +541,11 @@ 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_const_64( 0x1c1d1e1f18191a1b,
+                                                 0x1415161710111213,
+                                                 0x0c0d0e0f08090a0b,
+                                                 0x0405060700010203 );
 
     rnd512( state, zero, zero );
 
@@ -555,7 +559,9 @@ 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 +574,9 @@ 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

algo/lyra2/allium-4way.c

@@ -1,15 +1,206 @@
 #include "lyra2-gate.h"
 #include <memory.h>
 #include <mm_malloc.h>
-
-#if defined (ALLIUM_4WAY)	
-
 #include "algo/blake/blake-hash-4way.h"
 #include "algo/keccak/keccak-hash-4way.h"
 #include "algo/skein/skein-hash-4way.h"
 #include "algo/cubehash/cubehash_sse2.h"
+#include "algo/cubehash/cube-hash-2way.h"
 #include "algo/groestl/aes_ni/hash-groestl256.h"
 
+#if defined (ALLIUM_8WAY)  
+
+typedef struct {
+   blake256_8way_context     blake;
+   keccak256_8way_context    keccak;
+   cube_4way_context          cube;
+   skein256_8way_context     skein;
+   hashState_groestl256      groestl;
+} allium_8way_ctx_holder;
+
+static __thread allium_8way_ctx_holder allium_8way_ctx;
+
+bool init_allium_8way_ctx()
+{
+   keccak256_8way_init( &allium_8way_ctx.keccak );
+   cube_4way_init( &allium_8way_ctx.cube, 256, 16, 32 );
+   skein256_8way_init( &allium_8way_ctx.skein );
+   init_groestl256( &allium_8way_ctx.groestl, 32 );
+   return true;
+}
+
+void allium_8way_hash( void *state, const void *input )
+{
+   uint32_t vhash[8*8] __attribute__ ((aligned (128)));
+   uint32_t vhashA[8*8] __attribute__ ((aligned (64)));
+   uint32_t vhashB[8*8] __attribute__ ((aligned (64)));
+   uint32_t hash0[8] __attribute__ ((aligned (64)));
+   uint32_t hash1[8] __attribute__ ((aligned (64)));
+   uint32_t hash2[8] __attribute__ ((aligned (64)));
+   uint32_t hash3[8] __attribute__ ((aligned (64)));
+   uint32_t hash4[8] __attribute__ ((aligned (64)));
+   uint32_t hash5[8] __attribute__ ((aligned (64)));
+   uint32_t hash6[8] __attribute__ ((aligned (64)));
+   uint32_t hash7[8] __attribute__ ((aligned (64)));
+   allium_8way_ctx_holder ctx __attribute__ ((aligned (64)));
+
+   memcpy( &ctx, &allium_8way_ctx, sizeof(allium_8way_ctx) );
+   blake256_8way_update( &ctx.blake, input + (64<<3), 16 );
+   blake256_8way_close( &ctx.blake, vhash );
+
+   rintrlv_8x32_8x64( vhashA, vhash, 256 );
+   keccak256_8way_update( &ctx.keccak, vhashA, 32 );
+   keccak256_8way_close( &ctx.keccak, vhash );
+
+   dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+                 vhash, 256 );
+
+
+   intrlv_2x256( vhash, hash0, hash1, 256 );
+   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
+   dintrlv_2x256( hash0, hash1, vhash, 256 );
+   intrlv_2x256( vhash, hash2, hash3, 256 );
+   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
+   dintrlv_2x256( hash2, hash3, vhash, 256 );
+   intrlv_2x256( vhash, hash4, hash5, 256 );
+   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
+   dintrlv_2x256( hash4, hash5, vhash, 256 );
+   intrlv_2x256( vhash, hash6, hash7, 256 );
+   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
+   dintrlv_2x256( hash6, hash7, vhash, 256 );
+  
+/* 
+   LYRA2RE( hash0, 32, hash0, 32, hash0, 32, 1, 8, 8 );
+   LYRA2RE( hash1, 32, hash1, 32, hash1, 32, 1, 8, 8 );
+   LYRA2RE( hash2, 32, hash2, 32, hash2, 32, 1, 8, 8 );
+   LYRA2RE( hash3, 32, hash3, 32, hash3, 32, 1, 8, 8 );
+   LYRA2RE( hash4, 32, hash4, 32, hash4, 32, 1, 8, 8 );
+   LYRA2RE( hash5, 32, hash5, 32, hash5, 32, 1, 8, 8 );
+   LYRA2RE( hash6, 32, hash6, 32, hash6, 32, 1, 8, 8 );
+   LYRA2RE( hash7, 32, hash7, 32, hash7, 32, 1, 8, 8 );
+*/
+
+
+
+   intrlv_4x128( vhashA, hash0, hash1, hash2, hash3, 256 );
+   intrlv_4x128( vhashB, hash4, hash5, hash6, hash7, 256 );
+
+   cube_4way_update_close( &ctx.cube, vhashA, vhashA, 32 );
+   cube_4way_init( &ctx.cube, 256, 16, 32 );
+   cube_4way_update_close( &ctx.cube, vhashB, vhashB, 32 );
+
+   dintrlv_4x128( hash0, hash1, hash2, hash3, vhashA, 256 );
+   dintrlv_4x128( hash4, hash5, hash6, hash7, vhashB, 256 );
+
+   intrlv_2x256( vhash, hash0, hash1, 256 );
+   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
+   dintrlv_2x256( hash0, hash1, vhash, 256 );
+   intrlv_2x256( vhash, hash2, hash3, 256 );
+   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
+   dintrlv_2x256( hash2, hash3, vhash, 256 );
+   intrlv_2x256( vhash, hash4, hash5, 256 );
+   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
+   dintrlv_2x256( hash4, hash5, vhash, 256 );
+   intrlv_2x256( vhash, hash6, hash7, 256 );
+   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
+   dintrlv_2x256( hash6, hash7, vhash, 256 );
+
+
+/*
+   LYRA2RE( hash0, 32, hash0, 32, hash0, 32, 1, 8, 8 );
+   LYRA2RE( hash1, 32, hash1, 32, hash1, 32, 1, 8, 8 );
+   LYRA2RE( hash2, 32, hash2, 32, hash2, 32, 1, 8, 8 );
+   LYRA2RE( hash3, 32, hash3, 32, hash3, 32, 1, 8, 8 );
+   LYRA2RE( hash4, 32, hash4, 32, hash4, 32, 1, 8, 8 );
+   LYRA2RE( hash5, 32, hash5, 32, hash5, 32, 1, 8, 8 );
+   LYRA2RE( hash6, 32, hash6, 32, hash6, 32, 1, 8, 8 );
+   LYRA2RE( hash7, 32, hash7, 32, hash7, 32, 1, 8, 8 );
+*/
+
+
+
+   intrlv_8x64( vhash, hash0, hash1, hash2, hash3, hash4, hash5, hash6,
+                hash7, 256 );
+
+   skein256_8way_update( &ctx.skein, vhash, 32 );
+   skein256_8way_close( &ctx.skein, vhash );
+
+   dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+                 vhash, 256 );
+
+   update_and_final_groestl256( &ctx.groestl, state, hash0, 256 );
+   memcpy( &ctx.groestl, &allium_8way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, state+32, hash1, 256 );
+   memcpy( &ctx.groestl, &allium_8way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, state+64, hash2, 256 );
+   memcpy( &ctx.groestl, &allium_8way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, state+96, hash3, 256 );
+   memcpy( &ctx.groestl, &allium_8way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, state+128, hash4, 256 );
+   memcpy( &ctx.groestl, &allium_8way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, state+160, hash5, 256 );
+   memcpy( &ctx.groestl, &allium_8way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, state+192, hash6, 256 );
+   memcpy( &ctx.groestl, &allium_8way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, state+224, hash7, 256 );
+   memcpy( &ctx.groestl, &allium_8way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+}
+
+int scanhash_allium_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 *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   uint32_t n = first_nonce;
+   const uint32_t last_nonce = max_nonce - 8;
+   const uint32_t Htarg = ptarget[7];
+   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
+   int thr_id = mythr->id;  // thr_id arg is deprecated
+
+   if ( opt_benchmark )
+      ( (uint32_t*)ptarget )[7] = 0x0000ff;
+
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
+   blake256_8way_init( &allium_8way_ctx.blake );
+   blake256_8way_update( &allium_8way_ctx.blake, vdata, 64 );
+
+   do {
+     *noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
+                                                 n+3, n+2, n+1, n ) );
+
+     allium_8way_hash( hash, vdata );
+     pdata[19] = n;
+
+     for ( int lane = 0; lane < 8; lane++ ) if ( (hash+(lane<<3))[7] <= Htarg )
+     {
+        if ( fulltest( hash+(lane<<3), ptarget ) && !opt_benchmark )
+        {
+           pdata[19] = n + lane;
+           submit_lane_solution( work, hash+(lane<<3), mythr, lane );
+         }
+     }
+     n += 8;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce;
+   return 0;
+}
+
+
+#elif defined (ALLIUM_4WAY)  
+
+
 typedef struct {
    blake256_4way_context     blake;
    keccak256_4way_context    keccak;

algo/lyra2/lyra2-gate.c

@@ -44,8 +44,13 @@ bool lyra2rev3_thread_init()
 {
    const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * 4; // nCols
    const int64_t ROW_LEN_BYTES = ROW_LEN_INT64 * 8;
+   int size = ROW_LEN_BYTES * 4; // nRows;
 
-   int size = (int64_t)ROW_LEN_BYTES * 4; // nRows;
+#if defined(LYRA2REV3_16WAY)
+//   l2v3_wholeMatrix = _mm_malloc( 2*size, 128 );
+   l2v3_wholeMatrix = _mm_malloc( 2*size, 64 );
+   init_lyra2rev3_16way_ctx();;
+#else
    l2v3_wholeMatrix = _mm_malloc( size, 64 );
 #if defined (LYRA2REV3_8WAY)
    init_lyra2rev3_8way_ctx();;
@@ -53,13 +58,17 @@ bool lyra2rev3_thread_init()
    init_lyra2rev3_4way_ctx();;
 #else
    init_lyra2rev3_ctx();
+#endif
 #endif
    return l2v3_wholeMatrix;
 }
 
 bool register_lyra2rev3_algo( algo_gate_t* gate )
 {
-#if defined (LYRA2REV3_8WAY)
+#if defined(LYRA2REV3_16WAY)
+  gate->scanhash  = (void*)&scanhash_lyra2rev3_16way;
+  gate->hash      = (void*)&lyra2rev3_16way_hash;
+#elif defined (LYRA2REV3_8WAY)
   gate->scanhash  = (void*)&scanhash_lyra2rev3_8way;
   gate->hash      = (void*)&lyra2rev3_8way_hash;
 #elif defined (LYRA2REV3_4WAY)
@@ -69,7 +78,7 @@ bool register_lyra2rev3_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_lyra2rev3;
   gate->hash      = (void*)&lyra2rev3_hash;
 #endif
-  gate->optimizations = SSE2_OPT | SSE42_OPT | AVX2_OPT;
+  gate->optimizations = SSE2_OPT | SSE42_OPT | AVX2_OPT | AVX512_OPT;
   gate->miner_thread_init = (void*)&lyra2rev3_thread_init;
   opt_target_factor = 256.0;
   return true;
@@ -85,10 +94,14 @@ bool lyra2rev2_thread_init()
    const int64_t ROW_LEN_BYTES = ROW_LEN_INT64 * 8;
 
    int size = (int64_t)ROW_LEN_BYTES * 4; // nRows;
+#if defined (LYRA2REV2_8WAY)
+   l2v2_wholeMatrix = _mm_malloc( 2 * size, 64 );   // 2 way
+   init_lyra2rev2_8way_ctx();;
+#elif defined (LYRA2REV2_4WAY)
    l2v2_wholeMatrix = _mm_malloc( size, 64 );
-#if defined (LYRA2REV2_4WAY)
    init_lyra2rev2_4way_ctx();;
 #else
+   l2v2_wholeMatrix = _mm_malloc( size, 64 );
    init_lyra2rev2_ctx();
 #endif
    return l2v2_wholeMatrix;
@@ -96,14 +109,17 @@ bool lyra2rev2_thread_init()
 
 bool register_lyra2rev2_algo( algo_gate_t* gate )
 {
-#if defined (LYRA2REV2_4WAY)
+#if defined (LYRA2REV2_8WAY)
+  gate->scanhash  = (void*)&scanhash_lyra2rev2_8way;
+  gate->hash      = (void*)&lyra2rev2_8way_hash;
+#elif defined (LYRA2REV2_4WAY)
   gate->scanhash  = (void*)&scanhash_lyra2rev2_4way;
   gate->hash      = (void*)&lyra2rev2_4way_hash;
 #else
   gate->scanhash  = (void*)&scanhash_lyra2rev2;
   gate->hash      = (void*)&lyra2rev2_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | SSE42_OPT | AVX2_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | SSE42_OPT | AVX2_OPT | AVX512_OPT;
   gate->miner_thread_init = (void*)&lyra2rev2_thread_init;
   opt_target_factor = 256.0;
   return true;
@@ -113,7 +129,11 @@ bool register_lyra2rev2_algo( algo_gate_t* gate )
 
 bool register_lyra2z_algo( algo_gate_t* gate )
 {
-#if defined(LYRA2Z_8WAY)
+#if defined(LYRA2Z_16WAY)
+  gate->miner_thread_init = (void*)&lyra2z_16way_thread_init;
+  gate->scanhash   = (void*)&scanhash_lyra2z_16way;
+  gate->hash       = (void*)&lyra2z_16way_hash;
+#elif defined(LYRA2Z_8WAY)
   gate->miner_thread_init = (void*)&lyra2z_8way_thread_init;
   gate->scanhash   = (void*)&scanhash_lyra2z_8way;
   gate->hash       = (void*)&lyra2z_8way_hash;
@@ -126,7 +146,7 @@ bool register_lyra2z_algo( algo_gate_t* gate )
   gate->scanhash   = (void*)&scanhash_lyra2z;
   gate->hash       = (void*)&lyra2z_hash;
 #endif
-  gate->optimizations = SSE42_OPT | AVX2_OPT;
+  gate->optimizations = SSE42_OPT | AVX2_OPT | AVX512_OPT;
   opt_target_factor = 256.0;
   return true;
 };
@@ -154,7 +174,11 @@ bool register_lyra2h_algo( algo_gate_t* gate )
 
 bool register_allium_algo( algo_gate_t* gate )
 {
-#if defined (ALLIUM_4WAY)
+#if defined (ALLIUM_8WAY)
+  gate->miner_thread_init = (void*)&init_allium_8way_ctx;
+  gate->scanhash  = (void*)&scanhash_allium_8way;
+  gate->hash      = (void*)&allium_8way_hash;
+#elif defined (ALLIUM_4WAY)
   gate->miner_thread_init = (void*)&init_allium_4way_ctx;
   gate->scanhash  = (void*)&scanhash_allium_4way;
   gate->hash      = (void*)&allium_4way_hash;
@@ -163,7 +187,7 @@ bool register_allium_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_allium;
   gate->hash      = (void*)&allium_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | SSE42_OPT | AVX2_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | SSE42_OPT | AVX2_OPT | AVX512_OPT;
   opt_target_factor = 256.0;
   return true;
 };

algo/lyra2/lyra2-gate.h

@@ -5,18 +5,27 @@
 #include <stdint.h>
 #include "lyra2.h"
 
-#if defined(__AVX2__)
-  #define LYRA2REV3_8WAY
-#endif
 
-#if defined(__SSE2__)
-  #define LYRA2REV3_4WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define LYRA2REV3_16WAY 1
+#elif defined(__AVX2__)
+  #define LYRA2REV3_8WAY 1
+#elif defined(__SSE2__)
+  #define LYRA2REV3_4WAY 1
 #endif
 
 extern __thread uint64_t* l2v3_wholeMatrix;
 
 bool register_lyra2rev3_algo( algo_gate_t* gate );
-#if defined(LYRA2REV3_8WAY)
+
+#if defined(LYRA2REV3_16WAY)
+
+void lyra2rev3_16way_hash( void *state, const void *input );
+int scanhash_lyra2rev3_16way( struct work *work, uint32_t max_nonce,
+                             uint64_t *hashes_done, struct thr_info *mythr );
+bool init_lyra2rev3_16way_ctx();
+
+#elif defined(LYRA2REV3_8WAY)
 
 void lyra2rev3_8way_hash( void *state, const void *input );
 int scanhash_lyra2rev3_8way( struct work *work, uint32_t max_nonce,
@@ -41,15 +50,24 @@ bool init_lyra2rev3_ctx();
 
 //////////////////////////////////
 
-#if defined(__AVX2__)
-  #define LYRA2REV2_4WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define LYRA2REV2_8WAY 1
+#elif defined(__AVX2__)
+  #define LYRA2REV2_4WAY 1
 #endif
 
 extern __thread uint64_t* l2v2_wholeMatrix;
 
 bool register_lyra2rev2_algo( algo_gate_t* gate );
 
-#if defined(LYRA2REV2_4WAY)
+#if defined(LYRA2REV2_8WAY)
+
+void lyra2rev2_8way_hash( void *state, const void *input );
+int scanhash_lyra2rev2_8way( struct work *work, uint32_t max_nonce,
+                             uint64_t *hashes_done, struct thr_info *mythr );
+bool init_lyra2rev2_8way_ctx();
+
+#elif defined(LYRA2REV2_4WAY)
 
 void lyra2rev2_4way_hash( void *state, const void *input );
 int scanhash_lyra2rev2_4way( struct work *work, uint32_t max_nonce,
@@ -67,17 +85,25 @@ bool init_lyra2rev2_ctx();
 
 /////////////////////////
 
-#if defined(__SSE2__)
-  #define LYRA2Z_4WAY
-#endif
-#if defined(__AVX2__)
-  #define LYRA2Z_8WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define LYRA2Z_16WAY 1
+#elif defined(__AVX2__)
+  #define LYRA2Z_8WAY 1
+#elif defined(__SSE2__)
+  #define LYRA2Z_4WAY 1
 #endif
 
 
 #define LYRA2Z_MATRIX_SIZE  BLOCK_LEN_INT64 * 8 * 8 * 8
 
-#if defined(LYRA2Z_8WAY)
+#if defined(LYRA2Z_16WAY)
+
+void lyra2z_16way_hash( void *state, const void *input );
+int scanhash_lyra2z_16way( struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done, struct thr_info *mythr );
+bool lyra2z_16way_thread_init();
+
+#elif defined(LYRA2Z_8WAY)
 
 void lyra2z_8way_hash( void *state, const void *input );
 int scanhash_lyra2z_8way( struct work *work, uint32_t max_nonce,
@@ -126,13 +152,22 @@ bool lyra2h_thread_init();
 
 //////////////////////////////////
 
-#if defined(__AVX2__) && defined(__AES__)
-  #define ALLIUM_4WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define ALLIUM_8WAY 1
+#elif defined(__AVX2__) && defined(__AES__)
+  #define ALLIUM_4WAY 1
 #endif
 
 bool register_allium_algo( algo_gate_t* gate );
 
-#if defined(ALLIUM_4WAY)
+#if defined(ALLIUM_8WAY)
+
+void allium_8way_hash( void *state, const void *input );
+int scanhash_allium_8way( struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done, struct thr_info *mythr );
+bool init_allium_8way_ctx();
+
+#elif defined(ALLIUM_4WAY)
 
 void allium_4way_hash( void *state, const void *input );
 int scanhash_allium_4way( struct work *work, uint32_t max_nonce,

algo/lyra2/lyra2-hash-2way.c

@@ -0,0 +1,578 @@
+/**
+ * Implementation of the Lyra2 Password Hashing Scheme (PHS).
+ *
+ * Author: The Lyra PHC team (http://www.lyra-kdf.net/) -- 2014.
+ *
+ * This software is hereby placed in the public domain.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ''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 AUTHORS 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.
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <mm_malloc.h>
+#include "compat.h"
+#include "lyra2.h"
+#include "sponge.h"
+
+//  LYRA2RE 8 cols 8 rows used by lyra2re, allium, phi2, x22i, x25x, 
+//  dynamic matrix allocation.
+//
+//  LYRA2REV2 4 cols 4 rows used by lyra2rev2 and x21s, static matrix
+//  allocation.
+//
+//  LYRA2REV3 4 cols 4 rows with an extra twist in calculating
+//  rowa in the wandering phase. Used by lyra2rev3. Static matrix
+//  allocation.
+// 
+//  LYRA2Z various cols & rows and supports 80 byte input. Used by lyra2z,
+//  lyra2z330, lyra2h, 
+
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+/**
+ * Executes Lyra2 based on the G function from Blake2b. This version supports salts and passwords
+ * whose combined length is smaller than the size of the memory matrix, (i.e., (nRows x nCols x b) bits,
+ * where "b" is the underlying sponge's bitrate). In this implementation, the "basil" is composed by all
+ * integer parameters (treated as type "unsigned int") in the order they are provided, plus the value
+ * of nCols, (i.e., basil = kLen || pwdlen || saltlen || timeCost || nRows || nCols).
+ *
+ * @param K The derived key to be output by the algorithm
+ * @param kLen Desired key length
+ * @param pwd User password
+ * @param pwdlen Password length
+ * @param salt Salt
+ * @param saltlen Salt length
+ * @param timeCost Parameter to determine the processing time (T)
+ * @param nRows Number or rows of the memory matrix (R)
+ * @param nCols Number of columns of the memory matrix (C)
+ *
+ * @return 0 if the key is generated correctly; -1 if there is an error (usually due to lack of memory for allocation)
+ */
+
+// For lyra2rev3.
+// convert a simple offset to an index into 2x4 u64 interleaved data.
+// good for state and 4 row matrix. 
+// index = ( int( off / 4 ) * 2 ) + ( off mod 4 )
+
+#define offset_to_index( o ) \
+   ( ( ( (uint64_t)( (o) & 0xf) / 4 ) * 8 ) + ( (o) % 4 ) )
+
+
+int LYRA2REV2_2WAY( uint64_t* wholeMatrix, void *K, uint64_t kLen,
+             const void *pwd, const uint64_t pwdlen, const uint64_t timeCost,
+             const uint64_t nRows, const uint64_t nCols )
+{
+   //====================== Basic variables ============================//
+   uint64_t _ALIGN(256) state[32];
+   int64_t row = 2;
+   int64_t prev = 1;
+   int64_t rowa0 = 0;
+   int64_t rowa1 = 0;
+   int64_t tau; 
+   int64_t step = 1;
+   int64_t window = 2;
+   int64_t gap = 1;
+   //====================================================================/
+
+   const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * nCols;
+
+   // for Lyra2REv2, nCols = 4, v1 was using 8
+   const int64_t BLOCK_LEN = (nCols == 4) ? BLOCK_LEN_BLAKE2_SAFE_INT64
+                                          : BLOCK_LEN_BLAKE2_SAFE_BYTES;
+   uint64_t *ptrWord = wholeMatrix;
+
+   int64_t nBlocksInput = ( ( pwdlen + pwdlen + 6 * sizeof(uint64_t) )
+                              / BLOCK_LEN_BLAKE2_SAFE_BYTES ) + 1;
+
+   uint64_t *ptr = wholeMatrix;
+   uint64_t *pw = (uint64_t*)pwd;
+
+   memcpy( ptr, pw, 2*pwdlen ); // password 
+   ptr += pwdlen>>2;
+   memcpy( ptr, pw, 2*pwdlen ); // password lane 1
+   ptr += pwdlen>>2;
+
+   // now build the rest interleaving on the fly.
+
+   ptr[0] = ptr[ 4] = kLen;
+   ptr[1] = ptr[ 5] = pwdlen;
+   ptr[2] = ptr[ 6] = pwdlen;   // saltlen
+   ptr[3] = ptr[ 7] = timeCost;
+   ptr[8] = ptr[12] = nRows;
+   ptr[9] = ptr[13] = nCols;
+   ptr[10] = ptr[14] = 0x80;
+   ptr[11] = ptr[15] = 0x0100000000000000;
+
+   ptrWord = wholeMatrix;
+
+   absorbBlockBlake2Safe_2way( state, ptrWord, nBlocksInput, BLOCK_LEN );
+
+   //Initializes M[0] and M[1]
+   reducedSqueezeRow0_2way( state, &wholeMatrix[0], nCols );
+
+   reducedDuplexRow1_2way( state, &wholeMatrix[0],
+                           &wholeMatrix[ 2 * ROW_LEN_INT64 ],  nCols );
+
+   do
+   {
+     //M[row] = rand; //M[row*] = M[row*] XOR rotW(rand)
+
+     reducedDuplexRowSetup_2way( state, &wholeMatrix[ 2* prev * ROW_LEN_INT64],
+                                        &wholeMatrix[ 2* rowa0 * ROW_LEN_INT64],
+                                        &wholeMatrix[ 2* row*ROW_LEN_INT64],
+                                        nCols );
+
+     rowa0 = (rowa0 + step) & (window - 1);
+
+     prev = row;
+     row++;
+
+     if ( rowa0 == 0 )
+     {
+        step = window + gap;
+        window *= 2; 
+        gap = -gap;
+     }
+   } while ( row < nRows );
+
+   //===================== Wandering Phase =============================//
+   row = 0;
+   for ( tau = 1; tau <= timeCost; tau++ )
+   {
+      step = ( (tau & 1) == 0 ) ? -1 : ( nRows >> 1 ) - 1;
+      do
+      {
+        rowa0 = state[ 0 ] & (unsigned int)(nRows-1);
+        rowa1 = state[ 4 ] & (unsigned int)(nRows-1);
+
+        reducedDuplexRow_2way( state, &wholeMatrix[ 2* prev * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* rowa0 * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* rowa1 * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* row *ROW_LEN_INT64 ],
+                                      nCols );
+         prev = row;
+
+         row = (row + step) & (unsigned int)(nRows-1); //(USE THIS IF nRows IS A POWER OF 2)
+
+      } while (row != 0);
+   }
+
+   //===================== Wrap-up Phase ===============================//
+   //Absorbs the last block of the memory matrix
+   absorbBlock_2way( state, &wholeMatrix[ 2 * rowa0 *ROW_LEN_INT64 ],
+                            &wholeMatrix[ 2 * rowa1 *ROW_LEN_INT64 ] );
+   //Squeezes the key
+   squeeze_2way( state, K, (unsigned int) kLen );
+
+   return 0;
+}
+
+// This version is currently only used by REv3 and has some hard coding
+// specific to v3 such as input data size of 32 bytes.
+//
+// Similarly with REv2. Thedifference with REv3 isn't clear and maybe
+// they can be merged.
+//
+// RE is used by RE, allium. The main difference between RE and REv2
+// in the matrix size.
+//
+// Z also needs to support 80 byte input as well as 32 byte, and odd
+// matrix sizes like 330 rows. It is used by lyra2z330, lyra2z, lyra2h.
+
+
+/////////////////////////////////////////////////
+
+// 2 way 256
+// drop salt, salt len arguments, hard code some others.
+// Data is interleaved 2x256.
+
+int LYRA2REV3_2WAY( uint64_t* wholeMatrix, void *K, uint64_t kLen,
+                    const void *pwd, uint64_t pwdlen, uint64_t timeCost,
+                    uint64_t nRows, uint64_t nCols )
+
+// hard coded for 32 byte input as well as matrix size.
+// Other required versions include 80 byte input and different block
+// sizes.
+
+{
+   //====================== Basic variables ============================//
+   uint64_t _ALIGN(256) state[32];
+   int64_t row = 2; 
+   int64_t prev = 1;
+   int64_t rowa0 = 0;
+   int64_t rowa1 = 0;
+   int64_t tau; 
+   int64_t step = 1;
+   int64_t window = 2;
+   int64_t gap = 1; 
+   uint64_t instance0 = 0;
+   uint64_t instance1 = 0;
+   //====================================================================/
+
+   const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * nCols;
+   const int64_t BLOCK_LEN = BLOCK_LEN_BLAKE2_SAFE_INT64;
+
+   uint64_t *ptrWord = wholeMatrix;
+
+//  2 way 256 rewrite. Salt always == password, and data is interleaved,
+//  need to build in parallel as pw isalready interleaved.
+
+   
+//  {   password,    (64 or 80 bytes)
+//      salt,        (64 or 80 bytes) =  same as password
+//      Klen,        (u64)  = 32 bytes
+//      pwdlen,      (u64)
+//      saltlen,     (u64)
+//      timecost,    (u64)
+//      nrows,       (u64)
+//      ncols,       (u64)
+//      0x80,        (byte)
+//      { 0 .. 0 },
+//      1            (byte)
+//   }
+   
+// input is usually 32 maybe 64, both are aligned to 256 bit vector.
+// 80 byte inpput is not aligned complicating matters for lyra2z.   
+
+   int64_t nBlocksInput = ( ( pwdlen + pwdlen + 6 * sizeof(uint64_t) )
+                              / BLOCK_LEN_BLAKE2_SAFE_BYTES ) + 1;
+   
+   uint64_t *ptr = wholeMatrix;
+   uint64_t *pw = (uint64_t*)pwd;
+
+   memcpy( ptr, pw, 2*pwdlen ); // password 
+   ptr += pwdlen>>2;
+   memcpy( ptr, pw, 2*pwdlen ); // password lane 1
+   ptr += pwdlen>>2;
+ 
+   // now build the rest interleaving on the fly.
+
+   ptr[0] = ptr[ 4] = kLen;
+   ptr[1] = ptr[ 5] = pwdlen;
+   ptr[2] = ptr[ 6] = pwdlen;   // saltlen
+   ptr[3] = ptr[ 7] = timeCost;
+   ptr[8] = ptr[12] = nRows;
+   ptr[9] = ptr[13] = nCols;
+   ptr[10] = ptr[14] = 0x80;
+   ptr[11] = ptr[15] = 0x0100000000000000;
+
+   ptrWord = wholeMatrix;
+
+   absorbBlockBlake2Safe_2way( state, ptrWord, nBlocksInput, BLOCK_LEN );
+
+   reducedSqueezeRow0_2way( state, &wholeMatrix[0], nCols );
+
+   reducedDuplexRow1_2way( state, &wholeMatrix[0],
+                           &wholeMatrix[2*ROW_LEN_INT64],  nCols );
+
+   do
+   {
+
+      reducedDuplexRowSetup_2way( state, &wholeMatrix[ 2* prev*ROW_LEN_INT64 ],
+                                         &wholeMatrix[ 2* rowa0*ROW_LEN_INT64 ],
+                                         &wholeMatrix[ 2* row*ROW_LEN_INT64 ],
+                                         nCols );
+
+      rowa0 = (rowa0 + step) & (window - 1);
+
+      prev = row;
+      row++;
+
+      if (rowa0 == 0)
+      {
+         step = window + gap; //changes the step: approximately doubles its value
+         window *= 2; //doubles the size of the re-visitation window
+         gap = -gap; //inverts the modifier to the step
+      }
+
+   } while (row < nRows);
+
+   row = 0;
+   for (tau = 1; tau <= timeCost; tau++)
+   {
+      step = ( (tau & 1) == 0 ) ? -1 : ( nRows >> 1 ) - 1;
+      do
+      {
+        instance0 = state[ offset_to_index( instance0 ) ];
+        instance1 = (&state[4])[ offset_to_index( instance1 ) ];
+
+        rowa0 = state[ offset_to_index( instance0 )  ]
+                & (unsigned int)(nRows-1);
+        rowa1 = (state+4)[ offset_to_index( instance1 ) ]
+                & (unsigned int)(nRows-1);
+
+        reducedDuplexRow_2way( state, &wholeMatrix[ 2* prev * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* rowa0 * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* rowa1 * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* row*ROW_LEN_INT64 ],
+                                      nCols );
+
+        prev = row;
+        row = (row + step) & (unsigned int)(nRows-1); 
+
+       } while ( row != 0 );
+   }
+
+   absorbBlock_2way( state, &wholeMatrix[2*rowa0*ROW_LEN_INT64],
+                            &wholeMatrix[2*rowa1*ROW_LEN_INT64] );
+
+   squeeze_2way( state, K, (unsigned int) kLen );
+
+   return 0;
+}
+
+//////////////////////////////////////////////////
+
+int LYRA2Z_2WAY( uint64_t* wholeMatrix, void *K, uint64_t kLen,
+               const void *pwd, const uint64_t pwdlen, const uint64_t timeCost,
+               const uint64_t nRows, const uint64_t nCols )
+{
+    //========================== Basic variables ============================//
+    uint64_t _ALIGN(256) state[32];
+    int64_t row = 2;
+    int64_t prev = 1;
+    int64_t rowa0 = 0;
+    int64_t rowa1 = 0;
+    int64_t tau; 
+    int64_t step = 1;
+    int64_t window = 2;
+    int64_t gap = 1; 
+    //=======================================================================/
+
+    const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * nCols;
+
+    //First, we clean enough blocks for the password, salt, basil and padding
+    uint64_t nBlocksInput = ( ( pwdlen + pwdlen + 6 *
+                       sizeof (uint64_t) ) / BLOCK_LEN_BLAKE2_SAFE_BYTES ) + 1;
+
+   uint64_t *ptr = wholeMatrix;
+   uint64_t *pw = (uint64_t*)pwd;
+
+   memcpy( ptr, pw, 2*pwdlen ); // password 
+   ptr += pwdlen>>2;
+   memcpy( ptr, pw, 2*pwdlen ); // password lane 1
+   ptr += pwdlen>>2;
+
+   // now build the rest interleaving on the fly.
+   ptr[0] = ptr[ 4] = kLen;
+   ptr[1] = ptr[ 5] = pwdlen;
+   ptr[2] = ptr[ 6] = pwdlen;   // saltlen
+   ptr[3] = ptr[ 7] = timeCost;
+   ptr[8] = ptr[12] = nRows;
+   ptr[9] = ptr[13] = nCols;
+   ptr[10] = ptr[14] = 0x80;
+   ptr[11] = ptr[15] = 0x0100000000000000;
+
+   uint64_t *ptrWord = wholeMatrix;
+
+   absorbBlockBlake2Safe_2way( state, ptrWord, nBlocksInput,
+                               BLOCK_LEN_BLAKE2_SAFE_INT64 );
+
+   //Initializes M[0] and M[1]
+   reducedSqueezeRow0_2way( state, &wholeMatrix[0], nCols );
+
+   reducedDuplexRow1_2way( state, &wholeMatrix[0],
+                           &wholeMatrix[ 2 * ROW_LEN_INT64 ],  nCols );
+
+   do
+   {
+     //M[row] = rand; //M[row*] = M[row*] XOR rotW(rand)
+
+     reducedDuplexRowSetup_2way( state, &wholeMatrix[ 2* prev * ROW_LEN_INT64],
+                                        &wholeMatrix[ 2* rowa0 * ROW_LEN_INT64],
+                                        &wholeMatrix[ 2* row*ROW_LEN_INT64],
+                                        nCols );
+
+     rowa0 = (rowa0 + step) & (window - 1);
+     prev = row;
+     row++;
+
+     if ( rowa0 == 0 )
+     {
+        step = window + gap;
+        window *= 2;
+        gap = -gap;
+     }
+   } while ( row < nRows );
+
+   row = 0;
+   for ( tau = 1; tau <= timeCost; tau++ )
+   {
+        step = (tau % 2 == 0) ? -1 : nRows / 2 - 1;
+      do
+      {
+        rowa0 = state[ 0 ] % nRows;
+        rowa1 = state[ 4 ] % nRows;
+
+        reducedDuplexRow_2way( state, &wholeMatrix[ 2* prev * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* rowa0 * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* rowa1 * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* row *ROW_LEN_INT64 ],
+                                      nCols );
+
+        prev = row;
+        row = (row + step) % nRows;
+
+      } while (row != 0);
+   }
+
+   absorbBlock_2way( state, &wholeMatrix[ 2 * rowa0 *ROW_LEN_INT64 ],
+                            &wholeMatrix[ 2 * rowa1 *ROW_LEN_INT64 ] );
+
+   //Squeezes the key
+   squeeze_2way( state, K, (unsigned int) kLen );
+
+   return 0;
+}
+
+////////////////////////////////////////////////////
+
+// Lyra2RE doesn't like the new wholeMatrix implementation
+int LYRA2RE_2WAY( void *K, uint64_t kLen, const void *pwd,
+                  const uint64_t pwdlen, const uint64_t timeCost,
+                  const uint64_t nRows, const uint64_t nCols )
+{
+   //====================== Basic variables ============================//
+   uint64_t _ALIGN(256) state[32];
+   int64_t row = 2; //index of row to be processed
+   int64_t prev = 1; //index of prev (last row ever computed/modified)
+   int64_t rowa0 = 0;
+   int64_t rowa1 = 0;
+   int64_t tau; //Time Loop iterator
+   int64_t step = 1; //Visitation step (used during Setup and Wandering phases)
+   int64_t window = 2; //Visitation window (used to define which rows can be revisited during Setup)
+   int64_t gap = 1; //Modifier to the step, assuming the values 1 or -1
+   int64_t i; //auxiliary iteration counter
+   //====================================================================/
+
+   //=== Initializing the Memory Matrix and pointers to it =============//
+   //Tries to allocate enough space for the whole memory matrix
+
+   const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * nCols;
+   const int64_t ROW_LEN_BYTES = ROW_LEN_INT64 * 8;
+   // for Lyra2REv2, nCols = 4, v1 was using 8
+   const int64_t BLOCK_LEN = (nCols == 4) ? BLOCK_LEN_BLAKE2_SAFE_INT64
+                                          : BLOCK_LEN_BLAKE2_SAFE_BYTES;
+
+   i = (int64_t)ROW_LEN_BYTES * nRows;
+   uint64_t *wholeMatrix = _mm_malloc( 2*i, 64 );
+   if (wholeMatrix == NULL)
+      return -1;
+
+   memset_zero_512( (__m512i*)wholeMatrix, i>>5 );
+
+   uint64_t *ptrWord = wholeMatrix;
+   uint64_t *pw = (uint64_t*)pwd;
+
+   //First, we clean enough blocks for the password, salt, basil and padding
+   int64_t nBlocksInput = ( ( pwdlen + pwdlen + 6 * sizeof(uint64_t) )
+                              / BLOCK_LEN_BLAKE2_SAFE_BYTES ) + 1;
+
+   uint64_t *ptr = wholeMatrix;
+
+   memcpy( ptr, pw, 2*pwdlen ); // password 
+   ptr += pwdlen>>2;
+   memcpy( ptr, pw, 2*pwdlen ); // password lane 1
+   ptr += pwdlen>>2;
+
+   // now build the rest interleaving on the fly.
+
+   ptr[0] = ptr[ 4] = kLen;
+   ptr[1] = ptr[ 5] = pwdlen;
+   ptr[2] = ptr[ 6] = pwdlen;   // saltlen
+   ptr[3] = ptr[ 7] = timeCost;
+   ptr[8] = ptr[12] = nRows;
+   ptr[9] = ptr[13] = nCols;
+   ptr[10] = ptr[14] = 0x80;
+   ptr[11] = ptr[15] = 0x0100000000000000;
+
+   absorbBlockBlake2Safe_2way( state, ptrWord, nBlocksInput, BLOCK_LEN );
+
+   //Initializes M[0] and M[1]
+   reducedSqueezeRow0_2way( state, &wholeMatrix[0], nCols ); //The locally copied password is most likely overwritten here
+
+   reducedDuplexRow1_2way( state, &wholeMatrix[0],
+                                  &wholeMatrix[ 2 * ROW_LEN_INT64], nCols );
+
+   do
+   {
+      //M[row] = rand; //M[row*] = M[row*] XOR rotW(rand)
+
+      reducedDuplexRowSetup_2way( state, &wholeMatrix[ 2* prev*ROW_LEN_INT64 ],
+                                         &wholeMatrix[ 2* rowa0*ROW_LEN_INT64 ],
+                                         &wholeMatrix[ 2* row*ROW_LEN_INT64 ],
+                                         nCols );
+
+      //updates the value of row* (deterministically picked during Setup))
+      rowa0 = (rowa0 + step) & (window - 1);
+      //update prev: it now points to the last row ever computed
+
+      prev = row;
+      //updates row: goes to the next row to be computed
+      row++;
+
+      //Checks if all rows in the window where visited.
+      if (rowa0 == 0)
+      {
+         step = window + gap; //changes the step: approximately doubles its value
+         window *= 2; //doubles the size of the re-visitation window
+         gap = -gap; //inverts the modifier to the step
+      }
+
+   } while (row < nRows);
+
+   //===================== Wandering Phase =============================//
+   row = 0; //Resets the visitation to the first row of the memory matrix
+   for (tau = 1; tau <= timeCost; tau++)
+   {
+      step = ((tau & 1) == 0) ? -1 : (nRows >> 1) - 1;
+      do
+      {
+        rowa0 = state[ 0 ] & (unsigned int)(nRows-1);
+        rowa1 = state[ 4 ] & (unsigned int)(nRows-1);
+
+        reducedDuplexRow_2way( state, &wholeMatrix[ 2* prev * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* rowa0 * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* rowa1 * ROW_LEN_INT64 ],
+                                      &wholeMatrix[ 2* row *ROW_LEN_INT64 ],
+                                      nCols );
+
+           //update prev: it now points to the last row ever computed
+           prev = row;
+
+           //updates row: goes to the next row to be computed
+           //----------------------------------------------------
+           row = (row + step) & (unsigned int)(nRows-1); //(USE THIS IF nRows IS A POWER OF 2)
+           //row = (row + step) % nRows; //(USE THIS FOR THE "GENERIC" CASE)
+           //----------------------------------------------------
+
+       } while (row != 0);
+   }
+
+   //===================== Wrap-up Phase ===============================//
+   //Absorbs the last block of the memory matrix
+   absorbBlock_2way( state, &wholeMatrix[ 2 * rowa0 *ROW_LEN_INT64],
+                            &wholeMatrix[ 2 * rowa1 *ROW_LEN_INT64] );
+   //Squeezes the key
+   squeeze_2way( state, K, (unsigned int) kLen );
+
+   //================== Freeing the memory =============================//
+   _mm_free(wholeMatrix);
+
+   return 0;
+}
+
+#endif

algo/lyra2/lyra2.c

@@ -327,7 +327,6 @@ int LYRA2REV3( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
 
    reducedDuplexRow1( state, &wholeMatrix[0], &wholeMatrix[ROW_LEN_INT64],
                       nCols);
-
    do
    {
       //M[row] = rand; //M[row*] = M[row*] XOR rotW(rand)

algo/lyra2/lyra2.h

@@ -60,4 +60,20 @@ int LYRA2Z( uint64_t*, void *K, uint64_t kLen, const void *pwd,
 
 int LYRA2(void *K, int64_t kLen, const void *pwd, int32_t pwdlen, const void *salt, int32_t saltlen, int64_t timeCost, const int16_t nRows, const int16_t nCols);
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+int LYRA2RE_2WAY( void *K, uint64_t kLen, const void *pwd, uint64_t pwdlen,
+                  uint64_t timeCost, uint64_t nRows, uint64_t nCols );
+
+int LYRA2REV2_2WAY( uint64_t*, void *K, uint64_t kLen, const void *pwd,
+        uint64_t pwdlen, uint64_t timeCost, uint64_t nRows, uint64_t nCols );
+
+int LYRA2REV3_2WAY( uint64_t*, void *K, uint64_t kLen, const void *pwd,
+        uint64_t pwdlen, uint64_t timeCost, uint64_t nRows, uint64_t nCols );
+
+int LYRA2Z_2WAY( uint64_t*, void *K, uint64_t kLen, const void *pwd,
+          uint64_t pwdlen, uint64_t timeCost, uint64_t nRows, uint64_t nCols );
+
+#endif
+
 #endif /* LYRA2_H_ */

algo/lyra2/lyra2rev2-4way.c

@@ -1,13 +1,150 @@
 #include "lyra2-gate.h"
 #include <memory.h>
-
-#if defined (LYRA2REV2_4WAY)	
-
 #include "algo/blake/blake-hash-4way.h"
 #include "algo/keccak/keccak-hash-4way.h"
 #include "algo/skein/skein-hash-4way.h"
 #include "algo/bmw/bmw-hash-4way.h"
 #include "algo/cubehash/cubehash_sse2.h" 
+#include "algo/cubehash/cube-hash-2way.h"
+
+#if defined (LYRA2REV2_8WAY)
+
+typedef struct {
+   blake256_8way_context     blake;
+   keccak256_8way_context    keccak;
+   cube_4way_context          cube;
+   skein256_8way_context     skein;
+   bmw256_8way_context          bmw;
+} lyra2v2_8way_ctx_holder __attribute__ ((aligned (64)));
+
+static lyra2v2_8way_ctx_holder l2v2_8way_ctx;
+
+bool init_lyra2rev2_8way_ctx()
+{
+   keccak256_8way_init( &l2v2_8way_ctx.keccak );
+   cube_4way_init( &l2v2_8way_ctx.cube, 256, 16, 32 );
+   skein256_8way_init( &l2v2_8way_ctx.skein );
+   bmw256_8way_init( &l2v2_8way_ctx.bmw );
+   return true;
+}
+
+void lyra2rev2_8way_hash( void *state, const void *input )
+{
+   uint32_t vhash[8*8] __attribute__ ((aligned (128)));
+   uint32_t vhashA[8*8] __attribute__ ((aligned (64)));
+   uint32_t vhashB[8*8] __attribute__ ((aligned (64)));
+   uint32_t hash0[8] __attribute__ ((aligned (64)));
+   uint32_t hash1[8] __attribute__ ((aligned (64)));
+   uint32_t hash2[8] __attribute__ ((aligned (64)));
+   uint32_t hash3[8] __attribute__ ((aligned (64)));
+   uint32_t hash4[8] __attribute__ ((aligned (64)));
+   uint32_t hash5[8] __attribute__ ((aligned (64)));
+   uint32_t hash6[8] __attribute__ ((aligned (64)));
+   uint32_t hash7[8] __attribute__ ((aligned (64)));
+   lyra2v2_8way_ctx_holder ctx __attribute__ ((aligned (64)));
+   memcpy( &ctx, &l2v2_8way_ctx, sizeof(l2v2_8way_ctx) );
+
+   blake256_8way( &ctx.blake, input + (64<<3), 16 );
+   blake256_8way_close( &ctx.blake, vhash );
+
+   rintrlv_8x32_8x64( vhashA, vhash, 256 );
+
+   keccak256_8way_update( &ctx.keccak, vhashA, 32 );
+   keccak256_8way_close( &ctx.keccak, vhash );
+
+   rintrlv_8x64_4x128( vhashA, vhashB, vhash, 256 );
+
+   cube_4way_update_close( &ctx.cube, vhashA, vhashA, 32 );
+   cube_4way_init( &ctx.cube, 256, 16, 32 );
+   cube_4way_update_close( &ctx.cube, vhashB, vhashB, 32 );
+
+   dintrlv_4x128( hash0, hash1, hash2, hash3, vhashA, 256 );
+   dintrlv_4x128( hash4, hash5, hash6, hash7, vhashB, 256 );
+
+   intrlv_2x256( vhash, hash0, hash1, 256 );
+   LYRA2REV2_2WAY( l2v2_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash0, hash1, vhash, 256 );
+   intrlv_2x256( vhash, hash2, hash3, 256 );
+   LYRA2REV2_2WAY( l2v2_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash2, hash3, vhash, 256 );
+   intrlv_2x256( vhash, hash4, hash5, 256 );
+   LYRA2REV2_2WAY( l2v2_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash4, hash5, vhash, 256 );
+   intrlv_2x256( vhash, hash6, hash7, 256 );
+   LYRA2REV2_2WAY( l2v2_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash6, hash7, vhash, 256 );
+
+   intrlv_8x64( vhash, hash0, hash1, hash2, hash3, hash4, hash5, hash6,
+                hash7, 256 );
+
+   skein256_8way_update( &ctx.skein, vhash, 32 );
+   skein256_8way_close( &ctx.skein, vhash );
+
+   rintrlv_8x64_4x128( vhashA, vhashB, vhash, 256 );
+
+   cube_4way_init( &ctx.cube, 256, 16, 32 );
+   cube_4way_update_close( &ctx.cube, vhashA, vhashA, 32 );
+   cube_4way_init( &ctx.cube, 256, 16, 32 );
+   cube_4way_update_close( &ctx.cube, vhashB, vhashB, 32 );
+   
+   dintrlv_4x128( hash0, hash1, hash2, hash3, vhashA, 256 );
+   dintrlv_4x128( hash4, hash5, hash6, hash7, vhashB, 256 );
+
+   intrlv_8x32( vhash, hash0, hash1, hash2, hash3, hash4, hash5, hash6, 
+                hash7, 256 );
+
+   bmw256_8way_update( &ctx.bmw, vhash, 32 );
+   bmw256_8way_close( &ctx.bmw, state );
+}
+
+int scanhash_lyra2rev2_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 *hash7 = &(hash[7<<3]);
+   uint32_t lane_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 last_nonce = max_nonce - 8;
+   uint32_t n = first_nonce;
+   const uint32_t Htarg = ptarget[7];
+   __m256i *noncev = (__m256i*)vdata + 19;   // aligned
+   int thr_id = mythr->id; 
+
+   if ( opt_benchmark )
+      ( (uint32_t*)ptarget )[7] = 0x0000ff;
+
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
+
+   blake256_8way_init( &l2v2_8way_ctx.blake );
+   blake256_8way_update( &l2v2_8way_ctx.blake, vdata, 64 );
+
+   do
+   {
+      *noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
+                                                  n+3, n+2, n+1, n ) );
+
+      lyra2rev2_8way_hash( hash, vdata );
+      pdata[19] = n;
+
+      for ( int lane = 0; lane < 8; lane++ ) if ( hash7[lane] <= Htarg )
+      {
+         extr_lane_8x32( 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 < last_nonce) && !work_restart[thr_id].restart);
+   *hashes_done = n - first_nonce;
+   return 0;
+}
+
+#elif defined (LYRA2REV2_4WAY)
 
 typedef struct {
    blake256_4way_context     blake;

algo/lyra2/lyra2rev3-4way.c

@@ -4,9 +4,180 @@
 #include "algo/blake/blake-hash-4way.h"
 #include "algo/bmw/bmw-hash-4way.h"
 #include "algo/cubehash/cubehash_sse2.h" 
+#include "algo/cubehash/cube-hash-2way.h"
+
+#if defined (LYRA2REV3_16WAY)
+
+typedef struct {
+   blake256_16way_context     blake;
+   cube_4way_context          cube;
+   bmw256_16way_context       bmw;
+} lyra2v3_16way_ctx_holder;
+
+static __thread lyra2v3_16way_ctx_holder l2v3_16way_ctx;
+
+bool init_lyra2rev3_16way_ctx()
+{
+   blake256_16way_init( &l2v3_16way_ctx.blake );
+   cube_4way_init( &l2v3_16way_ctx.cube, 256, 16, 32 );
+   bmw256_16way_init( &l2v3_16way_ctx.bmw );
+   return true;
+}
+
+void lyra2rev3_16way_hash( void *state, const void *input )
+{
+   uint32_t vhash[16*8] __attribute__ ((aligned (128)));
+   uint32_t hash0[8] __attribute__ ((aligned (64)));
+   uint32_t hash1[8] __attribute__ ((aligned (64)));
+   uint32_t hash2[8] __attribute__ ((aligned (64)));
+   uint32_t hash3[8] __attribute__ ((aligned (64)));
+   uint32_t hash4[8] __attribute__ ((aligned (64)));
+   uint32_t hash5[8] __attribute__ ((aligned (64)));
+   uint32_t hash6[8] __attribute__ ((aligned (64)));
+   uint32_t hash7[8] __attribute__ ((aligned (64)));
+   uint32_t hash8[8] __attribute__ ((aligned (64)));
+   uint32_t hash9[8] __attribute__ ((aligned (64)));
+   uint32_t hash10[8] __attribute__ ((aligned (64)));
+   uint32_t hash11[8] __attribute__ ((aligned (64)));
+   uint32_t hash12[8] __attribute__ ((aligned (64)));
+   uint32_t hash13[8] __attribute__ ((aligned (64)));
+   uint32_t hash14[8] __attribute__ ((aligned (64)));
+   uint32_t hash15[8] __attribute__ ((aligned (64)));
+   lyra2v3_16way_ctx_holder ctx __attribute__ ((aligned (64)));
+   memcpy( &ctx, &l2v3_16way_ctx, sizeof(l2v3_16way_ctx) );
+
+   blake256_16way_update( &ctx.blake, input + (64*16), 16 );
+   blake256_16way_close( &ctx.blake, vhash );
+
+   dintrlv_16x32( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+           hash8, hash9, hash10, hash11 ,hash12, hash13, hash14, hash15,
+           vhash, 256 );
+
+   intrlv_2x256( vhash, hash0, hash1, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash0, hash1, vhash, 256 );
+   intrlv_2x256( vhash, hash2, hash3, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash2, hash3, vhash, 256 );
+   intrlv_2x256( vhash, hash4, hash5, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash4, hash5, vhash, 256 );
+   intrlv_2x256( vhash, hash6, hash7, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash6, hash7, vhash, 256 );
+   intrlv_2x256( vhash, hash8, hash9, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash8, hash9, vhash, 256 );
+   intrlv_2x256( vhash, hash10, hash11, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash10, hash11, vhash, 256 );
+   intrlv_2x256( vhash, hash12, hash13, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash12, hash13, vhash, 256 );
+   intrlv_2x256( vhash, hash14, hash15, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash14, hash15, vhash, 256 );
+
+   intrlv_4x128( vhash, hash0, hash1, hash2, hash3, 256 );
+   cube_4way_update_close( &ctx.cube, vhash, vhash, 32 );
+   dintrlv_4x128( hash0, hash1, hash2, hash3, vhash, 256 );
+   intrlv_4x128( vhash, hash4, hash5, hash6, hash7, 256 );
+   cube_4way_init( &ctx.cube, 256, 16, 32 );
+   cube_4way_update_close( &ctx.cube, vhash, vhash, 32 );
+   dintrlv_4x128( hash4, hash5, hash6, hash7, vhash, 256 );
+   intrlv_4x128( vhash, hash8, hash9, hash10, hash11, 256 );
+   cube_4way_init( &ctx.cube, 256, 16, 32 );
+   cube_4way_update_close( &ctx.cube, vhash, vhash, 32 );
+   dintrlv_4x128( hash8, hash9, hash10, hash11, vhash, 256 );
+   intrlv_4x128( vhash, hash12, hash13, hash14, hash15, 256 );
+   cube_4way_init( &ctx.cube, 256, 16, 32 );
+   cube_4way_update_close( &ctx.cube, vhash, vhash, 32 );
+   dintrlv_4x128( hash12, hash13, hash14, hash15, vhash, 256 );
+
+   intrlv_2x256( vhash, hash0, hash1, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash0, hash1, vhash, 256 );
+   intrlv_2x256( vhash, hash2, hash3, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash2, hash3, vhash, 256 );
+   intrlv_2x256( vhash, hash4, hash5, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash4, hash5, vhash, 256 );
+   intrlv_2x256( vhash, hash6, hash7, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash6, hash7, vhash, 256 );
+   intrlv_2x256( vhash, hash8, hash9, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash8, hash9, vhash, 256 );
+   intrlv_2x256( vhash, hash10, hash11, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash10, hash11, vhash, 256 );
+   intrlv_2x256( vhash, hash12, hash13, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash12, hash13, vhash, 256 );
+   intrlv_2x256( vhash, hash14, hash15, 256 );
+   LYRA2REV3_2WAY( l2v3_wholeMatrix, vhash, 32, vhash, 32, 1, 4, 4 );
+   dintrlv_2x256( hash14, hash15, vhash, 256 );
+
+   intrlv_16x32( vhash, hash0, hash1, hash2, hash3, hash4, hash5, hash6,
+             hash7, hash8, hash9, hash10, hash11, hash12, hash13, hash14,
+             hash15, 256 );
+
+   bmw256_16way_update( &ctx.bmw, vhash, 32 );
+   bmw256_16way_close( &ctx.bmw, state );
+}
 
 
-#if defined (LYRA2REV3_8WAY)
+int scanhash_lyra2rev3_16way( struct work *work, const uint32_t max_nonce,
+                             uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t hash[8*16] __attribute__ ((aligned (128)));
+   uint32_t vdata[20*16] __attribute__ ((aligned (64)));
+   uint32_t *hash7 = &hash[7<<4];
+   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
+   uint32_t *pdata = work->data;
+   const uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   uint32_t n = first_nonce;
+   const uint32_t last_nonce = max_nonce - 16;
+   const uint32_t Htarg = ptarget[7];
+   __m512i  *noncev = (__m512i*)vdata + 19;   // aligned
+   const int thr_id = mythr->id;
+
+   if ( opt_benchmark )  ( (uint32_t*)ptarget )[7] = 0x0000ff;
+
+   mm512_bswap32_intrlv80_16x32( vdata, pdata );
+
+   blake256_16way_init( &l2v3_16way_ctx.blake );
+   blake256_16way_update( &l2v3_16way_ctx.blake, 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 ) );
+
+      lyra2rev3_16way_hash( hash, vdata );
+      pdata[19] = n;
+
+      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 ( likely( (n < last_nonce) && !work_restart[thr_id].restart ) );
+   *hashes_done = n - first_nonce;
+   return 0;
+}
+
+#elif defined (LYRA2REV3_8WAY)
 
 typedef struct {
    blake256_8way_context     blake;
@@ -14,7 +185,7 @@ 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 +209,7 @@ 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 +262,7 @@ 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 +270,15 @@ 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 +287,18 @@ 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 +307,14 @@ 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 +334,8 @@ 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 +381,10 @@ 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 ) );

algo/lyra2/lyra2z-4way.c

@@ -1,13 +1,240 @@
 #include "lyra2-gate.h"
-
-#ifdef LYRA2Z_4WAY
-
 #include <memory.h>
 #include <mm_malloc.h>
 #include "lyra2.h"
 #include "algo/blake/sph_blake.h"
 #include "algo/blake/blake-hash-4way.h"
 
+#if defined(LYRA2Z_16WAY)
+
+__thread uint64_t* lyra2z_16way_matrix;
+
+bool lyra2z_16way_thread_init()
+{
+ return ( lyra2z_16way_matrix = _mm_malloc( 2*LYRA2Z_MATRIX_SIZE, 64 ) );
+}
+
+static __thread blake256_16way_context l2z_16way_blake_mid;
+
+void lyra2z_16way_midstate( const void* input )
+{
+       blake256_16way_init( &l2z_16way_blake_mid );
+       blake256_16way_update( &l2z_16way_blake_mid, input, 64 );
+}
+
+void lyra2z_16way_hash( void *state, const void *input )
+{
+    uint32_t vhash[8*16] __attribute__ ((aligned (128)));
+    uint32_t hash0[8] __attribute__ ((aligned (64)));
+    uint32_t hash1[8] __attribute__ ((aligned (64)));
+    uint32_t hash2[8] __attribute__ ((aligned (64)));
+    uint32_t hash3[8] __attribute__ ((aligned (64)));
+    uint32_t hash4[8] __attribute__ ((aligned (64)));
+    uint32_t hash5[8] __attribute__ ((aligned (64)));
+    uint32_t hash6[8] __attribute__ ((aligned (64)));
+    uint32_t hash7[8] __attribute__ ((aligned (64)));
+    uint32_t hash8[8] __attribute__ ((aligned (64)));
+    uint32_t hash9[8] __attribute__ ((aligned (64)));
+    uint32_t hash10[8] __attribute__ ((aligned (64)));
+    uint32_t hash11[8] __attribute__ ((aligned (64)));
+    uint32_t hash12[8] __attribute__ ((aligned (64)));
+    uint32_t hash13[8] __attribute__ ((aligned (64)));
+    uint32_t hash14[8] __attribute__ ((aligned (64)));
+    uint32_t hash15[8] __attribute__ ((aligned (64)));
+    blake256_16way_context ctx_blake __attribute__ ((aligned (64)));
+
+    memcpy( &ctx_blake, &l2z_16way_blake_mid, sizeof l2z_16way_blake_mid );
+    blake256_16way_update( &ctx_blake, input + (64*16), 16 );
+    blake256_16way_close( &ctx_blake, vhash );
+
+    dintrlv_16x32( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+              hash8, hash9, hash10, hash11 ,hash12, hash13, hash14, hash15,
+               vhash, 256 );
+
+    intrlv_2x256( vhash, hash0, hash1, 256 );
+    LYRA2Z_2WAY( lyra2z_16way_matrix, vhash, 32, vhash, 32, 8, 8, 8 );
+    dintrlv_2x256( hash0, hash1, vhash, 256 );
+    intrlv_2x256( vhash, hash2, hash3, 256 );
+    LYRA2Z_2WAY( lyra2z_16way_matrix, vhash, 32, vhash, 32, 8, 8, 8 );
+    dintrlv_2x256( hash2, hash3, vhash, 256 );
+    intrlv_2x256( vhash, hash4, hash5, 256 );
+    LYRA2Z_2WAY( lyra2z_16way_matrix, vhash, 32, vhash, 32, 8, 8, 8 );
+    dintrlv_2x256( hash4, hash5, vhash, 256 );
+    intrlv_2x256( vhash, hash6, hash7, 256 );
+    LYRA2Z_2WAY( lyra2z_16way_matrix, vhash, 32, vhash, 32, 8, 8, 8 );
+    dintrlv_2x256( hash6, hash7, vhash, 256 );
+    intrlv_2x256( vhash, hash8, hash9, 256 );
+    LYRA2Z_2WAY( lyra2z_16way_matrix, vhash, 32, vhash, 32, 8, 8, 8 );
+    dintrlv_2x256( hash8, hash9, vhash, 256 );
+    intrlv_2x256( vhash, hash10, hash11, 256 );
+    LYRA2Z_2WAY( lyra2z_16way_matrix, vhash, 32, vhash, 32, 8, 8, 8 );
+    dintrlv_2x256( hash10, hash11, vhash, 256 );
+    intrlv_2x256( vhash, hash12, hash13, 256 );
+    LYRA2Z_2WAY( lyra2z_16way_matrix, vhash, 32, vhash, 32, 8, 8, 8 );
+    dintrlv_2x256( hash12, hash13, vhash, 256 );
+    intrlv_2x256( vhash, hash14, hash15, 256 );
+    LYRA2Z_2WAY( lyra2z_16way_matrix, vhash, 32, vhash, 32, 8, 8, 8 );
+    dintrlv_2x256( hash14, hash15, vhash, 256 );
+   
+    memcpy( state,     hash0, 32 );
+    memcpy( state+ 32, hash1, 32 );
+    memcpy( state+ 64, hash2, 32 );
+    memcpy( state+ 96, hash3, 32 );
+    memcpy( state+128, hash4, 32 );
+    memcpy( state+160, hash5, 32 );
+    memcpy( state+192, hash6, 32 );
+    memcpy( state+224, hash7, 32 );
+    memcpy( state+256, hash8, 32 );
+    memcpy( state+288, hash9, 32 );
+    memcpy( state+320, hash10, 32 );
+    memcpy( state+352, hash11, 32 );
+    memcpy( state+384, hash12, 32 );
+    memcpy( state+416, hash13, 32 );
+    memcpy( state+448, hash14, 32 );
+    memcpy( state+480, hash15, 32 );
+}
+
+int scanhash_lyra2z_16way( struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t hash[8*16] __attribute__ ((aligned (128)));
+   uint32_t vdata[20*16] __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;
+   __m512i  *noncev = (__m512i*)vdata + 19;   // aligned
+   int thr_id = mythr->id;  // thr_id arg is deprecated
+
+   if ( opt_benchmark )
+      ptarget[7] = 0x0000ff;
+
+   mm512_bswap32_intrlv80_16x32( vdata, pdata );
+   lyra2z_16way_midstate( vdata );
+
+   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 ) );
+      lyra2z_16way_hash( hash, vdata );
+      pdata[19] = n;
+
+      for ( int i = 0; i < 16; i++ )
+      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget )
+           && !opt_benchmark )
+      {
+          pdata[19] = n+i;
+          submit_lane_solution( work, hash+(i<<3), mythr, i );
+      }
+      n += 16;
+   } while ( (n < max_nonce-16) && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+#elif defined(LYRA2Z_8WAY)
+
+__thread uint64_t* lyra2z_8way_matrix;
+
+bool lyra2z_8way_thread_init()
+{
+ return ( lyra2z_8way_matrix = _mm_malloc( LYRA2Z_MATRIX_SIZE, 64 ) );
+}
+
+static __thread blake256_8way_context l2z_8way_blake_mid;
+
+void lyra2z_8way_midstate( const void* input )
+{
+       blake256_8way_init( &l2z_8way_blake_mid );
+       blake256_8way( &l2z_8way_blake_mid, input, 64 );
+}
+
+void lyra2z_8way_hash( void *state, const void *input )
+{
+     uint32_t hash0[8] __attribute__ ((aligned (64)));
+     uint32_t hash1[8] __attribute__ ((aligned (64)));
+     uint32_t hash2[8] __attribute__ ((aligned (64)));
+     uint32_t hash3[8] __attribute__ ((aligned (64)));
+     uint32_t hash4[8] __attribute__ ((aligned (64)));
+     uint32_t hash5[8] __attribute__ ((aligned (64)));
+     uint32_t hash6[8] __attribute__ ((aligned (64)));
+     uint32_t hash7[8] __attribute__ ((aligned (64)));
+     uint32_t vhash[8*8] __attribute__ ((aligned (64)));
+     blake256_8way_context ctx_blake __attribute__ ((aligned (64)));
+
+     memcpy( &ctx_blake, &l2z_8way_blake_mid, sizeof l2z_8way_blake_mid );
+     blake256_8way( &ctx_blake, input + (64*8), 16 );
+     blake256_8way_close( &ctx_blake, vhash );
+
+     dintrlv_8x32( hash0, hash1, hash2, hash3,
+                   hash4, hash5, hash6, hash7, vhash, 256 );
+
+     LYRA2Z( lyra2z_8way_matrix, hash0, 32, hash0, 32, hash0, 32, 8, 8, 8 );
+     LYRA2Z( lyra2z_8way_matrix, hash1, 32, hash1, 32, hash1, 32, 8, 8, 8 );
+     LYRA2Z( lyra2z_8way_matrix, hash2, 32, hash2, 32, hash2, 32, 8, 8, 8 );
+     LYRA2Z( lyra2z_8way_matrix, hash3, 32, hash3, 32, hash3, 32, 8, 8, 8 );
+     LYRA2Z( lyra2z_8way_matrix, hash4, 32, hash4, 32, hash4, 32, 8, 8, 8 );
+     LYRA2Z( lyra2z_8way_matrix, hash5, 32, hash5, 32, hash5, 32, 8, 8, 8 );
+     LYRA2Z( lyra2z_8way_matrix, hash6, 32, hash6, 32, hash6, 32, 8, 8, 8 );
+     LYRA2Z( lyra2z_8way_matrix, hash7, 32, hash7, 32, hash7, 32, 8, 8, 8 );
+
+
+     memcpy( state,     hash0, 32 );
+     memcpy( state+ 32, hash1, 32 );
+     memcpy( state+ 64, hash2, 32 );
+     memcpy( state+ 96, hash3, 32 );
+     memcpy( state+128, hash4, 32 );
+     memcpy( state+160, hash5, 32 );
+     memcpy( state+192, hash6, 32 );
+     memcpy( state+224, hash7, 32 );
+}
+
+int scanhash_lyra2z_8way( struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t hash[8*8] __attribute__ ((aligned (64)));
+   uint32_t vdata[20*8] __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;
+   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
+   int thr_id = mythr->id;  // thr_id arg is deprecated
+
+   if ( opt_benchmark )
+      ptarget[7] = 0x0000ff;
+
+   mm256_bswap32_intrlv80_8x32( vdata, pdata );
+   lyra2z_8way_midstate( vdata );
+
+   do {
+      *noncev = mm256_bswap_32(
+                 _mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n ) );
+      lyra2z_8way_hash( hash, vdata );
+      pdata[19] = n;
+
+      for ( int i = 0; i < 8; i++ )
+      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget )
+           && !opt_benchmark )
+      {
+          pdata[19] = n+i;
+          submit_lane_solution( work, hash+(i<<3), mythr, i );
+      }
+      n += 8;
+   } while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+
+#elif defined(LYRA2Z_4WAY)
+
+
 __thread uint64_t* lyra2z_4way_matrix;
 
 bool lyra2z_4way_thread_init()
@@ -85,100 +312,3 @@ int scanhash_lyra2z_4way( struct work *work, uint32_t max_nonce,
 
 #endif
 
-#if defined(LYRA2Z_8WAY)
-
-__thread uint64_t* lyra2z_8way_matrix;
-
-bool lyra2z_8way_thread_init()
-{
- return ( lyra2z_8way_matrix = _mm_malloc( LYRA2Z_MATRIX_SIZE, 64 ) );
-}
-
-static __thread blake256_8way_context l2z_8way_blake_mid;
-
-void lyra2z_8way_midstate( const void* input )
-{
-       blake256_8way_init( &l2z_8way_blake_mid );
-       blake256_8way( &l2z_8way_blake_mid, input, 64 );
-}
-
-void lyra2z_8way_hash( void *state, const void *input )
-{
-     uint32_t hash0[8] __attribute__ ((aligned (64)));
-     uint32_t hash1[8] __attribute__ ((aligned (64)));
-     uint32_t hash2[8] __attribute__ ((aligned (64)));
-     uint32_t hash3[8] __attribute__ ((aligned (64)));
-     uint32_t hash4[8] __attribute__ ((aligned (64)));
-     uint32_t hash5[8] __attribute__ ((aligned (64)));
-     uint32_t hash6[8] __attribute__ ((aligned (64)));
-     uint32_t hash7[8] __attribute__ ((aligned (64)));
-     uint32_t vhash[8*8] __attribute__ ((aligned (64)));
-     blake256_8way_context ctx_blake __attribute__ ((aligned (64)));
-
-     memcpy( &ctx_blake, &l2z_8way_blake_mid, sizeof l2z_8way_blake_mid );
-     blake256_8way( &ctx_blake, input + (64*8), 16 );
-     blake256_8way_close( &ctx_blake, vhash );
-
-     dintrlv_8x32( hash0, hash1, hash2, hash3,
-                   hash4, hash5, hash6, hash7, vhash, 256 );
-
-     LYRA2Z( lyra2z_8way_matrix, hash0, 32, hash0, 32, hash0, 32, 8, 8, 8 );
-     LYRA2Z( lyra2z_8way_matrix, hash1, 32, hash1, 32, hash1, 32, 8, 8, 8 );
-     LYRA2Z( lyra2z_8way_matrix, hash2, 32, hash2, 32, hash2, 32, 8, 8, 8 );
-     LYRA2Z( lyra2z_8way_matrix, hash3, 32, hash3, 32, hash3, 32, 8, 8, 8 );
-     LYRA2Z( lyra2z_8way_matrix, hash4, 32, hash4, 32, hash4, 32, 8, 8, 8 );
-     LYRA2Z( lyra2z_8way_matrix, hash5, 32, hash5, 32, hash5, 32, 8, 8, 8 );
-     LYRA2Z( lyra2z_8way_matrix, hash6, 32, hash6, 32, hash6, 32, 8, 8, 8 );
-     LYRA2Z( lyra2z_8way_matrix, hash7, 32, hash7, 32, hash7, 32, 8, 8, 8 );
-
-     memcpy( state,     hash0, 32 );
-     memcpy( state+ 32, hash1, 32 );
-     memcpy( state+ 64, hash2, 32 );
-     memcpy( state+ 96, hash3, 32 );
-     memcpy( state+128, hash4, 32 );
-     memcpy( state+160, hash5, 32 );
-     memcpy( state+192, hash6, 32 );
-     memcpy( state+224, hash7, 32 );
-}
-
-int scanhash_lyra2z_8way( struct work *work, uint32_t max_nonce,
-                          uint64_t *hashes_done, struct thr_info *mythr )
-{
-   uint32_t hash[8*8] __attribute__ ((aligned (64)));
-   uint32_t vdata[20*8] __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;
-   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
-   int thr_id = mythr->id;  // thr_id arg is deprecated
-
-   if ( opt_benchmark )
-      ptarget[7] = 0x0000ff;
-
-   mm256_bswap32_intrlv80_8x32( vdata, pdata );
-   lyra2z_8way_midstate( vdata );
-
-   do {
-      *noncev = mm256_bswap_32(
-                 _mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n ) );
-      lyra2z_8way_hash( hash, vdata );
-      pdata[19] = n;
-
-      for ( int i = 0; i < 8; i++ )
-      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget )
-           && !opt_benchmark )
-      {
-          pdata[19] = n+i;         
-          submit_lane_solution( work, hash+(i<<3), mythr, i );
-      }
-      n += 8;
-   } while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
-
-   *hashes_done = n - first_nonce + 1;
-   return 0;
-}
-
-
-#endif

algo/lyra2/sponge-2way.c

@@ -0,0 +1,357 @@
+/**
+ * A simple implementation of Blake2b's internal permutation
+ * in the form of a sponge.
+ *
+ * Author: The Lyra PHC team (http://www.lyra-kdf.net/) -- 2014.
+ *
+ * This software is hereby placed in the public domain.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ''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 AUTHORS 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.
+ */
+
+//#include "algo-gate.h"
+#include <string.h>
+#include <stdio.h>
+#include <time.h>
+#include <immintrin.h>
+#include "sponge.h"
+#include "lyra2.h"
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+inline void squeeze_2way( uint64_t *State, byte *Out, unsigned int len )
+{
+    const int len_m256i = len / 32;
+    const int fullBlocks = len_m256i / BLOCK_LEN_M256I;
+    __m512i* state = (__m512i*)State;
+    __m512i* out   = (__m512i*)Out;
+    int i;
+
+    //Squeezes full blocks
+    for ( i = 0; i < fullBlocks; i++ )
+    {
+       memcpy_512( out, state, BLOCK_LEN_M256I );
+       LYRA_ROUND_2WAY_AVX512( state[0], state[1], state[2], state[3] );
+       out += BLOCK_LEN_M256I;
+    }
+    //Squeezes remaining bytes
+    memcpy_512( out, state, len_m256i % BLOCK_LEN_M256I );
+}
+
+inline void absorbBlock_2way( uint64_t *State, const uint64_t *In0,
+                                               const uint64_t *In1 ) 
+{
+    register __m512i state0, state1, state2, state3;
+    __m512i in[3];
+    casti_m256i( in, 0 ) = casti_m256i( In0, 0 );
+    casti_m256i( in, 1 ) = casti_m256i( In1, 1 );
+    casti_m256i( in, 2 ) = casti_m256i( In0, 2 );
+    casti_m256i( in, 3 ) = casti_m256i( In1, 3 );
+    casti_m256i( in, 4 ) = casti_m256i( In0, 4 );
+    casti_m256i( in, 5 ) = casti_m256i( In1, 5 );
+    
+    state0 = _mm512_load_si512( (__m512i*)State     );
+    state1 = _mm512_load_si512( (__m512i*)State + 1 );
+    state2 = _mm512_load_si512( (__m512i*)State + 2 );
+    state3 = _mm512_load_si512( (__m512i*)State + 3 );
+
+    state0 = _mm512_xor_si512( state0, in[0] );
+    state1 = _mm512_xor_si512( state1, in[1] );
+    state2 = _mm512_xor_si512( state2, in[2] );
+
+    LYRA_12_ROUNDS_2WAY_AVX512( state0, state1, state2, state3 );
+
+    _mm512_store_si512( (__m512i*)State,     state0 );
+    _mm512_store_si512( (__m512i*)State + 1, state1 );
+    _mm512_store_si512( (__m512i*)State + 2, state2 );
+    _mm512_store_si512( (__m512i*)State + 3, state3 );
+
+}
+
+inline void absorbBlockBlake2Safe_2way( uint64_t *State, const uint64_t *In,
+                      const uint64_t nBlocks, const uint64_t block_len )
+{
+  register __m512i state0, state1, state2, state3;
+
+  state0 = 
+  state1 = m512_zero;
+  state2 = m512_const4_64( 0xa54ff53a5f1d36f1ULL, 0x3c6ef372fe94f82bULL,
+                           0xbb67ae8584caa73bULL, 0x6a09e667f3bcc908ULL );
+  state3 = m512_const4_64( 0x5be0cd19137e2179ULL, 0x1f83d9abfb41bd6bULL,
+                           0x9b05688c2b3e6c1fULL, 0x510e527fade682d1ULL );
+
+  for ( int i = 0; i < nBlocks; i++ )
+  { 
+    __m512i *in = (__m512i*)In;
+    state0 = _mm512_xor_si512( state0, in[0] );
+    state1 = _mm512_xor_si512( state1, in[1] );
+
+    LYRA_12_ROUNDS_2WAY_AVX512( state0, state1, state2, state3 );
+    In += block_len*2;
+  }
+
+  _mm512_store_si512( (__m512i*)State,     state0 );
+  _mm512_store_si512( (__m512i*)State + 1, state1 );
+  _mm512_store_si512( (__m512i*)State + 2, state2 );
+  _mm512_store_si512( (__m512i*)State + 3, state3 );
+
+}
+
+inline void reducedSqueezeRow0_2way( uint64_t* State, uint64_t* rowOut,
+                                     uint64_t nCols )
+{
+    int i;
+
+    //M[row][C-1-col] = H.reduced_squeeze()
+
+
+    register __m512i state0, state1, state2, state3;
+    __m512i* out   = (__m512i*)rowOut + ( (nCols-1) * BLOCK_LEN_M256I );
+
+    state0 = _mm512_load_si512( (__m512i*)State     );
+    state1 = _mm512_load_si512( (__m512i*)State + 1 );
+    state2 = _mm512_load_si512( (__m512i*)State + 2 );
+    state3 = _mm512_load_si512( (__m512i*)State + 3 );
+
+    for ( i = 0; i < 9; i += 3)
+    {
+        _mm_prefetch( out - i,     _MM_HINT_T0 );
+        _mm_prefetch( out - i - 2, _MM_HINT_T0 );
+    }
+
+    for ( i = 0; i < nCols; i++ )
+    {
+       _mm_prefetch( out -  9, _MM_HINT_T0 );
+       _mm_prefetch( out - 11, _MM_HINT_T0 );
+
+       out[0] = state0;
+       out[1] = state1;
+       out[2] = state2;
+
+       //Goes to next block (column) that will receive the squeezed data
+       out -= BLOCK_LEN_M256I;
+
+       LYRA_ROUND_2WAY_AVX512( state0, state1, state2, state3 );
+    }
+
+    _mm512_store_si512( (__m512i*)State,     state0 );
+    _mm512_store_si512( (__m512i*)State + 1, state1 );
+    _mm512_store_si512( (__m512i*)State + 2, state2 );
+    _mm512_store_si512( (__m512i*)State + 3, state3 );
+}
+
+
+inline void reducedDuplexRow1_2way( uint64_t *State, uint64_t *rowIn,
+                 uint64_t *rowOut, uint64_t nCols )
+{
+    int i;
+    register __m512i state0, state1, state2, state3;
+    __m512i *in = (__m512i*)rowIn;
+    __m512i *out = (__m512i*)rowOut + ( (nCols-1) * BLOCK_LEN_M256I );
+
+    state0 = _mm512_load_si512( (__m512i*)State     );
+    state1 = _mm512_load_si512( (__m512i*)State + 1 );
+    state2 = _mm512_load_si512( (__m512i*)State + 2 );
+    state3 = _mm512_load_si512( (__m512i*)State + 3 );
+
+    for ( i = 0; i < nCols; i++ )
+    {
+         state0 = _mm512_xor_si512( state0, in[0] );
+         state1 = _mm512_xor_si512( state1, in[1] );
+         state2 = _mm512_xor_si512( state2, in[2] );
+
+         LYRA_ROUND_2WAY_AVX512( state0, state1, state2, state3 );
+
+         out[0] = _mm512_xor_si512( state0, in[0] );
+         out[1] = _mm512_xor_si512( state1, in[1] );
+         out[2] = _mm512_xor_si512( state2, in[2] );
+
+         //Input: next column (i.e., next block in sequence)
+         in += BLOCK_LEN_M256I;
+         //Output: goes to previous column
+         out -= BLOCK_LEN_M256I;
+    }
+
+    _mm512_store_si512( (__m512i*)State,     state0 );
+    _mm512_store_si512( (__m512i*)State + 1, state1 );
+    _mm512_store_si512( (__m512i*)State + 2, state2 );
+    _mm512_store_si512( (__m512i*)State + 3, state3 );
+}
+
+inline void reducedDuplexRowSetup_2way( uint64_t *State, uint64_t *rowIn,
+                       uint64_t *rowInOut, uint64_t *rowOut, uint64_t nCols )
+{
+    int i;
+    register __m512i state0, state1, state2, state3;
+    __m512i* in    = (__m512i*)rowIn;
+    __m512i* inout = (__m512i*)rowInOut;
+    __m512i* out   = (__m512i*)rowOut + ( (nCols-1) * BLOCK_LEN_M256I );
+    __m512i  t0, t1, t2;
+
+    state0 = _mm512_load_si512( (__m512i*)State     );
+    state1 = _mm512_load_si512( (__m512i*)State + 1 );
+    state2 = _mm512_load_si512( (__m512i*)State + 2 );
+    state3 = _mm512_load_si512( (__m512i*)State + 3 );
+
+    for ( i = 0; i < nCols; i++ )
+    {
+       state0 = _mm512_xor_si512( state0,
+                                  _mm512_add_epi64( in[0], inout[0] ) );
+       state1 = _mm512_xor_si512( state1,
+                                  _mm512_add_epi64( in[1], inout[1] ) );
+       state2 = _mm512_xor_si512( state2,
+                                  _mm512_add_epi64( in[2], inout[2] ) );
+
+       LYRA_ROUND_2WAY_AVX512( state0, state1, state2, state3 );
+
+       out[0] = _mm512_xor_si512( state0, in[0] );
+       out[1] = _mm512_xor_si512( state1, in[1] );
+       out[2] = _mm512_xor_si512( state2, in[2] );
+
+       //M[row*][col] = M[row*][col] XOR rotW(rand)
+       t0 = _mm512_permutex_epi64( state0, 0x93 );
+       t1 = _mm512_permutex_epi64( state1, 0x93 );
+       t2 = _mm512_permutex_epi64( state2, 0x93 );
+
+       inout[0] = _mm512_xor_si512( inout[0],
+                                 _mm512_mask_blend_epi32( 0x0303, t0, t2 ) );
+       inout[1] = _mm512_xor_si512( inout[1],
+                                 _mm512_mask_blend_epi32( 0x0303, t1, t0 ) );
+       inout[2] = _mm512_xor_si512( inout[2],
+                                 _mm512_mask_blend_epi32( 0x0303, t2, t1 ) );
+
+
+       //Inputs: next column (i.e., next block in sequence)
+       in    += BLOCK_LEN_M256I;
+       inout += BLOCK_LEN_M256I;
+       //Output: goes to previous column
+       out   -= BLOCK_LEN_M256I;
+    }
+
+    _mm512_store_si512( (__m512i*)State,     state0 );
+    _mm512_store_si512( (__m512i*)State + 1, state1 );
+    _mm512_store_si512( (__m512i*)State + 2, state2 );
+    _mm512_store_si512( (__m512i*)State + 3, state3 );
+}
+
+// big ugly workaound for pointer aliasing, use a union of pointers.
+// Access matrix using m512i for in and out, m256i for inout
+
+inline void reducedDuplexRow_2way( uint64_t *State, uint64_t *rowIn,
+                            uint64_t *rowInOut0, uint64_t *rowInOut1,
+                            uint64_t *rowOut, uint64_t nCols)
+{
+   int i;
+   register __m512i state0, state1, state2, state3;
+   __m512i *in = (__m512i*)rowIn;
+   __m256i *inout0 = (__m256i*)rowInOut0;
+   __m256i *inout1 = (__m256i*)rowInOut1;
+   __m512i *out = (__m512i*)rowOut;
+   __m512i io[3];
+   povly inout;
+   inout.v512 = &io[0];
+    __m512i t0, t1, t2;
+
+   state0 = _mm512_load_si512( (__m512i*)State     );
+   state1 = _mm512_load_si512( (__m512i*)State + 1 );
+   state2 = _mm512_load_si512( (__m512i*)State + 2 );
+   state3 = _mm512_load_si512( (__m512i*)State + 3 );
+    
+    _mm_prefetch( in,     _MM_HINT_T0 );
+    _mm_prefetch( inout0,     _MM_HINT_T0 );
+    _mm_prefetch( inout1,     _MM_HINT_T0 );
+    _mm_prefetch( in     + 2, _MM_HINT_T0 );
+    _mm_prefetch( inout0 + 2, _MM_HINT_T0 );
+    _mm_prefetch( inout1 + 2, _MM_HINT_T0 );
+    _mm_prefetch( in     + 4, _MM_HINT_T0 );
+    _mm_prefetch( inout0 + 4, _MM_HINT_T0 );
+    _mm_prefetch( inout1 + 4, _MM_HINT_T0 );
+    _mm_prefetch( in     + 6, _MM_HINT_T0 );
+    _mm_prefetch( inout0 + 6, _MM_HINT_T0 );
+    _mm_prefetch( inout1 + 6, _MM_HINT_T0 );
+
+    
+    for ( i = 0; i < nCols; i++ )
+    {
+
+      //Absorbing "M[prev] [+] M[row*]"
+      inout.v256[0] = inout0[0];
+      inout.v256[1] = inout1[1];
+      inout.v256[2] = inout0[2];
+      inout.v256[3] = inout1[3];
+      inout.v256[4] = inout0[4];
+      inout.v256[5] = inout1[5];
+
+      state0 = _mm512_xor_si512( state0,
+                                 _mm512_add_epi64( in[0], inout.v512[0] ) );
+      state1 = _mm512_xor_si512( state1,
+                                 _mm512_add_epi64( in[1], inout.v512[1] ) );
+      state2 = _mm512_xor_si512( state2,
+                                 _mm512_add_epi64( in[2], inout.v512[2] ) );
+
+
+      //Applies the reduced-round transformation f to the sponge's state
+      LYRA_ROUND_2WAY_AVX512( state0, state1, state2, state3 );
+
+      //M[rowOut][col] = M[rowOut][col] XOR rand
+      out[0] = _mm512_xor_si512( out[0], state0 );
+      out[1] = _mm512_xor_si512( out[1], state1 );
+      out[2] = _mm512_xor_si512( out[2], state2 );
+
+      // if inout is the same row as out it was just overwritten, reload.
+      if ( rowOut == rowInOut0 )
+      {
+         inout.v256[0] = inout0[0];
+         inout.v256[2] = inout0[2];
+         inout.v256[4] = inout0[4];
+      }
+      if ( rowOut == rowInOut1 )
+      {
+         inout.v256[1] = inout1[1];
+         inout.v256[3] = inout1[3];
+         inout.v256[5] = inout1[5];
+      }
+
+      //M[rowInOut][col] = M[rowInOut][col] XOR rotW(rand)
+      t0 = _mm512_permutex_epi64( state0, 0x93 );
+      t1 = _mm512_permutex_epi64( state1, 0x93 );
+      t2 = _mm512_permutex_epi64( state2, 0x93 );
+
+      inout.v512[0] = _mm512_xor_si512( inout.v512[0],
+                                   _mm512_mask_blend_epi32( 0x0303, t0, t2 ) );
+      inout.v512[1] = _mm512_xor_si512( inout.v512[1],
+                                   _mm512_mask_blend_epi32( 0x0303, t1, t0 ) );
+      inout.v512[2] = _mm512_xor_si512( inout.v512[2],
+                                   _mm512_mask_blend_epi32( 0x0303, t2, t1 ) );
+      
+      inout0[0] = inout.v256[0];
+      inout1[1] = inout.v256[1];
+      inout0[2] = inout.v256[2];
+      inout1[3] = inout.v256[3];
+      inout0[4] = inout.v256[4];
+      inout1[5] = inout.v256[5];
+
+       //Goes to next block
+       in     += BLOCK_LEN_M256I;
+       inout0 += BLOCK_LEN_M256I * 2;
+       inout1 += BLOCK_LEN_M256I * 2;
+       out    += BLOCK_LEN_M256I;
+   }
+
+   _mm512_store_si512( (__m512i*)State,     state0 );
+   _mm512_store_si512( (__m512i*)State + 1, state1 );
+   _mm512_store_si512( (__m512i*)State + 2, state2 );
+   _mm512_store_si512( (__m512i*)State + 3, state3 );
+}
+
+#endif // AVX512

algo/lyra2/sponge.c

@@ -375,7 +375,10 @@ inline void reducedSqueezeRow0( uint64_t* State, uint64_t* rowOut,
     {
        _mm_prefetch( out -  9, _MM_HINT_T0 );
        _mm_prefetch( out - 11, _MM_HINT_T0 );
-                   
+
+//printf("S RSR0 col= %d, out= %x\n",i,out);
+
+
        out[0] = state0;
        out[1] = state1;
        out[2] = state2;
@@ -706,11 +709,34 @@ inline void reducedDuplexRowSetup( uint64_t *State, uint64_t *rowIn,
        out[1] = _mm256_xor_si256( state1, in[1] );
        out[2] = _mm256_xor_si256( state2, in[2] );
 
+/*
+printf("s duplexsetup col= %d\n",i); 
+uint64_t * o = (uint64_t*)out;
+printf("S out %016lx %016lx %016lx %016lx\n",o[0],o[1],o[2],o[3]);
+printf("S out %016lx %016lx %016lx %016lx\n",o[4],o[5],o[6],o[7]);
+printf("S out %016lx %016lx %016lx %016lx\n",o[8],o[9],o[10],o[11]);
+printf("S out %016lx %016lx %016lx %016lx\n",o[12],o[13],o[14],o[15]);
+printf("S out %016lx %016lx %016lx %016lx\n",o[16],o[17],o[18],o[19]);
+printf("S out %016lx %016lx %016lx %016lx\n",o[20],o[21],o[22],o[23]);
+*/
+
        //M[row*][col] = M[row*][col] XOR rotW(rand)
        t0 = _mm256_permute4x64_epi64( state0, 0x93 );
        t1 = _mm256_permute4x64_epi64( state1, 0x93 );
        t2 = _mm256_permute4x64_epi64( state2, 0x93 );
 
+/*
+uint64_t *t = (uint64_t*)&t0;
+printf("S t0 %016lx %016lx %016lx %016lx\n",t[0],t[1],t[2],t[3]);
+
+o = (uint64_t*)inout;
+printf("S inout0 %016lx %016lx %016lx %016lx\n",o[0],o[1],o[2],o[3]);
+printf("S inout0 %016lx %016lx %016lx %016lx\n",o[4],o[5],o[6],o[7]);
+printf("S inout0 %016lx %016lx %016lx %016lx\n",o[8],o[9],o[10],o[11]);
+printf("S inout0 %016lx %016lx %016lx %016lx\n",o[12],o[13],o[14],o[15]);
+printf("S inout0 %016lx %016lx %016lx %016lx\n",o[16],o[17],o[18],o[19]);
+printf("S inout0 %016lx %016lx %016lx %016lx\n",o[20],o[21],o[22],o[23]);
+*/       
        inout[0] = _mm256_xor_si256( inout[0],
                                     _mm256_blend_epi32( t0, t2, 0x03 ) );
        inout[1] = _mm256_xor_si256( inout[1],
@@ -718,7 +744,17 @@ inline void reducedDuplexRowSetup( uint64_t *State, uint64_t *rowIn,
        inout[2] = _mm256_xor_si256( inout[2],
                                     _mm256_blend_epi32( t2, t1, 0x03 ) );
 
-       //Inputs: next column (i.e., next block in sequence)
+/*
+o = (uint64_t*)inout;
+printf("S inout1 %016lx %016lx %016lx %016lx\n",o[0],o[1],o[2],o[3]);
+printf("S inout1 %016lx %016lx %016lx %016lx\n",o[4],o[5],o[6],o[7]);
+printf("S inout1 %016lx %016lx %016lx %016lx\n",o[8],o[9],o[10],o[11]);
+printf("S inout1 %016lx %016lx %016lx %016lx\n",o[12],o[13],o[14],o[15]);
+printf("S inout1 %016lx %016lx %016lx %016lx\n",o[16],o[17],o[18],o[19]);
+printf("S inout1 %016lx %016lx %016lx %016lx\n",o[20],o[21],o[22],o[23]);
+*/
+
+//Inputs: next column (i.e., next block in sequence)
        in    += BLOCK_LEN_M256I;
        inout += BLOCK_LEN_M256I;
        //Output: goes to previous column
@@ -949,6 +985,22 @@ inline void reducedDuplexRow( uint64_t *State, uint64_t *rowIn,
       _mm_prefetch( inout +  9, _MM_HINT_T0 );
       _mm_prefetch( inout + 11, _MM_HINT_T0 );
 
+/*
+uint64_t *io = (uint64_t*)inout;
+uint64_t *ii = (uint64_t*)in;
+
+printf("RDRS1 col= %d\n", i);
+printf("RDRS1 IO %016lx %016lx %016lx %016lx\n",io[0],io[1],io[2],io[3]);
+printf("RDRS1 IO %016lx %016lx %016lx %016lx\n",io[4],io[5],io[6],io[7]);
+printf("RDRS1 IO %016lx %016lx %016lx %016lx\n",io[8],io[9],io[10],io[11]);
+printf("RDRS1 IO %016lx %016lx %016lx %016lx\n",io[12],io[13],io[14],io[15]);
+printf("RDRS1 IN %016lx %016lx %016lx %016lx\n",ii[0],ii[1],ii[2],ii[3]);
+printf("RDRS1 IN %016lx %016lx %016lx %016lx\n",ii[4],ii[5],ii[6],ii[7]);
+printf("RDRS1 IN %016lx %016lx %016lx %016lx\n",ii[8],ii[9],ii[10],ii[11]);
+printf("RDRS1 IN %016lx %016lx %016lx %016lx\n",ii[12],ii[13],ii[14],ii[15]);
+*/
+
+
       //Absorbing "M[prev] [+] M[row*]"
       state0 = _mm256_xor_si256( state0,
                                      _mm256_add_epi64( in[0], inout[0] ) );

algo/lyra2/sponge.h

@@ -52,8 +52,46 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
 // However, 2 way parallel looks trivial to code for AVX512 except for
 // a data dependency with rowa.
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+#define G2W_4X64(a,b,c,d) \
+   a = _mm512_add_epi64( a, b ); \
+   d = mm512_ror_64( _mm512_xor_si512( d, a ), 32 ); \
+   c = _mm512_add_epi64( c, d ); \
+   b = mm512_ror_64( _mm512_xor_si512( b, c ), 24 ); \
+   a = _mm512_add_epi64( a, b ); \
+   d = mm512_ror_64( _mm512_xor_si512( d, a ), 16 ); \
+   c = _mm512_add_epi64( c, d ); \
+   b = mm512_ror_64( _mm512_xor_si512( b, c ), 63 );
+
+#define LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   G2W_4X64( s0, s1, s2, s3 ); \
+   s1 = mm512_ror256_64( s1); \
+   s2 = mm512_swap256_128( s2 ); \
+   s3 = mm512_rol256_64( s3 ); \
+   G2W_4X64( s0, s1, s2, s3 ); \
+   s1 = mm512_rol256_64( s1 ); \
+   s2 = mm512_swap256_128( s2 ); \
+   s3 = mm512_ror256_64( s3 );
+
+#define LYRA_12_ROUNDS_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
+   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 )
+
+
+#endif  // AVX512
+
 #if defined __AVX2__
-// only available with avx2
 
 // process 4 columns in parallel
 // returns void, updates all args
@@ -89,9 +127,11 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
    LYRA_ROUND_AVX2( s0, s1, s2, s3 ) \
    LYRA_ROUND_AVX2( s0, s1, s2, s3 ) \
    LYRA_ROUND_AVX2( s0, s1, s2, s3 ) \
-   LYRA_ROUND_AVX2( s0, s1, s2, s3 ) \
+   LYRA_ROUND_AVX2( s0, s1, s2, s3 )
 
-#elif defined(__SSE2__)
+#endif
+
+#if defined(__SSE2__)
 
 // process 2 columns in parallel
 // returns void, all args updated
@@ -108,14 +148,14 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
 #define LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
    G_2X64( s0, s2, s4, s6 ); \
    G_2X64( s1, s3, s5, s7 ); \
-   mm128_ror1x64_256( s2, s3 ); \
-   mm128_swap128_256( s4, s5 ); \
-   mm128_rol1x64_256( s6, s7 ); \
+   mm128_ror256_64( s2, s3 ); \
+   mm128_swap256_128( s4, s5 ); \
+   mm128_rol256_64( s6, s7 ); \
    G_2X64( s0, s2, s4, s6 ); \
    G_2X64( s1, s3, s5, s7 ); \
-   mm128_rol1x64_256( s2, s3 ); \
-   mm128_swap128_256( s4, s5 ); \
-   mm128_ror1x64_256( s6, s7 );
+   mm128_rol256_64( s2, s3 ); \
+   mm128_swap256_128( s4, s5 ); \
+   mm128_ror256_64( s6, s7 );
 
 #define LYRA_12_ROUNDS_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
    LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
@@ -129,7 +169,7 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
    LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
    LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
    LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
-   LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
+   LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7)
 
 
 #endif // AVX2 else SSE2
@@ -161,6 +201,42 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
     G(r,7,v[ 3],v[ 4],v[ 9],v[14]);
 
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+union _povly
+{
+   __m512i *v512;
+   __m256i *v256;
+   uint64_t *u64;
+};
+typedef union _povly povly;
+
+//---- Housekeeping
+void initState_2way( uint64_t State[/*16*/] );
+
+//---- Squeezes
+void squeeze_2way( uint64_t *State, unsigned char *out, unsigned int len );
+void reducedSqueezeRow0_2way( uint64_t* state, uint64_t* row, uint64_t nCols );
+
+//---- Absorbs
+void absorbBlock_2way( uint64_t *State, const uint64_t *In0,
+                       const uint64_t *In1 );
+void absorbBlockBlake2Safe_2way( uint64_t *State, const uint64_t *In,
+                            const uint64_t nBlocks, const uint64_t block_len );
+
+//---- Duplexes
+void reducedDuplexRow1_2way( uint64_t *State, uint64_t *rowIn,
+                             uint64_t *rowOut, uint64_t nCols);
+void reducedDuplexRowSetup_2way( uint64_t *State, uint64_t *rowIn,
+                    uint64_t *rowInOut, uint64_t *rowOut, uint64_t nCols );
+
+void reducedDuplexRow_2way( uint64_t *State, uint64_t *rowIn,
+                            uint64_t *rowInOut0, uint64_t *rowInOut1,
+                            uint64_t *rowOut, uint64_t nCols);
+
+#endif
+
+
 //---- Housekeeping
 void initState(uint64_t state[/*16*/]);
 
@@ -178,20 +254,4 @@ void reducedDuplexRow1(uint64_t *state, uint64_t *rowIn, uint64_t *rowOut, uint6
 void reducedDuplexRowSetup(uint64_t *state, uint64_t *rowIn, uint64_t *rowInOut, uint64_t *rowOut, uint64_t nCols);
 void reducedDuplexRow(uint64_t *state, uint64_t *rowIn, uint64_t *rowInOut, uint64_t *rowOut, uint64_t nCols);
 
-//---- Misc
-void printArray(unsigned char *array, unsigned int size, char *name);
-
-////////////////////////////////////////////////////////////////////////////////////////////////
-
-
-////TESTS////
-//void reducedDuplexRowc(uint64_t *state, uint64_t *rowIn, uint64_t *rowInOut, uint64_t *rowOut);
-//void reducedDuplexRowd(uint64_t *state, uint64_t *rowIn, uint64_t *rowInOut, uint64_t *rowOut);
-//void reducedDuplexRowSetupv4(uint64_t *state, uint64_t *rowIn1, uint64_t *rowIn2, uint64_t *rowOut1, uint64_t *rowOut2);
-//void reducedDuplexRowSetupv5(uint64_t *state, uint64_t *rowIn, uint64_t *rowInOut, uint64_t *rowOut);
-//void reducedDuplexRowSetupv5c(uint64_t *state, uint64_t *rowIn, uint64_t *rowInOut, uint64_t *rowOut);
-//void reducedDuplexRowSetupv5d(uint64_t *state, uint64_t *rowIn, uint64_t *rowInOut, uint64_t *rowOut);
-/////////////
-
-
 #endif /* SPONGE_H_ */

algo/nist5/nist5-4way.c

@@ -3,22 +3,129 @@
 #include <stdint.h>
 #include <string.h>
 #include <stdio.h>
-
-#if defined(NIST5_4WAY)
-
 #include "algo/blake/blake-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/groestl/aes_ni/hash-groestl.h"
 
-void nist5hash_4way( void *out, const void *input )
+#if defined(NIST5_8WAY)
+
+void nist5hash_8way( void *out, const void *input )
 {
+     uint64_t vhash[8*16] __attribute__ ((aligned (128)));
+     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 hash4[8] __attribute__ ((aligned (64)));
+     uint64_t hash5[8] __attribute__ ((aligned (64)));
+     uint64_t hash6[8] __attribute__ ((aligned (64)));
+     uint64_t hash7[8] __attribute__ ((aligned (64)));
+
+     blake512_8way_context  ctx_blake;
+     hashState_groestl      ctx_groestl;
+     jh512_8way_context     ctx_jh;
+     skein512_8way_context  ctx_skein;
+     keccak512_8way_context ctx_keccak;
+
+     blake512_8way_init( &ctx_blake );
+     blake512_8way_update( &ctx_blake, input, 80 );
+     blake512_8way_close( &ctx_blake, vhash );
+
+     dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+                   vhash, 512 );
+
+     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 );
+     init_groestl( &ctx_groestl, 64 );
+     update_and_final_groestl( &ctx_groestl, (char*)hash4,
+                               (const char*)hash4, 512 );
+     init_groestl( &ctx_groestl, 64 );
+     update_and_final_groestl( &ctx_groestl, (char*)hash5,
+                               (const char*)hash5, 512 );
+     init_groestl( &ctx_groestl, 64 );
+     update_and_final_groestl( &ctx_groestl, (char*)hash6,
+                               (const char*)hash6, 512 );
+     init_groestl( &ctx_groestl, 64 );
+     update_and_final_groestl( &ctx_groestl, (char*)hash7,
+                               (const char*)hash7, 512 );
+
+     intrlv_8x64( vhash, hash0, hash1, hash2, hash3, hash4, hash5, hash6,
+                         hash7, 512 );
+
+     jh512_8way_init( &ctx_jh );
+     jh512_8way_update( &ctx_jh, vhash, 64 );
+     jh512_8way_close( &ctx_jh, vhash );
+
+     keccak512_8way_init( &ctx_keccak );
+     keccak512_8way_update( &ctx_keccak, vhash, 64 );
+     keccak512_8way_close( &ctx_keccak, vhash );
+
+     skein512_8way_init( &ctx_skein );
+     skein512_8way_update( &ctx_skein, vhash, 64 );
+     skein512_8way_close( &ctx_skein, out );
+}
+
+int scanhash_nist5_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[24*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;
+     uint32_t n = pdata[19];
+     const uint32_t first_nonce = pdata[19];
+     const uint32_t Htarg = ptarget[7];
+     __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 );
+
+        nist5hash_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;
+     return 0;
+}
+
+#elif defined(NIST5_4WAY)
+
+void nist5hash_4way( void *out, const void *input )
+{
+     uint64_t vhash[8*4] __attribute__ ((aligned (128)));
      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 vhash[8*4] __attribute__ ((aligned (64)));
      blake512_4way_context  ctx_blake;
      hashState_groestl      ctx_groestl;
      jh512_4way_context     ctx_jh;
@@ -62,62 +169,39 @@ void nist5hash_4way( void *out, const void *input )
 int scanhash_nist5_4way( struct work *work, uint32_t max_nonce,
                          uint64_t *hashes_done, struct thr_info *mythr )
 {
+     uint32_t vdata[4*24] __attribute__ ((aligned (128)));
      uint32_t hash[4*16] __attribute__ ((aligned (64)));
      uint32_t *hash7 = &(hash[25]);
      uint32_t lane_hash[8] __attribute__ ((aligned (32)));
-     uint32_t vdata[24*4] __attribute__ ((aligned (64)));
      uint32_t *pdata = work->data;
      uint32_t *ptarget = work->target;
      uint32_t n = pdata[19];
      const uint32_t first_nonce = pdata[19];
      const uint32_t Htarg = ptarget[7];
      __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
-     int thr_id = mythr->id;  // thr_id arg is deprecated
-
-     uint64_t htmax[] = {          0,
-                                 0xF,
-                                0xFF,
-                               0xFFF,
-                              0xFFFF,
-                          0x10000000 };
-
-     uint32_t masks[] = { 0xFFFFFFFF,
-                          0xFFFFFFF0,
-                          0xFFFFFF00,
-                          0xFFFFF000,
-                          0xFFFF0000,
-                                   0 };
+     int thr_id = mythr->id;  
 
      mm256_bswap32_intrlv80_4x64( vdata, pdata );
 
-     for ( int m=0; m < 6; m++ )
-     {
-        if (Htarg <= htmax[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 );
+
+        nist5hash_4way( hash, vdata );
+
+        for ( int lane = 0; lane < 4; lane++ )
+        if ( hash7[ lane<<1 ] < Htarg )
         {
-           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 );
-
-              nist5hash_4way( hash, vdata );
-
-              for ( int lane = 0; lane < 4; lane++ )
-              if ( ( hash7[ lane ] & mask ) == 0 )
-              {
-                 extr_lane_4x64( 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 );
-           break;
+           extr_lane_4x64( lane_hash, hash, lane, 256 );
+           if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
+           {
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
+           }
         }
-     }
-     *hashes_done = n - first_nonce + 1;
+        n += 4;
+     } while ( ( n < max_nonce-4 ) && !work_restart[thr_id].restart );
+     *hashes_done = n - first_nonce;
      return 0;
 }
 

algo/nist5/nist5-gate.c

@@ -2,8 +2,11 @@
 
 bool register_nist5_algo( algo_gate_t* gate )
 {
-    gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
-#if defined (NIST5_4WAY)
+    gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT;
+#if defined (NIST5_8WAY)
+    gate->scanhash = (void*)&scanhash_nist5_8way;
+    gate->hash     = (void*)&nist5hash_8way;
+#elif defined (NIST5_4WAY)
     gate->scanhash = (void*)&scanhash_nist5_4way;
     gate->hash     = (void*)&nist5hash_4way;
 #else

algo/nist5/nist5-gate.h

@@ -1,14 +1,23 @@
 #ifndef __NIST5_GATE_H__
-#define __NIST5_GATE_H__
+#define __NIST5_GATE_H__ 1
 
 #include "algo-gate-api.h"
 #include <stdint.h>
 
-#if defined(__AVX2__) && defined(__AES__)
-  #define NIST5_4WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define NIST5_8WAY 1
+#elif defined(__AVX2__) && defined(__AES__)
+  #define NIST5_4WAY 1
 #endif
 
-#if defined(NIST5_4WAY)
+#if defined(NIST5_8WAY)
+
+void nist5hash_8way( void *state, const void *input );
+
+int scanhash_nist5_8way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+
+#elif defined(NIST5_4WAY)
 
 void nist5hash_4way( void *state, const void *input );
 

algo/quark/hmq1725-gate.c

@@ -2,7 +2,10 @@
 
 bool register_hmq1725_algo( algo_gate_t* gate )
 {
-#if defined(HMQ1725_4WAY)
+#if defined(HMQ1725_8WAY)
+  gate->scanhash  = (void*)&scanhash_hmq1725_8way;
+  gate->hash      = (void*)&hmq1725_8way_hash;
+#elif defined(HMQ1725_4WAY)
   gate->scanhash  = (void*)&scanhash_hmq1725_4way;
   gate->hash      = (void*)&hmq1725_4way_hash;
 #else
@@ -10,7 +13,7 @@ bool register_hmq1725_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_hmq1725;
   gate->hash      = (void*)&hmq1725hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT;
   opt_target_factor = 65536.0;
   return true;
 };

algo/quark/hmq1725-gate.h

@@ -4,13 +4,21 @@
 #include "algo-gate-api.h"
 #include <stdint.h>
 
-#if defined(__AVX2__) && defined(__AES__)
-//  #define HMQ1725_4WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define HMQ1725_8WAY 1
+#elif defined(__AVX2__) && defined(__AES__)
+  #define HMQ1725_4WAY 1
 #endif
 
 bool register_hmq1725_algo( algo_gate_t* gate );
 
-#if defined(HMQ1725_4WAY)
+#if defined(HMQ1725_8WAY)
+
+void hmq1725_8way_hash( void *state, const void *input );
+int scanhash_hmq1725_8way( struct work *work, uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr );
+
+#elif defined(HMQ1725_4WAY)
 
 void hmq1725_4way_hash( void *state, const void *input );
 int scanhash_hmq1725_4way( struct work *work, uint32_t max_nonce,

algo/quark/hmq1725.c

@@ -333,6 +333,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
 			if (((hash64[7]&0xFFFFFFFF)==0) && 
 					fulltest(hash64, ptarget)) {
 				*hashes_done = n - first_nonce + 1;
+            work_set_target_ratio( work, hash64 );
 				return true;
 			}
 		} while (n < max_nonce && !work_restart[thr_id].restart);	
@@ -346,6 +347,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
 			if (((hash64[7]&0xFFFFFFF0)==0) && 
 					fulltest(hash64, ptarget)) {
 				*hashes_done = n - first_nonce + 1;
+            work_set_target_ratio( work, hash64 );
 				return true;
 			}
 		} while (n < max_nonce && !work_restart[thr_id].restart);	
@@ -359,6 +361,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
 			if (((hash64[7]&0xFFFFFF00)==0) && 
 					fulltest(hash64, ptarget)) {
 				*hashes_done = n - first_nonce + 1;
+            work_set_target_ratio( work, hash64 );
 				return true;
 			}
 		} while (n < max_nonce && !work_restart[thr_id].restart);	
@@ -372,6 +375,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
 			if (((hash64[7]&0xFFFFF000)==0) && 
 					fulltest(hash64, ptarget)) {
 				*hashes_done = n - first_nonce + 1;
+            work_set_target_ratio( work, hash64 );
 				return true;
 			}
 		} while (n < max_nonce && !work_restart[thr_id].restart);	
@@ -386,6 +390,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
 			if (((hash64[7]&0xFFFF0000)==0) && 
 					fulltest(hash64, ptarget)) {
 				*hashes_done = n - first_nonce + 1;
+            work_set_target_ratio( work, hash64 );
 				return true;
 			}
 		} while (n < max_nonce && !work_restart[thr_id].restart);	
@@ -399,6 +404,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
 			hmq1725hash(hash64, endiandata);
 			if (fulltest(hash64, ptarget)) {
 				*hashes_done = n - first_nonce + 1;
+            work_set_target_ratio( work, hash64 );
 				return true;
 			}
 		} while (n < max_nonce && !work_restart[thr_id].restart);	

algo/quark/quark-4way.c

@@ -1,12 +1,8 @@
 #include "cpuminer-config.h"
 #include "quark-gate.h"
-
-#if defined (QUARK_4WAY)
-
 #include <stdio.h>
 #include <string.h>
 #include <stdint.h>
-
 #include "algo/blake/blake-hash-4way.h"
 #include "algo/bmw/bmw-hash-4way.h"
 #include "algo/skein/skein-hash-4way.h"
@@ -14,6 +10,244 @@
 #include "algo/keccak/keccak-hash-4way.h"
 #include "algo/groestl/aes_ni/hash-groestl.h"
 
+#if defined (QUARK_8WAY)
+
+typedef struct {
+    blake512_8way_context  blake;
+    bmw512_8way_context    bmw;
+    hashState_groestl      groestl;
+    jh512_8way_context     jh;
+    skein512_8way_context  skein;
+    keccak512_8way_context keccak;
+} quark_8way_ctx_holder;
+
+quark_8way_ctx_holder quark_8way_ctx __attribute__ ((aligned (128)));
+
+void init_quark_8way_ctx()
+{
+     blake512_8way_init( &quark_8way_ctx.blake );
+     bmw512_8way_init( &quark_8way_ctx.bmw );
+     init_groestl( &quark_8way_ctx.groestl, 64 );
+     skein512_8way_init( &quark_8way_ctx.skein );
+     jh512_8way_init( &quark_8way_ctx.jh );
+     keccak512_8way_init( &quark_8way_ctx.keccak );
+}
+
+void quark_8way_hash( void *state, const void *input )
+{
+    uint64_t vhash[8*8] __attribute__ ((aligned (128)));
+    uint64_t vhashA[8*8] __attribute__ ((aligned (64)));
+    uint64_t vhashB[8*8] __attribute__ ((aligned (64)));
+    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 hash4[8] __attribute__ ((aligned (64)));
+    uint64_t hash5[8] __attribute__ ((aligned (64)));
+    uint64_t hash6[8] __attribute__ ((aligned (64)));
+    uint64_t hash7[8] __attribute__ ((aligned (64)));
+    __m512i* vh  = (__m512i*)vhash;
+    __m512i* vhA = (__m512i*)vhashA;
+    __m512i* vhB = (__m512i*)vhashB;
+    __mmask8 vh_mask;
+    quark_8way_ctx_holder ctx;
+    const uint32_t mask = 8;
+    const __m512i bit3_mask = m512_const1_64( mask );
+    const __m512i zero = _mm512_setzero_si512();
+
+    memcpy( &ctx, &quark_8way_ctx, sizeof(quark_8way_ctx) );
+
+    blake512_8way_update( &ctx.blake, input, 80 );
+    blake512_8way_close( &ctx.blake, vhash );
+
+    bmw512_8way_update( &ctx.bmw, vhash, 64 );
+    bmw512_8way_close( &ctx.bmw, vhash );
+
+    vh_mask = _mm512_cmpeq_epi64_mask( _mm512_and_si512( vh[0], bit3_mask ),
+                                       zero );
+
+    dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+                  vhash, 512 );
+
+    if ( hash0[0] & mask )
+    {
+       update_and_final_groestl( &ctx.groestl, (char*)hash0,
+                                               (char*)hash0, 512 );
+    }
+    if ( hash1[0] & mask )
+    {
+       reinit_groestl( &ctx.groestl );
+       update_and_final_groestl( &ctx.groestl, (char*)hash1,
+                                               (char*)hash1, 512 );
+    }
+    if ( hash2[0] & mask )
+    {
+       reinit_groestl( &ctx.groestl );
+       update_and_final_groestl( &ctx.groestl, (char*)hash2,
+                                               (char*)hash2, 512 );
+    }
+    if ( hash3[0] & mask )
+    {
+       reinit_groestl( &ctx.groestl );
+       update_and_final_groestl( &ctx.groestl, (char*)hash3,
+                                               (char*)hash3, 512 );
+    }
+    if ( hash4[0] & mask )
+    {
+       reinit_groestl( &ctx.groestl );
+       update_and_final_groestl( &ctx.groestl, (char*)hash4,
+                                               (char*)hash4, 512 );
+    }
+    if ( hash5[0] & mask )
+    {
+       reinit_groestl( &ctx.groestl );
+       update_and_final_groestl( &ctx.groestl, (char*)hash5,
+                                               (char*)hash5, 512 );
+    }
+    if ( hash6[0] & mask )
+    {
+       reinit_groestl( &ctx.groestl );
+       update_and_final_groestl( &ctx.groestl, (char*)hash6,
+                                               (char*)hash6, 512 );
+    }
+    if ( hash7[0] & mask )
+    {
+       reinit_groestl( &ctx.groestl );
+       update_and_final_groestl( &ctx.groestl, (char*)hash7,
+                                               (char*)hash7, 512 );
+    }
+
+    intrlv_8x64( vhashA, hash0, hash1, hash2, hash3, hash4, hash5, hash6,
+                         hash7, 512 );
+
+    if ( vh_mask & 0xff )
+    {
+       skein512_8way_update( &ctx.skein, vhash, 64 );
+       skein512_8way_close( &ctx.skein, vhashB );
+    }
+
+    mm512_blend_hash_8x64( vh, vhA, vhB, vh_mask );
+
+    dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+                  vhash, 512 );
+
+    reinit_groestl( &ctx.groestl );
+    update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
+    reinit_groestl( &ctx.groestl );
+    update_and_final_groestl( &ctx.groestl, (char*)hash1, (char*)hash1, 512 );
+    reinit_groestl( &ctx.groestl );
+    update_and_final_groestl( &ctx.groestl, (char*)hash2, (char*)hash2, 512 );
+    reinit_groestl( &ctx.groestl );
+    update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
+    reinit_groestl( &ctx.groestl );
+    update_and_final_groestl( &ctx.groestl, (char*)hash4, (char*)hash4, 512 );
+    reinit_groestl( &ctx.groestl );
+    update_and_final_groestl( &ctx.groestl, (char*)hash5, (char*)hash5, 512 );
+    reinit_groestl( &ctx.groestl );
+    update_and_final_groestl( &ctx.groestl, (char*)hash6, (char*)hash6, 512 );
+    reinit_groestl( &ctx.groestl );
+    update_and_final_groestl( &ctx.groestl, (char*)hash7, (char*)hash7, 512 );
+
+    intrlv_8x64( vhash, hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
+                 512 );
+
+    jh512_8way_update( &ctx.jh, vhash, 64 );
+    jh512_8way_close( &ctx.jh, vhash );
+
+    vh_mask = _mm512_cmpeq_epi64_mask( _mm512_and_si512( vh[0], bit3_mask ),
+                                       zero );
+
+    if ( ( vh_mask & 0xff ) != 0xff )
+    {
+       blake512_8way_init( &ctx.blake );
+       blake512_8way_update( &ctx.blake, vhash, 64 );
+       blake512_8way_close( &ctx.blake, vhashA );
+    }
+
+    if ( vh_mask & 0xff )
+    {
+       bmw512_8way_init( &ctx.bmw );
+       bmw512_8way_update( &ctx.bmw, vhash, 64 );
+       bmw512_8way_close( &ctx.bmw, vhashB );
+    }
+
+    mm512_blend_hash_8x64( vh, vhA, vhB, vh_mask );
+
+    keccak512_8way_update( &ctx.keccak, vhash, 64 );
+    keccak512_8way_close( &ctx.keccak, vhash );
+
+    skein512_8way_init( &ctx.skein );
+    skein512_8way_update( &ctx.skein, vhash, 64 );
+    skein512_8way_close( &ctx.skein, vhash );
+
+    vh_mask = _mm512_cmpeq_epi64_mask( _mm512_and_si512( vh[0], bit3_mask ),
+                                       zero );
+
+    if ( ( vh_mask & 0xff ) != 0xff )
+    {
+       keccak512_8way_init( &ctx.keccak );
+       keccak512_8way_update( &ctx.keccak, vhash, 64 );
+       keccak512_8way_close( &ctx.keccak, vhashA );
+    }
+
+    if ( vh_mask & 0xff )
+    {
+       jh512_8way_init( &ctx.jh );
+       jh512_8way_update( &ctx.jh, vhash, 64 );
+       jh512_8way_close( &ctx.jh, vhashB );
+    }
+
+    // Final blend, directly to state, only need 32 bytes.
+    casti_m512i( state,0 ) = _mm512_mask_blend_epi64( vh_mask, vhA[0], vhB[0] );
+    casti_m512i( state,1 ) = _mm512_mask_blend_epi64( vh_mask, vhA[1], vhB[1] );
+    casti_m512i( state,2 ) = _mm512_mask_blend_epi64( vh_mask, vhA[2], vhB[2] );
+    casti_m512i( state,3 ) = _mm512_mask_blend_epi64( vh_mask, vhA[3], vhB[3] );
+}
+
+int scanhash_quark_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[24*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;
+    uint32_t n = pdata[19];
+    const uint32_t first_nonce = pdata[19];
+    __m512i  *noncev = (__m512i*)vdata + 9;   // aligned
+    int thr_id = mythr->id;  // thr_id arg is deprecated
+
+    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 );
+
+       quark_8way_hash( hash, vdata );
+       pdata[19] = n;
+
+       for ( int i = 0; i < 8; i++ )
+       if ( ( hash7[ i<<1 ] & 0xFFFFFF00 ) == 0 )
+       {
+          extr_lane_8x64( lane_hash, hash, i, 256 );
+          if ( fulltest( lane_hash, ptarget ) && !opt_benchmark  )
+          {
+            pdata[19] = n+i;
+            submit_lane_solution( work, lane_hash, mythr, i );
+          }
+       }
+       n += 8;
+    } while ( ( n < max_nonce-8 ) && !work_restart[thr_id].restart );
+
+    *hashes_done = n - first_nonce;
+    return 0;
+}
+
+
+#elif defined (QUARK_4WAY)
+
 typedef struct {
     blake512_4way_context  blake;
     bmw512_4way_context    bmw;
@@ -91,7 +325,7 @@ void quark_4way_hash( void *state, const void *input )
 
     intrlv_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
 
-    if ( mm256_anybits0( vh_mask ) )   
+    if ( mm256_anybits1( vh_mask ) )   
     {
        skein512_4way( &ctx.skein, vhash, 64 );
        skein512_4way_close( &ctx.skein, vhashB );
@@ -117,14 +351,14 @@ void quark_4way_hash( void *state, const void *input )
 
     vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ), zero );
 
-    if ( mm256_anybits1( vh_mask ) )
+    if ( mm256_anybits0( vh_mask ) )   
     {
        blake512_4way_init( &ctx.blake );
        blake512_4way( &ctx.blake, vhash, 64 );
        blake512_4way_close( &ctx.blake, vhashA );
     }
 
-    if ( mm256_anybits0( vh_mask ) )
+    if ( mm256_anybits1( vh_mask ) )
     {
        bmw512_4way_init( &ctx.bmw );
        bmw512_4way( &ctx.bmw, vhash, 64 );
@@ -142,14 +376,14 @@ void quark_4way_hash( void *state, const void *input )
 
     vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ), zero );
 
-    if ( mm256_anybits1( vh_mask ) )
+    if ( mm256_anybits0( vh_mask ) )    
     {
        keccak512_4way_init( &ctx.keccak );
        keccak512_4way( &ctx.keccak, vhash, 64 );
        keccak512_4way_close( &ctx.keccak, vhashA );
     }
 
-    if ( mm256_anybits0( vh_mask ) )
+    if ( mm256_anybits1( vh_mask ) )
     {
        jh512_4way_init( &ctx.jh );
        jh512_4way( &ctx.jh, vhash, 64 );

algo/quark/quark-gate.c

@@ -2,7 +2,11 @@
 
 bool register_quark_algo( algo_gate_t* gate )
 {
-#if defined (QUARK_4WAY)
+#if defined (QUARK_8WAY)
+  init_quark_8way_ctx();
+  gate->scanhash  = (void*)&scanhash_quark_8way;
+  gate->hash      = (void*)&quark_8way_hash;
+#elif defined (QUARK_4WAY)
   init_quark_4way_ctx();
   gate->scanhash  = (void*)&scanhash_quark_4way;
   gate->hash      = (void*)&quark_4way_hash;
@@ -11,7 +15,7 @@ bool register_quark_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_quark;
   gate->hash      = (void*)&quark_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT;
   return true;
 };
 

algo/quark/quark-gate.h

@@ -4,13 +4,22 @@
 #include "algo-gate-api.h"
 #include <stdint.h>
 
-#if defined(__AVX2__) && defined(__AES__)
-  #define QUARK_4WAY
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define QUARK_8WAY 1
+#elif defined(__AVX2__) && defined(__AES__)
+  #define QUARK_4WAY 1
 #endif
 
 bool register_quark_algo( algo_gate_t* gate );
 
-#if defined(QUARK_4WAY)
+#if defined(QUARK_8WAY)
+
+void quark_8way_hash( void *state, const void *input );
+int scanhash_quark_8way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+void init_quark_8way_ctx();
+
+#elif defined(QUARK_4WAY)
 
 void quark_4way_hash( void *state, const void *input );
 int scanhash_quark_4way( struct work *work, uint32_t max_nonce,

algo/qubit/qubit-2way.c

@@ -1,17 +1,134 @@
 #include "qubit-gate.h"
-
-#if defined(QUBIT_2WAY)
-
 #include <stdlib.h>
 #include <stdint.h>
 #include <string.h>
 #include <stdio.h>
 #include "algo/luffa/luffa-hash-2way.h"
-#include "algo/cubehash/cubehash_sse2.h" 
+#include "algo/cubehash/cube-hash-2way.h"
+#include "algo/cubehash/cubehash_sse2.h"
 #include "algo/simd/simd-hash-2way.h"
 #include "algo/shavite/sph_shavite.h"
 #include "algo/echo/aes_ni/hash_api.h"
 
+#if defined(QUBIT_4WAY)
+
+typedef struct
+{
+    luffa_4way_context      luffa;
+    cube_4way_context       cube;
+    sph_shavite512_context  shavite;
+    simd_4way_context       simd;
+    simd_2way_context       simd2;
+    hashState_echo          echo;
+} qubit_4way_ctx_holder;
+
+qubit_4way_ctx_holder qubit_4way_ctx;
+
+void init_qubit_4way_ctx()
+{
+    cube_4way_init( &qubit_4way_ctx.cube, 512, 16, 32 );
+    sph_shavite512_init(&qubit_4way_ctx.shavite);
+    simd_4way_init( &qubit_4way_ctx.simd, 512 );
+    simd_2way_init( &qubit_4way_ctx.simd2, 512 );
+    init_echo(&qubit_4way_ctx.echo, 512);
+};
+
+void qubit_4way_hash( void *output, const void *input )
+{
+     uint32_t vhash[16*4] __attribute__ ((aligned (128)));
+     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)));
+     qubit_4way_ctx_holder ctx;
+
+     memcpy( &ctx, &qubit_4way_ctx, sizeof(qubit_4way_ctx) );
+
+     luffa_4way_update( &ctx.luffa, input + (64<<2), 16 );
+     luffa_4way_close( &ctx.luffa, vhash );
+     
+     cube_4way_update_close( &ctx.cube, vhash, vhash, 64 );
+     dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
+     
+     sph_shavite512( &ctx.shavite, hash0, 64 );
+     sph_shavite512_close( &ctx.shavite, hash0 );
+     memcpy( &ctx.shavite, &qubit_4way_ctx.shavite,
+             sizeof(sph_shavite512_context) );
+     sph_shavite512( &ctx.shavite, hash1, 64 );
+     sph_shavite512_close( &ctx.shavite, hash1 );
+     memcpy( &ctx.shavite, &qubit_4way_ctx.shavite,
+             sizeof(sph_shavite512_context) );
+     sph_shavite512( &ctx.shavite, hash2, 64 );
+     sph_shavite512_close( &ctx.shavite, hash2 );
+     memcpy( &ctx.shavite, &qubit_4way_ctx.shavite,
+             sizeof(sph_shavite512_context) );
+     sph_shavite512( &ctx.shavite, hash3, 64 );
+     sph_shavite512_close( &ctx.shavite, hash3 );
+
+     intrlv_4x128_512( vhash, hash0, hash1, hash2, hash3 );
+     simd_4way_update_close( &ctx.simd, vhash, vhash, 512 );
+     dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
+
+     update_final_echo( &ctx.echo, (BitSequence *)hash0,
+                       (const BitSequence *) hash0, 512 );
+     memcpy( &ctx.echo, &qubit_4way_ctx.echo, sizeof(hashState_echo) );
+     update_final_echo( &ctx.echo, (BitSequence *)hash1,
+                       (const BitSequence *) hash1, 512 );
+     memcpy( &ctx.echo, &qubit_4way_ctx.echo, sizeof(hashState_echo) );
+     update_final_echo( &ctx.echo, (BitSequence *)hash2,
+                       (const BitSequence *) hash2, 512 );
+     memcpy( &ctx.echo, &qubit_4way_ctx.echo, sizeof(hashState_echo) );
+     update_final_echo( &ctx.echo, (BitSequence *)hash3,
+                       (const BitSequence *) hash3, 512 );
+
+     memcpy( output,    hash0, 32 );
+     memcpy( output+32, hash1, 32 );
+     memcpy( output+64, hash2, 32 );
+     memcpy( output+96, hash3, 32 );
+}
+
+int scanhash_qubit_4way( struct work *work,uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr )
+{
+     uint32_t hash[4*8] __attribute__ ((aligned (128)));
+     uint32_t vdata[24*4] __attribute__ ((aligned (64)));
+     uint32_t *pdata = work->data;
+     uint32_t *ptarget = work->target;
+     uint32_t n = pdata[19];
+     const uint32_t first_nonce = pdata[19];
+     uint32_t *noncep = vdata + 64+3;   // 4*16 + 3
+     int thr_id = mythr->id;
+     const uint32_t Htarg = ptarget[7];
+
+     mm512_bswap32_intrlv80_4x128( vdata, pdata );
+     luffa_4way_init( &qubit_4way_ctx.luffa, 512 );
+     luffa_4way_update( &qubit_4way_ctx.luffa, vdata, 64 );
+
+     do
+     {
+        be32enc( noncep,    n   );
+        be32enc( noncep+ 4, n+1 );
+        be32enc( noncep+ 8, n+2 );
+        be32enc( noncep+12, n+3 );
+
+        qubit_4way_hash( hash, vdata );
+        pdata[19] = n;
+
+        for ( int lane = 0; lane < 4; lane++ )
+        if ( ( hash+(lane<<3) )[7] < Htarg )
+        if ( fulltest( hash+(lane<<3), ptarget) && !opt_benchmark )
+        {
+           pdata[19] = n + lane;
+           submit_lane_solution( work, hash+(lane<<3), mythr, lane );
+        }
+        n += 4;
+     } while ( ( n < max_nonce-4 ) && !work_restart[thr_id].restart );
+     *hashes_done = n - first_nonce;
+     return 0;
+}
+
+#elif defined(QUBIT_2WAY)
+
 typedef struct
 {
         luffa_2way_context      luffa;

algo/qubit/qubit-gate.c

@@ -2,7 +2,12 @@
 
 bool register_qubit_algo( algo_gate_t* gate )
 {
-#if defined (QUBIT_2WAY)
+   
+#if defined (QUBIT_4WAY)
+  init_qubit_4way_ctx();
+  gate->scanhash  = (void*)&scanhash_qubit_4way;
+  gate->hash      = (void*)&qubit_4way_hash;
+#elif defined (QUBIT_2WAY)
   init_qubit_2way_ctx();
   gate->scanhash  = (void*)&scanhash_qubit_2way;
   gate->hash      = (void*)&qubit_2way_hash;
@@ -11,7 +16,7 @@ bool register_qubit_algo( algo_gate_t* gate )
   gate->scanhash  = (void*)&scanhash_qubit;
   gate->hash      = (void*)&qubit_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT;
   return true;
 };
 

algo/qubit/qubit-gate.h

@@ -4,13 +4,23 @@
 #include "algo-gate-api.h"
 #include <stdint.h>
 
-#if defined(__AVX2__) && defined(__AES__)
-  #define QUBIT_2WAY
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define QUBIT_4WAY 1
+#elif defined(__AVX2__) && defined(__AES__)
+  #define QUBIT_2WAY 1
 #endif
 
 bool register_qubit_algo( algo_gate_t* gate );
 
-#if defined(QUBIT_2WAY)
+#if defined(QUBIT_4WAY)
+
+void qubit_4way_hash( void *state, const void *input );
+int scanhash_qubit_4way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+void init_qubit_4way_ctx();
+
+#elif defined(QUBIT_2WAY)
 
 void qubit_2way_hash( void *state, const void *input );
 int scanhash_qubit_2way( struct work *work, uint32_t max_nonce,

algo/ripemd/lbry-4way.c

@@ -10,7 +10,140 @@
 #define LBRY_MIDSTATE    64
 #define LBRY_TAIL (LBRY_INPUT_SIZE) - (LBRY_MIDSTATE)
 
-#if defined(LBRY_8WAY)
+#if defined(LBRY_16WAY)
+
+static __thread sha256_16way_context sha256_16w_mid;
+
+void lbry_16way_hash( void* output, const void* input )
+{
+   uint32_t _ALIGN(128) vhashA[16<<4];
+   uint32_t _ALIGN(64) vhashB[16<<4];
+   uint32_t _ALIGN(64) vhashC[16<<4];
+   uint32_t _ALIGN(64) h0[32];
+   uint32_t _ALIGN(64) h1[32];
+   uint32_t _ALIGN(64) h2[32];
+   uint32_t _ALIGN(64) h3[32];
+   uint32_t _ALIGN(64) h4[32];
+   uint32_t _ALIGN(64) h5[32];
+   uint32_t _ALIGN(64) h6[32];
+   uint32_t _ALIGN(64) h7[32];
+   uint32_t _ALIGN(64) h8[32];
+   uint32_t _ALIGN(64) h9[32];
+   uint32_t _ALIGN(64) h10[32];
+   uint32_t _ALIGN(64) h11[32];
+   uint32_t _ALIGN(64) h12[32];
+   uint32_t _ALIGN(64) h13[32];
+   uint32_t _ALIGN(64) h14[32];
+   uint32_t _ALIGN(64) h15[32];
+   sha256_16way_context     ctx_sha256 __attribute__ ((aligned (64)));
+   sha512_8way_context     ctx_sha512;
+   ripemd160_16way_context  ctx_ripemd;
+
+   memcpy( &ctx_sha256, &sha256_16w_mid, sizeof(ctx_sha256) );
+   sha256_16way_update( &ctx_sha256, input + (LBRY_MIDSTATE<<4), LBRY_TAIL );
+   sha256_16way_close( &ctx_sha256, vhashA );
+
+   sha256_16way_init( &ctx_sha256 );
+   sha256_16way_update( &ctx_sha256, vhashA, 32 );
+   sha256_16way_close( &ctx_sha256, vhashA );
+
+   // reinterleave to do sha512 4-way 64 bit twice.
+   dintrlv_16x32( h0, h1, h2, h3, h4, h5, h6, h7,
+                  h8, h9, h10, h11, h12, h13, h14, h15, vhashA, 256 );
+   intrlv_8x64( vhashA, h0, h1, h2, h3, h4, h5, h6, h7, 256 );
+   intrlv_8x64( vhashB, h8, h9, h10, h11, h12, h13, h14, h15, 256 );
+
+   sha512_8way_init( &ctx_sha512 );
+   sha512_8way_update( &ctx_sha512, vhashA, 32 );
+   sha512_8way_close( &ctx_sha512, vhashA );
+
+   sha512_8way_init( &ctx_sha512 );
+   sha512_8way_update( &ctx_sha512, vhashB, 32 );
+   sha512_8way_close( &ctx_sha512, vhashB );
+
+   // back to 8-way 32 bit
+   dintrlv_8x64( h0, h1, h2, h3,h4, h5, h6, h7, vhashA, 512 );
+   dintrlv_8x64( h8, h9, h10, h11, h12, h13, h14, h15, vhashB, 512 );
+   intrlv_16x32( vhashA, h0, h1, h2, h3, h4, h5, h6, h7,
+                         h8, h9, h10, h11, h12, h13, h14, h15, 512 );
+
+   ripemd160_16way_init( &ctx_ripemd );
+   ripemd160_16way_update( &ctx_ripemd, vhashA, 32 );
+   ripemd160_16way_close( &ctx_ripemd, vhashB );
+
+   ripemd160_16way_init( &ctx_ripemd );
+   ripemd160_16way_update( &ctx_ripemd, vhashA+(8<<4), 32 );
+   ripemd160_16way_close( &ctx_ripemd, vhashC );
+
+   sha256_16way_init( &ctx_sha256 );
+   sha256_16way_update( &ctx_sha256, vhashB, 20 );
+   sha256_16way_update( &ctx_sha256, vhashC, 20 );
+   sha256_16way_close( &ctx_sha256, vhashA );
+
+   sha256_16way_init( &ctx_sha256 );
+   sha256_16way_update( &ctx_sha256, vhashA, 32 );
+   sha256_16way_close( &ctx_sha256, output );
+}
+
+int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
+                        uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t hash[8*16] __attribute__ ((aligned (128)));
+   uint32_t vdata[32*16] __attribute__ ((aligned (64)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash7 = &(hash[7<<4]);
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   uint32_t n = pdata[27];
+   const uint32_t first_nonce = pdata[27];
+   const uint32_t Htarg = ptarget[7];
+   uint32_t edata[32] __attribute__ ((aligned (64)));
+   __m512i  *noncev = (__m512i*)vdata + 27;   // aligned
+   int thr_id = mythr->id;  // thr_id arg is deprecated
+
+   // we need bigendian data...
+   casti_m128i( edata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
+   casti_m128i( edata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
+   casti_m128i( edata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
+   casti_m128i( edata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
+   casti_m128i( edata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
+   casti_m128i( edata, 5 ) = mm128_bswap_32( casti_m128i( pdata, 5 ) );
+   casti_m128i( edata, 6 ) = mm128_bswap_32( casti_m128i( pdata, 6 ) );
+   casti_m128i( edata, 7 ) = mm128_bswap_32( casti_m128i( pdata, 7 ) );
+   intrlv_16x32( vdata, edata, edata, edata, edata, edata, edata, edata,
+        edata, edata, edata, edata, edata, edata, edata, edata, edata, 1024 );
+
+   sha256_16way_init( &sha256_16w_mid );
+   sha256_16way( &sha256_16w_mid, vdata, LBRY_MIDSTATE );
+
+   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 ) );
+      lbry_16way_hash( hash, vdata );
+
+      for ( int i = 0; i < 16; i++ )
+      if ( unlikely( hash7[ i ] <= Htarg ) )
+      {
+         // deinterleave hash for lane
+         extr_lane_16x32( lane_hash, hash, i, 256 );
+         if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
+         {
+            pdata[27] = n + i;
+            submit_lane_solution( work, lane_hash, mythr, i );
+         }
+      }
+      n += 16;
+   } while ( (n < max_nonce-16) && !work_restart[thr_id].restart );
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+
+
+#elif defined(LBRY_8WAY)
 
 static __thread sha256_8way_context sha256_8w_mid;
 
@@ -91,11 +224,6 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
    __m256i  *noncev = (__m256i*)vdata + 27;   // aligned
    int thr_id = mythr->id;  // thr_id arg is deprecated
 
-   uint64_t htmax[] = {          0,        0xF,       0xFF,
-                             0xFFF,     0xFFFF, 0x10000000 };
-   uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
-                        0xFFFFF000, 0xFFFF0000,          0 };
-
    // we need bigendian data...
    casti_m128i( edata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
    casti_m128i( edata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
@@ -106,33 +234,30 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
    casti_m128i( edata, 6 ) = mm128_bswap_32( casti_m128i( pdata, 6 ) );
    casti_m128i( edata, 7 ) = mm128_bswap_32( casti_m128i( pdata, 7 ) );
    intrlv_8x32( vdata, edata, edata, edata, edata,
-                             edata, edata, edata, edata, 1024 );
+                       edata, edata, edata, edata, 1024 );
+
    sha256_8way_init( &sha256_8w_mid );
    sha256_8way( &sha256_8w_mid, vdata, LBRY_MIDSTATE );
 
-   for ( int m = 0; m < sizeof(masks); m++ ) if ( Htarg <= htmax[m] )
+   do
    {
-      uint32_t mask = masks[m];
-      do
-      {
-        *noncev = mm256_bswap_32( _mm256_set_epi32(
-                                          n+7,n+6,n+5,n+4,n+3,n+2,n+1,n ) );
-         lbry_8way_hash( hash, vdata );
+      *noncev = mm256_bswap_32( _mm256_set_epi32(
+                                       n+7,n+6,n+5,n+4,n+3,n+2,n+1,n ) );
+      lbry_8way_hash( hash, vdata );
 
-         for ( int i = 0; i < 8; i++ )  if ( !( hash7[ i ] & mask ) )
+      for ( int i = 0; i < 8; i++ )
+      if ( unlikely( hash7[ i ] <= Htarg ) )
+      {
+         // deinterleave hash for lane
+         extr_lane_8x32( lane_hash, hash, i, 256 );
+         if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
          {
-            // deinterleave hash for lane
-            extr_lane_8x32( lane_hash, hash, i, 256 );
-            if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
-            {
-              pdata[27] = n + i;
-              submit_lane_solution( work, lane_hash, mythr, i );
-            }
+            pdata[27] = n + i;
+            submit_lane_solution( work, lane_hash, mythr, i );
          }
-         n += 8;
-      } while ( (n < max_nonce-10) && !work_restart[thr_id].restart );
-      break;
-   }
+      }
+      n += 8;
+   } while ( (n < max_nonce-10) && !work_restart[thr_id].restart );
    *hashes_done = n - first_nonce + 1;
    return 0;
 }

algo/ripemd/lbry-gate.c

@@ -98,16 +98,23 @@ int lbry_get_work_data_size() { return LBRY_WORK_DATA_SIZE; }
 
 bool register_lbry_algo( algo_gate_t* gate )
 {
-  gate->optimizations = AVX2_OPT | SHA_OPT;
-#if defined (LBRY_8WAY)
+  gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
+#if defined (LBRY_16WAY)
+  gate->scanhash              = (void*)&scanhash_lbry_16way;
+  gate->hash                  = (void*)&lbry_16way_hash;
+  gate->optimizations = AVX2_OPT | AVX512_OPT;
+#elif defined (LBRY_8WAY)
   gate->scanhash              = (void*)&scanhash_lbry_8way;
   gate->hash                  = (void*)&lbry_8way_hash;
+  gate->optimizations = AVX2_OPT | AVX512_OPT;
 #elif defined (LBRY_4WAY)
   gate->scanhash              = (void*)&scanhash_lbry_4way;
   gate->hash                  = (void*)&lbry_4way_hash;
+  gate->optimizations = AVX2_OPT | AVX512_OPT;
 #else 
   gate->scanhash              = (void*)&scanhash_lbry;
   gate->hash                  = (void*)&lbry_hash;
+  gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
 #endif
   gate->calc_network_diff     = (void*)&lbry_calc_network_diff;
   gate->build_stratum_request = (void*)&lbry_le_build_stratum_request;

algo/ripemd/lbry-gate.h

@@ -4,11 +4,20 @@
 #include "algo-gate-api.h"
 #include <stdint.h>
 
+
+// 16 way needs sha256 16 way
+//#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+//  #define LBRY_16WAY
+#if defined(__AVX2__)
+  #define LBRY_8WAY
+#endif
+/*
 #if !defined(__SHA__)
  #if defined(__AVX2__)
   #define LBRY_8WAY
  #endif
 #endif
+*/
 
 #define LBRY_NTIME_INDEX 25
 #define LBRY_NBITS_INDEX 26
@@ -18,7 +27,12 @@
 
 bool register_lbry_algo( algo_gate_t* gate );
 
-#if defined(LBRY_8WAY)
+#if defined(LBRY_16WAY)
+
+void lbry_16way_hash( void *state, const void *input );
+int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+#elif defined(LBRY_8WAY)
 
 void lbry_8way_hash( void *state, const void *input );
 int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,

algo/ripemd/lbry.c

@@ -80,9 +80,6 @@ int scanhash_lbry( struct work *work, uint32_t max_nonce,
 	// we need bigendian data...
         swab32_array( endiandata, pdata, 32 );
 
-#ifdef DEBUG_ALGO
-	printf("[%d] Htarg=%X\n", thr_id, Htarg);
-#endif
 	for (int m=0; m < sizeof(masks); m++) {
 		if (Htarg <= htmax[m]) {
 			uint32_t mask = masks[m];
@@ -90,23 +87,11 @@ int scanhash_lbry( struct work *work, uint32_t max_nonce,
 				pdata[27] = ++n;
 				be32enc(&endiandata[27], n);
 				lbry_hash(hash64, &endiandata);
-#ifndef DEBUG_ALGO
 				if ((!(hash64[7] & mask)) && fulltest(hash64, ptarget)) {
-					*hashes_done = n - first_nonce + 1;
-					return true;
+               pdata[27] = n;
+               submit_solution( work, hash64, mythr );
 				}
-#else
-				if (!(n % 0x1000) && !thr_id) printf(".");
-				if (!(hash64[7] & mask)) {
-					printf("[%d]",thr_id);
-					if (fulltest(hash64, ptarget)) {
-						*hashes_done = n - first_nonce + 1;
-						return true;
-					}
-				}
-#endif
-			} while (n < max_nonce && !work_restart[thr_id].restart);
-			// see blake.c if else to understand the loop on htmax => mask
+			} while ( (n < max_nonce -8) && !work_restart[thr_id].restart);
 			break;
 		}
 	}

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

@@ -623,3 +623,303 @@ void ripemd160_8way_close( ripemd160_8way_context  *sc, void *dst )
 
 #endif // __AVX2__
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+//  RIPEMD-160 16 way
+
+
+#define F16W_1(x, y, z) \
+   _mm512_xor_si512( _mm512_xor_si512( x, y ), z )
+
+#define F16W_2(x, y, z) \
+   _mm512_xor_si512( _mm512_and_si512( _mm512_xor_si512( y, z ), x ), z )
+
+#define F16W_3(x, y, z) \
+   _mm512_xor_si512( _mm512_or_si512( x, mm512_not( y ) ), z )
+
+#define F16W_4(x, y, z) \
+   _mm512_xor_si512( _mm512_and_si512( _mm512_xor_si512( x, y ), z ), y )
+
+#define F16W_5(x, y, z) \
+   _mm512_xor_si512( x, _mm512_or_si512( y, mm512_not( z ) ) )
+
+#define RR_16W(a, b, c, d, e, f, s, r, k) \
+do{ \
+   a = _mm512_add_epi32( mm512_rol_32( _mm512_add_epi32( _mm512_add_epi32( \
+                _mm512_add_epi32( a, f( b ,c, d ) ), r ), \
+                                 m512_const1_64( k ) ), s ), e ); \
+   c = mm512_rol_32( c, 10 );\
+} while (0)
+
+#define ROUND1_16W(a, b, c, d, e, f, s, r, k)  \
+        RR_16W(a ## 1, b ## 1, c ## 1, d ## 1, e ## 1, f, s, r, K1 ## k)
+
+#define ROUND2_16W(a, b, c, d, e, f, s, r, k)  \
+        RR_16W(a ## 2, b ## 2, c ## 2, d ## 2, e ## 2, f, s, r, K2 ## k)
+
+static void ripemd160_16way_round( ripemd160_16way_context *sc )
+{
+   const __m512i *in = (__m512i*)sc->buf;
+   __m512i *h  = (__m512i*)sc->val;
+   register __m512i A1, B1, C1, D1, E1;
+   register __m512i A2, B2, C2, D2, E2;
+   __m512i tmp;
+
+   A1 = A2 = h[0];
+   B1 = B2 = h[1];
+   C1 = C2 = h[2];
+   D1 = D2 = h[3];
+   E1 = E2 = h[4];
+
+   ROUND1_16W( A, B, C, D, E, F16W_1, 11, in[ 0], 1 );
+   ROUND1_16W( E, A, B, C, D, F16W_1, 14, in[ 1], 1 );
+   ROUND1_16W( D, E, A, B, C, F16W_1, 15, in[ 2], 1 );
+   ROUND1_16W( C, D, E, A, B, F16W_1, 12, in[ 3], 1 );
+   ROUND1_16W( B, C, D, E, A, F16W_1,  5, in[ 4], 1 );
+   ROUND1_16W( A, B, C, D, E, F16W_1,  8, in[ 5], 1 );
+   ROUND1_16W( E, A, B, C, D, F16W_1,  7, in[ 6], 1 );
+   ROUND1_16W( D, E, A, B, C, F16W_1,  9, in[ 7], 1 );
+   ROUND1_16W( C, D, E, A, B, F16W_1, 11, in[ 8], 1 );
+   ROUND1_16W( B, C, D, E, A, F16W_1, 13, in[ 9], 1 );
+   ROUND1_16W( A, B, C, D, E, F16W_1, 14, in[10], 1 );
+   ROUND1_16W( E, A, B, C, D, F16W_1, 15, in[11], 1 );
+   ROUND1_16W( D, E, A, B, C, F16W_1,  6, in[12], 1 );
+   ROUND1_16W( C, D, E, A, B, F16W_1,  7, in[13], 1 );
+   ROUND1_16W( B, C, D, E, A, F16W_1,  9, in[14], 1 );
+   ROUND1_16W( A, B, C, D, E, F16W_1,  8, in[15], 1 );
+
+   ROUND1_16W( E, A, B, C, D, F16W_2,  7, in[ 7], 2 );
+   ROUND1_16W( D, E, A, B, C, F16W_2,  6, in[ 4], 2 );
+   ROUND1_16W( C, D, E, A, B, F16W_2,  8, in[13], 2 );
+   ROUND1_16W( B, C, D, E, A, F16W_2, 13, in[ 1], 2 );
+   ROUND1_16W( A, B, C, D, E, F16W_2, 11, in[10], 2 );
+   ROUND1_16W( E, A, B, C, D, F16W_2,  9, in[ 6], 2 );
+   ROUND1_16W( D, E, A, B, C, F16W_2,  7, in[15], 2 );
+   ROUND1_16W( C, D, E, A, B, F16W_2, 15, in[ 3], 2 );
+   ROUND1_16W( B, C, D, E, A, F16W_2,  7, in[12], 2 );
+   ROUND1_16W( A, B, C, D, E, F16W_2, 12, in[ 0], 2 );
+   ROUND1_16W( E, A, B, C, D, F16W_2, 15, in[ 9], 2 );
+   ROUND1_16W( D, E, A, B, C, F16W_2,  9, in[ 5], 2 );
+   ROUND1_16W( C, D, E, A, B, F16W_2, 11, in[ 2], 2 );
+   ROUND1_16W( B, C, D, E, A, F16W_2,  7, in[14], 2 );
+   ROUND1_16W( A, B, C, D, E, F16W_2, 13, in[11], 2 );
+   ROUND1_16W( E, A, B, C, D, F16W_2, 12, in[ 8], 2 );
+
+   ROUND1_16W( D, E, A, B, C, F16W_3, 11, in[ 3], 3 );
+   ROUND1_16W( C, D, E, A, B, F16W_3, 13, in[10], 3 );
+   ROUND1_16W( B, C, D, E, A, F16W_3,  6, in[14], 3 );
+   ROUND1_16W( A, B, C, D, E, F16W_3,  7, in[ 4], 3 );
+   ROUND1_16W( E, A, B, C, D, F16W_3, 14, in[ 9], 3 );
+   ROUND1_16W( D, E, A, B, C, F16W_3,  9, in[15], 3 );
+   ROUND1_16W( C, D, E, A, B, F16W_3, 13, in[ 8], 3 );
+   ROUND1_16W( B, C, D, E, A, F16W_3, 15, in[ 1], 3 );
+   ROUND1_16W( A, B, C, D, E, F16W_3, 14, in[ 2], 3 );
+   ROUND1_16W( E, A, B, C, D, F16W_3,  8, in[ 7], 3 );
+   ROUND1_16W( D, E, A, B, C, F16W_3, 13, in[ 0], 3 );
+   ROUND1_16W( C, D, E, A, B, F16W_3,  6, in[ 6], 3 );
+   ROUND1_16W( B, C, D, E, A, F16W_3,  5, in[13], 3 );
+   ROUND1_16W( A, B, C, D, E, F16W_3, 12, in[11], 3 );
+   ROUND1_16W( E, A, B, C, D, F16W_3,  7, in[ 5], 3 );
+   ROUND1_16W( D, E, A, B, C, F16W_3,  5, in[12], 3 );
+
+   ROUND1_16W( C, D, E, A, B, F16W_4, 11, in[ 1], 4 );
+   ROUND1_16W( B, C, D, E, A, F16W_4, 12, in[ 9], 4 );
+   ROUND1_16W( A, B, C, D, E, F16W_4, 14, in[11], 4 );
+   ROUND1_16W( E, A, B, C, D, F16W_4, 15, in[10], 4 );
+   ROUND1_16W( D, E, A, B, C, F16W_4, 14, in[ 0], 4 );
+   ROUND1_16W( C, D, E, A, B, F16W_4, 15, in[ 8], 4 );
+   ROUND1_16W( B, C, D, E, A, F16W_4,  9, in[12], 4 );
+   ROUND1_16W( A, B, C, D, E, F16W_4,  8, in[ 4], 4 );
+   ROUND1_16W( E, A, B, C, D, F16W_4,  9, in[13], 4 );
+   ROUND1_16W( D, E, A, B, C, F16W_4, 14, in[ 3], 4 );
+   ROUND1_16W( C, D, E, A, B, F16W_4,  5, in[ 7], 4 );
+   ROUND1_16W( B, C, D, E, A, F16W_4,  6, in[15], 4 );
+   ROUND1_16W( A, B, C, D, E, F16W_4,  8, in[14], 4 );
+   ROUND1_16W( E, A, B, C, D, F16W_4,  6, in[ 5], 4 );
+   ROUND1_16W( D, E, A, B, C, F16W_4,  5, in[ 6], 4 );
+   ROUND1_16W( C, D, E, A, B, F16W_4, 12, in[ 2], 4 );
+
+   ROUND1_16W( B, C, D, E, A, F16W_5,  9, in[ 4], 5 );
+   ROUND1_16W( A, B, C, D, E, F16W_5, 15, in[ 0], 5 );
+   ROUND1_16W( E, A, B, C, D, F16W_5,  5, in[ 5], 5 );
+   ROUND1_16W( D, E, A, B, C, F16W_5, 11, in[ 9], 5 );
+   ROUND1_16W( C, D, E, A, B, F16W_5,  6, in[ 7], 5 );
+   ROUND1_16W( B, C, D, E, A, F16W_5,  8, in[12], 5 );
+   ROUND1_16W( A, B, C, D, E, F16W_5, 13, in[ 2], 5 );
+   ROUND1_16W( E, A, B, C, D, F16W_5, 12, in[10], 5 );
+   ROUND1_16W( D, E, A, B, C, F16W_5,  5, in[14], 5 );
+   ROUND1_16W( C, D, E, A, B, F16W_5, 12, in[ 1], 5 );
+   ROUND1_16W( B, C, D, E, A, F16W_5, 13, in[ 3], 5 );
+   ROUND1_16W( A, B, C, D, E, F16W_5, 14, in[ 8], 5 );
+   ROUND1_16W( E, A, B, C, D, F16W_5, 11, in[11], 5 );
+   ROUND1_16W( D, E, A, B, C, F16W_5,  8, in[ 6], 5 );
+   ROUND1_16W( C, D, E, A, B, F16W_5,  5, in[15], 5 );
+   ROUND1_16W( B, C, D, E, A, F16W_5,  6, in[13], 5 );
+
+   ROUND2_16W( A, B, C, D, E, F16W_5,  8, in[ 5], 1 );
+   ROUND2_16W( E, A, B, C, D, F16W_5,  9, in[14], 1 );
+   ROUND2_16W( D, E, A, B, C, F16W_5,  9, in[ 7], 1 );
+   ROUND2_16W( C, D, E, A, B, F16W_5, 11, in[ 0], 1 );
+   ROUND2_16W( B, C, D, E, A, F16W_5, 13, in[ 9], 1 );
+   ROUND2_16W( A, B, C, D, E, F16W_5, 15, in[ 2], 1 );
+   ROUND2_16W( E, A, B, C, D, F16W_5, 15, in[11], 1 );
+   ROUND2_16W( D, E, A, B, C, F16W_5,  5, in[ 4], 1 );
+   ROUND2_16W( C, D, E, A, B, F16W_5,  7, in[13], 1 );
+   ROUND2_16W( B, C, D, E, A, F16W_5,  7, in[ 6], 1 );
+   ROUND2_16W( A, B, C, D, E, F16W_5,  8, in[15], 1 );
+   ROUND2_16W( E, A, B, C, D, F16W_5, 11, in[ 8], 1 );
+   ROUND2_16W( D, E, A, B, C, F16W_5, 14, in[ 1], 1 );
+   ROUND2_16W( C, D, E, A, B, F16W_5, 14, in[10], 1 );
+   ROUND2_16W( B, C, D, E, A, F16W_5, 12, in[ 3], 1 );
+   ROUND2_16W( A, B, C, D, E, F16W_5,  6, in[12], 1 );
+
+   ROUND2_16W( E, A, B, C, D, F16W_4,  9, in[ 6], 2 );
+   ROUND2_16W( D, E, A, B, C, F16W_4, 13, in[11], 2 );
+   ROUND2_16W( C, D, E, A, B, F16W_4, 15, in[ 3], 2 );
+   ROUND2_16W( B, C, D, E, A, F16W_4,  7, in[ 7], 2 );
+   ROUND2_16W( A, B, C, D, E, F16W_4, 12, in[ 0], 2 );
+   ROUND2_16W( E, A, B, C, D, F16W_4,  8, in[13], 2 );
+   ROUND2_16W( D, E, A, B, C, F16W_4,  9, in[ 5], 2 );
+   ROUND2_16W( C, D, E, A, B, F16W_4, 11, in[10], 2 );
+   ROUND2_16W( B, C, D, E, A, F16W_4,  7, in[14], 2 );
+   ROUND2_16W( A, B, C, D, E, F16W_4,  7, in[15], 2 );
+   ROUND2_16W( E, A, B, C, D, F16W_4, 12, in[ 8], 2 );
+   ROUND2_16W( D, E, A, B, C, F16W_4,  7, in[12], 2 );
+   ROUND2_16W( C, D, E, A, B, F16W_4,  6, in[ 4], 2 );
+   ROUND2_16W( B, C, D, E, A, F16W_4, 15, in[ 9], 2 );
+   ROUND2_16W( A, B, C, D, E, F16W_4, 13, in[ 1], 2 );
+   ROUND2_16W( E, A, B, C, D, F16W_4, 11, in[ 2], 2 );
+
+   ROUND2_16W( D, E, A, B, C, F16W_3,  9, in[15], 3 );
+   ROUND2_16W( C, D, E, A, B, F16W_3,  7, in[ 5], 3 );
+   ROUND2_16W( B, C, D, E, A, F16W_3, 15, in[ 1], 3 );
+   ROUND2_16W( A, B, C, D, E, F16W_3, 11, in[ 3], 3 );
+   ROUND2_16W( E, A, B, C, D, F16W_3,  8, in[ 7], 3 );
+   ROUND2_16W( D, E, A, B, C, F16W_3,  6, in[14], 3 );
+   ROUND2_16W( C, D, E, A, B, F16W_3,  6, in[ 6], 3 );
+   ROUND2_16W( B, C, D, E, A, F16W_3, 14, in[ 9], 3 );
+   ROUND2_16W( A, B, C, D, E, F16W_3, 12, in[11], 3 );
+   ROUND2_16W( E, A, B, C, D, F16W_3, 13, in[ 8], 3 );
+   ROUND2_16W( D, E, A, B, C, F16W_3,  5, in[12], 3 );
+   ROUND2_16W( C, D, E, A, B, F16W_3, 14, in[ 2], 3 );
+   ROUND2_16W( B, C, D, E, A, F16W_3, 13, in[10], 3 );
+   ROUND2_16W( A, B, C, D, E, F16W_3, 13, in[ 0], 3 );
+   ROUND2_16W( E, A, B, C, D, F16W_3,  7, in[ 4], 3 );
+   ROUND2_16W( D, E, A, B, C, F16W_3,  5, in[13], 3 );
+
+   ROUND2_16W( C, D, E, A, B, F16W_2, 15, in[ 8], 4 );
+   ROUND2_16W( B, C, D, E, A, F16W_2,  5, in[ 6], 4 );
+   ROUND2_16W( A, B, C, D, E, F16W_2,  8, in[ 4], 4 );
+   ROUND2_16W( E, A, B, C, D, F16W_2, 11, in[ 1], 4 );
+   ROUND2_16W( D, E, A, B, C, F16W_2, 14, in[ 3], 4 );
+   ROUND2_16W( C, D, E, A, B, F16W_2, 14, in[11], 4 );
+   ROUND2_16W( B, C, D, E, A, F16W_2,  6, in[15], 4 );
+   ROUND2_16W( A, B, C, D, E, F16W_2, 14, in[ 0], 4 );
+   ROUND2_16W( E, A, B, C, D, F16W_2,  6, in[ 5], 4 );
+   ROUND2_16W( D, E, A, B, C, F16W_2,  9, in[12], 4 );
+   ROUND2_16W( C, D, E, A, B, F16W_2, 12, in[ 2], 4 );
+   ROUND2_16W( B, C, D, E, A, F16W_2,  9, in[13], 4 );
+   ROUND2_16W( A, B, C, D, E, F16W_2, 12, in[ 9], 4 );
+   ROUND2_16W( E, A, B, C, D, F16W_2,  5, in[ 7], 4 );
+   ROUND2_16W( D, E, A, B, C, F16W_2, 15, in[10], 4 );
+   ROUND2_16W( C, D, E, A, B, F16W_2,  8, in[14], 4 );
+
+   ROUND2_16W( B, C, D, E, A, F16W_1,  8, in[12], 5 );
+   ROUND2_16W( A, B, C, D, E, F16W_1,  5, in[15], 5 );
+   ROUND2_16W( E, A, B, C, D, F16W_1, 12, in[10], 5 );
+   ROUND2_16W( D, E, A, B, C, F16W_1,  9, in[ 4], 5 );
+   ROUND2_16W( C, D, E, A, B, F16W_1, 12, in[ 1], 5 );
+   ROUND2_16W( B, C, D, E, A, F16W_1,  5, in[ 5], 5 );
+   ROUND2_16W( A, B, C, D, E, F16W_1, 14, in[ 8], 5 );
+   ROUND2_16W( E, A, B, C, D, F16W_1,  6, in[ 7], 5 );
+   ROUND2_16W( D, E, A, B, C, F16W_1,  8, in[ 6], 5 );
+   ROUND2_16W( C, D, E, A, B, F16W_1, 13, in[ 2], 5 );
+   ROUND2_16W( B, C, D, E, A, F16W_1,  6, in[13], 5 );
+   ROUND2_16W( A, B, C, D, E, F16W_1,  5, in[14], 5 );
+   ROUND2_16W( E, A, B, C, D, F16W_1, 15, in[ 0], 5 );
+   ROUND2_16W( D, E, A, B, C, F16W_1, 13, in[ 3], 5 );
+   ROUND2_16W( C, D, E, A, B, F16W_1, 11, in[ 9], 5 );
+   ROUND2_16W( B, C, D, E, A, F16W_1, 11, in[11], 5 );
+
+   tmp =  _mm512_add_epi32( _mm512_add_epi32( h[1], C1 ), D2 );
+   h[1] = _mm512_add_epi32( _mm512_add_epi32( h[2], D1 ), E2 );
+   h[2] = _mm512_add_epi32( _mm512_add_epi32( h[3], E1 ), A2 );
+   h[3] = _mm512_add_epi32( _mm512_add_epi32( h[4], A1 ), B2 );
+   h[4] = _mm512_add_epi32( _mm512_add_epi32( h[0], B1 ), C2 );
+   h[0] = tmp;
+}
+
+void ripemd160_16way_init( ripemd160_16way_context *sc )
+{
+   sc->val[0] = m512_const1_64( 0x6745230167452301 );
+   sc->val[1] = m512_const1_64( 0xEFCDAB89EFCDAB89 );
+   sc->val[2] = m512_const1_64( 0x98BADCFE98BADCFE );
+   sc->val[3] = m512_const1_64( 0x1032547610325476 );
+   sc->val[4] = m512_const1_64( 0xC3D2E1F0C3D2E1F0 );
+   sc->count_high = sc->count_low = 0;
+}
+
+void ripemd160_16way( ripemd160_16way_context *sc, const void *data,
+                      size_t len )
+{
+   __m512i *vdata = (__m512i*)data;
+   size_t ptr;
+   const int block_size = 64;
+
+   ptr = (unsigned)sc->count_low & (block_size - 1U);
+   while ( len > 0 )
+   {
+      size_t clen;
+      uint32_t clow, clow2;
+
+      clen = block_size - ptr;
+      if ( clen > len )
+         clen = len;
+      memcpy_512( sc->buf + (ptr>>2), vdata, clen>>2 );
+      vdata = vdata + (clen>>2);
+      ptr += clen;
+      len -= clen;
+      if ( ptr == block_size )
+      {
+         ripemd160_16way_round( sc );
+         ptr = 0;
+      }
+      clow = sc->count_low;
+      clow2 = clow + clen;
+      sc->count_low = clow2;
+      if ( clow2 < clow )
+         sc->count_high++;
+   }
+}
+
+void ripemd160_16way_close( ripemd160_16way_context  *sc, void *dst )
+{
+   unsigned ptr, u;
+   uint32_t low, high;
+   const int block_size = 64;
+   const int pad = block_size - 8;
+
+   ptr = (unsigned)sc->count_low & ( block_size - 1U);
+   sc->buf[ ptr>>2 ] = m512_const1_32( 0x80 );
+   ptr += 4;
+
+   if ( ptr > pad )
+   {
+       memset_zero_512( sc->buf + (ptr>>2), (block_size - ptr) >> 2 );
+       ripemd160_16way_round( sc );
+       memset_zero_512( sc->buf, pad>>2 );
+   }
+   else
+       memset_zero_512( sc->buf + (ptr>>2), (pad - ptr) >> 2 );
+
+    low = sc->count_low;
+    high = (sc->count_high << 3) | (low >> 29);
+    low = low << 3;
+    sc->buf[  pad>>2      ] = _mm512_set1_epi32( low  );
+    sc->buf[ (pad>>2) + 1 ] = _mm512_set1_epi32( high );
+    ripemd160_16way_round( sc );
+    for (u = 0; u < 5; u ++)
+        casti_m512i( dst, u ) = sc->val[u];
+}
+
+#endif  // AVX512

algo/ripemd/ripemd-hash-4way.h

@@ -32,7 +32,21 @@ void ripemd160_8way_init( ripemd160_8way_context *sc );
 void ripemd160_8way( ripemd160_8way_context *sc, const void *data, size_t len );
 void ripemd160_8way_close( ripemd160_8way_context *sc, void *dst );
 
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
 
+typedef struct
+{
+   __m512i buf[64>>2];
+   __m512i val[5];
+   uint32_t count_high, count_low;
+} __attribute__ ((aligned (128))) ripemd160_16way_context;
+
+void ripemd160_16way_init( ripemd160_16way_context *sc );
+void ripemd160_16way( ripemd160_16way_context *sc, const void *data,
+                      size_t len );
+void ripemd160_16way_close( ripemd160_16way_context *sc, void *dst );
+
+#endif // AVX512
 #endif // __AVX2__
 #endif // __SSE4_2__
 #endif // RIPEMD_HASH_4WAY_H__

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

@@ -56,7 +56,7 @@ typedef struct {
    __m128i val[8];
    uint32_t count_high, count_low;
    bool initialized;
-} sha256_4way_context;
+} sha256_4way_context __attribute__ ((aligned (64)));
 
 void sha256_4way_init( sha256_4way_context *sc );
 void sha256_4way( sha256_4way_context *sc, const void *data, size_t len );
@@ -71,10 +71,11 @@ typedef struct {
    __m256i val[8];
    uint32_t count_high, count_low;
    bool initialized;
-} sha256_8way_context;
+} sha256_8way_context __attribute__ ((aligned (128)));
 
 void sha256_8way_init( sha256_8way_context *sc );
-void sha256_8way( sha256_8way_context *sc, const void *data, size_t len );
+void sha256_8way_update( sha256_8way_context *sc, const void *data, size_t len );
+#define sha256_8way sha256_8way_update
 void sha256_8way_close( sha256_8way_context *sc, void *dst );
 
 //#define SPH_SIZE_sha512   512
@@ -86,30 +87,32 @@ typedef struct {
    __m256i val[8];
    uint64_t count;
    bool initialized;
-} sha512_4way_context;
+} sha512_4way_context __attribute__ ((aligned (128)));
 
 void sha512_4way_init( sha512_4way_context *sc);
-void sha512_4way( sha512_4way_context *sc, const void *data, size_t len );
+void sha512_4way_update( sha512_4way_context *sc, const void *data,
+                         size_t len );
+#define sha512_4way sha512_4way_update
 void sha512_4way_close( sha512_4way_context *sc, void *dst );
 
-// SHA-256 11 way hybrid
-// Combines AVX2, MMX and scalar data to do 8 + 2 + 1 parallel.
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// SHA-512 8 way
+
 typedef struct {
-   __m256i  bufx[64>>2];
-   __m256i  valx[8];
-   __m64    bufy[64>>2];
-   __m64    valy[8];
-   uint32_t bufz[64>>2];
-   uint32_t valz[8];
-   uint32_t count_high, count_low;
-} sha256_11way_context;
+   __m512i buf[128>>3];
+   __m512i val[8];
+   uint64_t count;
+   bool initialized;
+} sha512_8way_context __attribute__ ((aligned (128)));
 
-void sha256_11way_init( sha256_11way_context *ctx );
-void sha256_11way_update( sha256_11way_context *ctx, const void *datax,
-	                 const void *datay, const void *dataz, size_t len );
-void sha256_11way_close( sha256_11way_context *ctx, void *dstx, void *dstyx,
-	                 void *dstz  );
+void sha512_8way_init( sha512_8way_context *sc);
+void sha512_8way_update( sha512_8way_context *sc, const void *data, 
+                         size_t len );
+void sha512_8way_close( sha512_8way_context *sc, void *dst );
 
+
+#endif  // AVX512
 #endif  // __AVX2__
 #endif  // __SSE2__
 #endif  // SHA256_4WAY_H__

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 );
 

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

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)));

algo/sha/sha512-hash-4way.c

@@ -36,8 +36,6 @@
 #include <string.h>
 #include "sha-hash-4way.h"
 
-// SHA-512 4 way 64 bit
-
 /*
 static const sph_u64 H512[8] = {
         SPH_C64(0x6A09E667F3BCC908), SPH_C64(0xBB67AE8584CAA73B),
@@ -90,6 +88,236 @@ static const sph_u64 K512[80] = {
 	SPH_C64(0x5FCB6FAB3AD6FAEC), SPH_C64(0x6C44198C4A475817)
 };
 
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+// SHA-512 8 way 64 bit
+
+#define CH8W(X, Y, Z) \
+   _mm512_xor_si512( _mm512_and_si512( _mm512_xor_si512( Y, Z ), X ), Z ) 
+
+#define MAJ8W(X, Y, Z) \
+   _mm512_or_si512( _mm512_and_si512( X, Y ), \
+                    _mm512_and_si512( _mm512_or_si512( X, Y ), Z ) )
+
+#define BSG8W_5_0(x) \
+   _mm512_xor_si512( _mm512_xor_si512( \
+        mm512_ror_64(x, 28), mm512_ror_64(x, 34) ), mm512_ror_64(x, 39) )
+
+#define BSG8W_5_1(x) \
+   _mm512_xor_si512( _mm512_xor_si512( \
+        mm512_ror_64(x, 14), mm512_ror_64(x, 18) ), mm512_ror_64(x, 41) )
+
+#define SSG8W_5_0(x) \
+   _mm512_xor_si512( _mm512_xor_si512( \
+        mm512_ror_64(x,  1), mm512_ror_64(x,  8) ), _mm512_srli_epi64(x, 7) ) 
+
+#define SSG8W_5_1(x) \
+   _mm512_xor_si512( _mm512_xor_si512( \
+        mm512_ror_64(x, 19), mm512_ror_64(x, 61) ), _mm512_srli_epi64(x, 6) )
+
+static inline __m512i ssg8w_512_add( __m512i w0, __m512i w1 )
+{
+   __m512i w0a, w1a, w0b, w1b;
+   w0a = mm512_ror_64( w0, 1 );
+   w1a = mm512_ror_64( w1,19 );
+   w0b = mm512_ror_64( w0, 8 );
+   w1b = mm512_ror_64( w1,61 );
+   w0a = _mm512_xor_si512( w0a, w0b );
+   w1a = _mm512_xor_si512( w1a, w1b );
+   w0b = _mm512_srli_epi64( w0, 7 );
+   w1b = _mm512_srli_epi64( w1, 6 );
+   w0a = _mm512_xor_si512( w0a, w0b );
+   w1a = _mm512_xor_si512( w1a, w1b );
+   return _mm512_add_epi64( w0a, w1a );
+}
+
+
+#define SSG8W_512x2_0( w0, w1, i ) do \
+{ \
+   __m512i X0a, X1a, X0b, X1b; \
+  X0a = mm512_ror_64( W[i-15], 1 ); \
+  X1a = mm512_ror_64( W[i-14], 1 ); \
+  X0b = mm512_ror_64( W[i-15], 8 ); \
+  X1b = mm512_ror_64( W[i-14], 8 ); \
+  X0a = _mm512_xor_si512( X0a, X0b ); \
+  X1a = _mm512_xor_si512( X1a, X1b ); \
+  X0b = _mm512_srli_epi64( W[i-15], 7 ); \
+  X1b = _mm512_srli_epi64( W[i-14], 7 ); \
+  w0  = _mm512_xor_si512( X0a, X0b ); \
+  w1  = _mm512_xor_si512( X1a, X1b ); \
+} while(0)
+
+#define SSG8W_512x2_1( w0, w1, i ) do \
+{ \
+   __m512i X0a, X1a, X0b, X1b; \
+  X0a = mm512_ror_64( W[i-2],19 ); \
+  X1a = mm512_ror_64( W[i-1],19 ); \
+  X0b = mm512_ror_64( W[i-2],61 ); \
+  X1b = mm512_ror_64( W[i-1],61 ); \
+  X0a = _mm512_xor_si512( X0a, X0b ); \
+  X1a = _mm512_xor_si512( X1a, X1b ); \
+  X0b = _mm512_srli_epi64( W[i-2], 6 ); \
+  X1b = _mm512_srli_epi64( W[i-1], 6 ); \
+  w0  = _mm512_xor_si512( X0a, X0b ); \
+  w1  = _mm512_xor_si512( X1a, X1b ); \
+} while(0)
+
+#define SHA3_8WAY_STEP(A, B, C, D, E, F, G, H, i) \
+do { \
+  __m512i T1, T2; \
+  __m512i K = _mm512_set1_epi64( K512[ i ] ); \
+  T1 = _mm512_add_epi64( H, mm512_add4_64( BSG8W_5_1(E), CH8W(E, F, G), \
+                                           K, W[i] ) ); \
+  T2 = _mm512_add_epi64( BSG8W_5_0(A), MAJ8W(A, B, C) ); \
+  D  = _mm512_add_epi64( D, T1 ); \
+  H  = _mm512_add_epi64( T1, T2 ); \
+} while (0)
+
+static void
+sha512_8way_round( sha512_8way_context *ctx,  __m512i *in, __m512i r[8] )
+{
+   int i;
+   register __m512i A, B, C, D, E, F, G, H;
+   __m512i W[80];
+
+   mm512_block_bswap_64( W  , in );
+   mm512_block_bswap_64( W+8, in+8 );
+
+   for ( i = 16; i < 80; i++ )
+      W[i] = _mm512_add_epi64( ssg8w_512_add( W[i-15], W[i-2] ),
+                               _mm512_add_epi64( W[ i- 7 ], W[ i-16 ] ) );
+
+   if ( ctx->initialized )
+   {
+      A = r[0];
+      B = r[1];
+      C = r[2];
+      D = r[3];
+      E = r[4];
+      F = r[5];
+      G = r[6];
+      H = r[7];
+   }
+   else
+   {
+      A = m512_const1_64( 0x6A09E667F3BCC908 );
+      B = m512_const1_64( 0xBB67AE8584CAA73B );
+      C = m512_const1_64( 0x3C6EF372FE94F82B );
+      D = m512_const1_64( 0xA54FF53A5F1D36F1 );
+      E = m512_const1_64( 0x510E527FADE682D1 );
+      F = m512_const1_64( 0x9B05688C2B3E6C1F );
+      G = m512_const1_64( 0x1F83D9ABFB41BD6B );
+      H = m512_const1_64( 0x5BE0CD19137E2179 );
+   }
+
+   for ( i = 0; i < 80; i += 8 )
+   {
+      SHA3_8WAY_STEP( A, B, C, D, E, F, G, H, i + 0 );
+      SHA3_8WAY_STEP( H, A, B, C, D, E, F, G, i + 1 );
+      SHA3_8WAY_STEP( G, H, A, B, C, D, E, F, i + 2 );
+      SHA3_8WAY_STEP( F, G, H, A, B, C, D, E, i + 3 );
+      SHA3_8WAY_STEP( E, F, G, H, A, B, C, D, i + 4 );
+      SHA3_8WAY_STEP( D, E, F, G, H, A, B, C, i + 5 );
+      SHA3_8WAY_STEP( C, D, E, F, G, H, A, B, i + 6 );
+      SHA3_8WAY_STEP( B, C, D, E, F, G, H, A, i + 7 );
+   }
+
+   if ( ctx->initialized )
+   {
+      r[0] = _mm512_add_epi64( r[0], A );
+      r[1] = _mm512_add_epi64( r[1], B );
+      r[2] = _mm512_add_epi64( r[2], C );
+      r[3] = _mm512_add_epi64( r[3], D );
+      r[4] = _mm512_add_epi64( r[4], E );
+      r[5] = _mm512_add_epi64( r[5], F );
+      r[6] = _mm512_add_epi64( r[6], G );
+      r[7] = _mm512_add_epi64( r[7], H );
+   }
+   else
+   {
+      ctx->initialized = true;
+      r[0] = _mm512_add_epi64( A, m512_const1_64( 0x6A09E667F3BCC908 ) );
+      r[1] = _mm512_add_epi64( B, m512_const1_64( 0xBB67AE8584CAA73B ) );
+      r[2] = _mm512_add_epi64( C, m512_const1_64( 0x3C6EF372FE94F82B ) );
+      r[3] = _mm512_add_epi64( D, m512_const1_64( 0xA54FF53A5F1D36F1 ) );
+      r[4] = _mm512_add_epi64( E, m512_const1_64( 0x510E527FADE682D1 ) );
+      r[5] = _mm512_add_epi64( F, m512_const1_64( 0x9B05688C2B3E6C1F ) );
+      r[6] = _mm512_add_epi64( G, m512_const1_64( 0x1F83D9ABFB41BD6B ) );
+      r[7] = _mm512_add_epi64( H, m512_const1_64( 0x5BE0CD19137E2179 ) );
+   }
+}
+
+void sha512_8way_init( sha512_8way_context *sc )
+{
+   sc->initialized = false;
+   sc->count = 0;
+}
+
+void sha512_8way_update( sha512_8way_context *sc, const void *data, size_t len )
+{
+   __m512i *vdata = (__m512i*)data;
+   size_t ptr;
+   const int buf_size = 128;
+
+   ptr = (unsigned)sc->count & (buf_size - 1U);
+   while ( len > 0 )
+   {
+      size_t clen;
+      clen = buf_size - ptr;
+      if ( clen > len )
+         clen = len;
+      memcpy_512( sc->buf + (ptr>>3), vdata, clen>>3 );
+      vdata = vdata + (clen>>3);
+      ptr += clen;
+      len -= clen;
+      if ( ptr == buf_size )
+      {
+         sha512_8way_round( sc, sc->buf, sc->val );
+         ptr = 0;
+      }
+      sc->count += clen;
+   }
+}
+
+void sha512_8way_close( sha512_8way_context *sc, void *dst )
+{
+    unsigned ptr;
+    const int buf_size = 128;
+    const int pad = buf_size - 16;
+    const __m512i shuff_bswap64 = m512_const_64(
+                                    0x38393a3b3c3d3e3f, 0x3031323334353637,
+                                    0x28292a2b2c2d2e2f, 0x2021222324252627,
+                                    0x18191a1b1c1d1e1f, 0x1011121314151617,
+                                    0x08090a0b0c0d0e0f, 0x0001020304050607 );
+
+    ptr = (unsigned)sc->count & (buf_size - 1U);
+    sc->buf[ ptr>>3 ] = m512_const1_64( 0x80 );
+    ptr += 8;
+    if ( ptr > pad )
+    {
+         memset_zero_512( sc->buf + (ptr>>3), (buf_size - ptr) >> 3 );
+         sha512_8way_round( sc, sc->buf, sc->val );
+         memset_zero_512( sc->buf, pad >> 3 );
+    }
+    else
+         memset_zero_512( sc->buf + (ptr>>3), (pad - ptr) >> 3 );
+
+    sc->buf[ pad >> 3 ] = _mm512_shuffle_epi8(
+                       _mm512_set1_epi64( sc->count >> 61 ), shuff_bswap64 );
+    sc->buf[ ( pad+8 ) >> 3 ] = _mm512_shuffle_epi8(
+                       _mm512_set1_epi64( sc->count <<  3 ), shuff_bswap64 );
+    sha512_8way_round( sc, sc->buf, sc->val );
+
+    mm512_block_bswap_64( dst, sc->val );
+}
+
+
+#endif   // AVX512
+
+// SHA-512 4 way 64 bit
+
+
 #define CH(X, Y, Z) \
    _mm256_xor_si256( _mm256_and_si256( _mm256_xor_si256( Y, Z ), X ), Z ) 
 
@@ -252,19 +480,9 @@ 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 )
+void sha512_4way_update( sha512_4way_context *sc, const void *data, size_t len )
 {
    __m256i *vdata = (__m256i*)data;
    size_t ptr;
@@ -295,6 +513,10 @@ 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_const_64( 0x18191a1b1c1d1e1f,
+                                                 0x1011121314151617,
+                                                 0x08090a0b0c0d0e0f,
+                                                 0x0001020304050607 );
 
     ptr = (unsigned)sc->count & (buf_size - 1U);
     sc->buf[ ptr>>3 ] = m256_const1_64( 0x80 );
@@ -308,10 +530,10 @@ 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 );

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

@@ -33,7 +33,7 @@
 #include <stddef.h>
 #include <string.h>
 
-#ifdef __AVX2__
+#ifdef __SSE4_1__
 
 #include "shabal-hash-4way.h"
 #ifdef __cplusplus
@@ -58,6 +58,599 @@ extern "C"{
 #define O2    9
 #define O3    6
 
+
+#if defined(__AVX2__)
+
+#define DECL_STATE8   \
+   __m256i A00, A01, A02, A03, A04, A05, A06, A07, \
+           A08, A09, A0A, A0B; \
+   __m256i B0, B1, B2, B3, B4, B5, B6, B7, \
+           B8, B9, BA, BB, BC, BD, BE, BF; \
+   __m256i C0, C1, C2, C3, C4, C5, C6, C7, \
+           C8, C9, CA, CB, CC, CD, CE, CF; \
+   __m256i M0, M1, M2, M3, M4, M5, M6, M7, \
+           M8, M9, MA, MB, MC, MD, ME, MF; \
+   sph_u32 Wlow, Whigh;
+
+#define READ_STATE8(state) do \
+{ \
+   if ( (state)->state_loaded ) \
+   { \
+      A00 = (state)->A[0]; \
+      A01 = (state)->A[1]; \
+      A02 = (state)->A[2]; \
+      A03 = (state)->A[3]; \
+      A04 = (state)->A[4]; \
+      A05 = (state)->A[5]; \
+      A06 = (state)->A[6]; \
+      A07 = (state)->A[7]; \
+      A08 = (state)->A[8]; \
+      A09 = (state)->A[9]; \
+      A0A = (state)->A[10]; \
+      A0B = (state)->A[11]; \
+      B0 = (state)->B[0]; \
+      B1 = (state)->B[1]; \
+      B2 = (state)->B[2]; \
+      B3 = (state)->B[3]; \
+      B4 = (state)->B[4]; \
+      B5 = (state)->B[5]; \
+      B6 = (state)->B[6]; \
+      B7 = (state)->B[7]; \
+      B8 = (state)->B[8]; \
+      B9 = (state)->B[9]; \
+      BA = (state)->B[10]; \
+      BB = (state)->B[11]; \
+      BC = (state)->B[12]; \
+      BD = (state)->B[13]; \
+      BE = (state)->B[14]; \
+      BF = (state)->B[15]; \
+      C0 = (state)->C[0]; \
+      C1 = (state)->C[1]; \
+      C2 = (state)->C[2]; \
+      C3 = (state)->C[3]; \
+      C4 = (state)->C[4]; \
+      C5 = (state)->C[5]; \
+      C6 = (state)->C[6]; \
+      C7 = (state)->C[7]; \
+      C8 = (state)->C[8]; \
+      C9 = (state)->C[9]; \
+      CA = (state)->C[10]; \
+      CB = (state)->C[11]; \
+      CC = (state)->C[12]; \
+      CD = (state)->C[13]; \
+      CE = (state)->C[14]; \
+      CF = (state)->C[15]; \
+   } \
+   else \
+   { \
+       (state)->state_loaded = true; \
+       A00 = m256_const1_64( 0x20728DFD20728DFD ); \
+       A01 = m256_const1_64( 0x46C0BD5346C0BD53 ); \
+       A02 = m256_const1_64( 0xE782B699E782B699 ); \
+       A03 = m256_const1_64( 0x5530463255304632 ); \
+       A04 = m256_const1_64( 0x71B4EF9071B4EF90 ); \
+       A05 = m256_const1_64( 0x0EA9E82C0EA9E82C ); \
+       A06 = m256_const1_64( 0xDBB930F1DBB930F1 ); \
+       A07 = m256_const1_64( 0xFAD06B8BFAD06B8B ); \
+       A08 = m256_const1_64( 0xBE0CAE40BE0CAE40 ); \
+       A09 = m256_const1_64( 0x8BD144108BD14410 ); \
+       A0A = m256_const1_64( 0x76D2ADAC76D2ADAC ); \
+       A0B = m256_const1_64( 0x28ACAB7F28ACAB7F ); \
+       B0 = m256_const1_64( 0xC1099CB7C1099CB7 ); \
+       B1 = m256_const1_64( 0x07B385F307B385F3 ); \
+       B2 = m256_const1_64( 0xE7442C26E7442C26 ); \
+       B3 = m256_const1_64( 0xCC8AD640CC8AD640 ); \
+       B4 = m256_const1_64( 0xEB6F56C7EB6F56C7 ); \
+       B5 = m256_const1_64( 0x1EA81AA91EA81AA9 ); \
+       B6 = m256_const1_64( 0x73B9D31473B9D314 ); \
+       B7 = m256_const1_64( 0x1DE85D081DE85D08 ); \
+       B8 = m256_const1_64( 0x48910A5A48910A5A ); \
+       B9 = m256_const1_64( 0x893B22DB893B22DB ); \
+       BA = m256_const1_64( 0xC5A0DF44C5A0DF44 ); \
+       BB = m256_const1_64( 0xBBC4324EBBC4324E ); \
+       BC = m256_const1_64( 0x72D2F24072D2F240 ); \
+       BD = m256_const1_64( 0x75941D9975941D99 ); \
+       BE = m256_const1_64( 0x6D8BDE826D8BDE82 ); \
+       BF = m256_const1_64( 0xA1A7502BA1A7502B ); \
+       C0 = m256_const1_64( 0xD9BF68D1D9BF68D1 ); \
+       C1 = m256_const1_64( 0x58BAD75058BAD750 ); \
+       C2 = m256_const1_64( 0x56028CB256028CB2 ); \
+       C3 = m256_const1_64( 0x8134F3598134F359 ); \
+       C4 = m256_const1_64( 0xB5D469D8B5D469D8 ); \
+       C5 = m256_const1_64( 0x941A8CC2941A8CC2 ); \
+       C6 = m256_const1_64( 0x418B2A6E418B2A6E ); \
+       C7 = m256_const1_64( 0x0405278004052780 ); \
+       C8 = m256_const1_64( 0x7F07D7877F07D787 ); \
+       C9 = m256_const1_64( 0x5194358F5194358F ); \
+       CA = m256_const1_64( 0x3C60D6653C60D665 ); \
+       CB = m256_const1_64( 0xBE97D79ABE97D79A ); \
+       CC = m256_const1_64( 0x950C3434950C3434 ); \
+       CD = m256_const1_64( 0xAED9A06DAED9A06D ); \
+       CE = m256_const1_64( 0x2537DC8D2537DC8D ); \
+       CF = m256_const1_64( 0x7CDB59697CDB5969 ); \
+   } \
+   Wlow = (state)->Wlow; \
+   Whigh = (state)->Whigh; \
+} while (0)
+
+#define WRITE_STATE8(state)   do { \
+      (state)->A[0] = A00; \
+      (state)->A[1] = A01; \
+      (state)->A[2] = A02; \
+      (state)->A[3] = A03; \
+      (state)->A[4] = A04; \
+      (state)->A[5] = A05; \
+      (state)->A[6] = A06; \
+      (state)->A[7] = A07; \
+      (state)->A[8] = A08; \
+      (state)->A[9] = A09; \
+      (state)->A[10] = A0A; \
+      (state)->A[11] = A0B; \
+      (state)->B[0] = B0; \
+      (state)->B[1] = B1; \
+      (state)->B[2] = B2; \
+      (state)->B[3] = B3; \
+      (state)->B[4] = B4; \
+      (state)->B[5] = B5; \
+      (state)->B[6] = B6; \
+      (state)->B[7] = B7; \
+      (state)->B[8] = B8; \
+      (state)->B[9] = B9; \
+      (state)->B[10] = BA; \
+      (state)->B[11] = BB; \
+      (state)->B[12] = BC; \
+      (state)->B[13] = BD; \
+      (state)->B[14] = BE; \
+      (state)->B[15] = BF; \
+      (state)->C[0] = C0; \
+      (state)->C[1] = C1; \
+      (state)->C[2] = C2; \
+      (state)->C[3] = C3; \
+      (state)->C[4] = C4; \
+      (state)->C[5] = C5; \
+      (state)->C[6] = C6; \
+      (state)->C[7] = C7; \
+      (state)->C[8] = C8; \
+      (state)->C[9] = C9; \
+      (state)->C[10] = CA; \
+      (state)->C[11] = CB; \
+      (state)->C[12] = CC; \
+      (state)->C[13] = CD; \
+      (state)->C[14] = CE; \
+      (state)->C[15] = CF; \
+      (state)->Wlow = Wlow; \
+      (state)->Whigh = Whigh; \
+   } while (0)
+
+#define DECODE_BLOCK8 \
+do { \
+   M0 = buf[ 0]; \
+   M1 = buf[ 1]; \
+   M2 = buf[ 2]; \
+   M3 = buf[ 3]; \
+   M4 = buf[ 4]; \
+   M5 = buf[ 5]; \
+   M6 = buf[ 6]; \
+   M7 = buf[ 7]; \
+   M8 = buf[ 8]; \
+   M9 = buf[ 9]; \
+   MA = buf[10]; \
+   MB = buf[11]; \
+   MC = buf[12]; \
+   MD = buf[13]; \
+   ME = buf[14]; \
+   MF = buf[15]; \
+} while (0)
+
+#define INPUT_BLOCK_ADD8 \
+do { \
+    B0 = _mm256_add_epi32( B0, M0 );\
+    B1 = _mm256_add_epi32( B1, M1 );\
+    B2 = _mm256_add_epi32( B2, M2 );\
+    B3 = _mm256_add_epi32( B3, M3 );\
+    B4 = _mm256_add_epi32( B4, M4 );\
+    B5 = _mm256_add_epi32( B5, M5 );\
+    B6 = _mm256_add_epi32( B6, M6 );\
+    B7 = _mm256_add_epi32( B7, M7 );\
+    B8 = _mm256_add_epi32( B8, M8 );\
+    B9 = _mm256_add_epi32( B9, M9 );\
+    BA = _mm256_add_epi32( BA, MA );\
+    BB = _mm256_add_epi32( BB, MB );\
+    BC = _mm256_add_epi32( BC, MC );\
+    BD = _mm256_add_epi32( BD, MD );\
+    BE = _mm256_add_epi32( BE, ME );\
+    BF = _mm256_add_epi32( BF, MF );\
+} while (0)
+
+#define INPUT_BLOCK_SUB8 \
+do { \
+    C0 = _mm256_sub_epi32( C0, M0 ); \
+    C1 = _mm256_sub_epi32( C1, M1 ); \
+    C2 = _mm256_sub_epi32( C2, M2 ); \
+    C3 = _mm256_sub_epi32( C3, M3 ); \
+    C4 = _mm256_sub_epi32( C4, M4 ); \
+    C5 = _mm256_sub_epi32( C5, M5 ); \
+    C6 = _mm256_sub_epi32( C6, M6 ); \
+    C7 = _mm256_sub_epi32( C7, M7 ); \
+    C8 = _mm256_sub_epi32( C8, M8 ); \
+    C9 = _mm256_sub_epi32( C9, M9 ); \
+    CA = _mm256_sub_epi32( CA, MA ); \
+    CB = _mm256_sub_epi32( CB, MB ); \
+    CC = _mm256_sub_epi32( CC, MC ); \
+    CD = _mm256_sub_epi32( CD, MD ); \
+    CE = _mm256_sub_epi32( CE, ME ); \
+    CF = _mm256_sub_epi32( CF, MF ); \
+} while (0)
+
+#define XOR_W8 \
+do { \
+   A00 = _mm256_xor_si256( A00, _mm256_set1_epi32( Wlow ) ); \
+   A01 = _mm256_xor_si256( A01, _mm256_set1_epi32( Whigh ) ); \
+} while (0)
+
+#define SWAP_BC8 \
+do { \
+    mm256_swap512_256( B0, C0 ); \
+    mm256_swap512_256( B1, C1 ); \
+    mm256_swap512_256( B2, C2 ); \
+    mm256_swap512_256( B3, C3 ); \
+    mm256_swap512_256( B4, C4 ); \
+    mm256_swap512_256( B5, C5 ); \
+    mm256_swap512_256( B6, C6 ); \
+    mm256_swap512_256( B7, C7 ); \
+    mm256_swap512_256( B8, C8 ); \
+    mm256_swap512_256( B9, C9 ); \
+    mm256_swap512_256( BA, CA ); \
+    mm256_swap512_256( BB, CB ); \
+    mm256_swap512_256( BC, CC ); \
+    mm256_swap512_256( BD, CD ); \
+    mm256_swap512_256( BE, CE ); \
+    mm256_swap512_256( BF, CF ); \
+} while (0)
+
+#define PERM_ELT8(xa0, xa1, xb0, xb1, xb2, xb3, xc, xm) \
+do { \
+   xa0 = _mm256_xor_si256( xm, _mm256_xor_si256( xb1, _mm256_xor_si256(  \
+            _mm256_andnot_si256( xb3, xb2 ), \
+            _mm256_mullo_epi32( _mm256_xor_si256( xa0, _mm256_xor_si256( xc, \
+               _mm256_mullo_epi32(  mm256_rol_32( xa1, 15 ), _mm256_set1_epi32(5UL) ) \
+                   ) ), _mm256_set1_epi32(3UL) ) ) ) ); \
+   xb0 = mm256_not( _mm256_xor_si256( xa0, mm256_rol_32( xb0, 1 ) ) ); \
+} while (0)
+
+#define PERM_STEP_0_8   do { \
+      PERM_ELT8(A00, A0B, B0, BD, B9, B6, C8, M0); \
+      PERM_ELT8(A01, A00, B1, BE, BA, B7, C7, M1); \
+      PERM_ELT8(A02, A01, B2, BF, BB, B8, C6, M2); \
+      PERM_ELT8(A03, A02, B3, B0, BC, B9, C5, M3); \
+      PERM_ELT8(A04, A03, B4, B1, BD, BA, C4, M4); \
+      PERM_ELT8(A05, A04, B5, B2, BE, BB, C3, M5); \
+      PERM_ELT8(A06, A05, B6, B3, BF, BC, C2, M6); \
+      PERM_ELT8(A07, A06, B7, B4, B0, BD, C1, M7); \
+      PERM_ELT8(A08, A07, B8, B5, B1, BE, C0, M8); \
+      PERM_ELT8(A09, A08, B9, B6, B2, BF, CF, M9); \
+      PERM_ELT8(A0A, A09, BA, B7, B3, B0, CE, MA); \
+      PERM_ELT8(A0B, A0A, BB, B8, B4, B1, CD, MB); \
+      PERM_ELT8(A00, A0B, BC, B9, B5, B2, CC, MC); \
+      PERM_ELT8(A01, A00, BD, BA, B6, B3, CB, MD); \
+      PERM_ELT8(A02, A01, BE, BB, B7, B4, CA, ME); \
+      PERM_ELT8(A03, A02, BF, BC, B8, B5, C9, MF); \
+   } while (0)
+
+#define PERM_STEP_1_8   do { \
+      PERM_ELT8(A04, A03, B0, BD, B9, B6, C8, M0); \
+      PERM_ELT8(A05, A04, B1, BE, BA, B7, C7, M1); \
+      PERM_ELT8(A06, A05, B2, BF, BB, B8, C6, M2); \
+      PERM_ELT8(A07, A06, B3, B0, BC, B9, C5, M3); \
+      PERM_ELT8(A08, A07, B4, B1, BD, BA, C4, M4); \
+      PERM_ELT8(A09, A08, B5, B2, BE, BB, C3, M5); \
+      PERM_ELT8(A0A, A09, B6, B3, BF, BC, C2, M6); \
+      PERM_ELT8(A0B, A0A, B7, B4, B0, BD, C1, M7); \
+      PERM_ELT8(A00, A0B, B8, B5, B1, BE, C0, M8); \
+      PERM_ELT8(A01, A00, B9, B6, B2, BF, CF, M9); \
+      PERM_ELT8(A02, A01, BA, B7, B3, B0, CE, MA); \
+      PERM_ELT8(A03, A02, BB, B8, B4, B1, CD, MB); \
+      PERM_ELT8(A04, A03, BC, B9, B5, B2, CC, MC); \
+      PERM_ELT8(A05, A04, BD, BA, B6, B3, CB, MD); \
+      PERM_ELT8(A06, A05, BE, BB, B7, B4, CA, ME); \
+      PERM_ELT8(A07, A06, BF, BC, B8, B5, C9, MF); \
+   } while (0)
+
+#define PERM_STEP_2_8   do { \
+      PERM_ELT8(A08, A07, B0, BD, B9, B6, C8, M0); \
+      PERM_ELT8(A09, A08, B1, BE, BA, B7, C7, M1); \
+      PERM_ELT8(A0A, A09, B2, BF, BB, B8, C6, M2); \
+      PERM_ELT8(A0B, A0A, B3, B0, BC, B9, C5, M3); \
+      PERM_ELT8(A00, A0B, B4, B1, BD, BA, C4, M4); \
+      PERM_ELT8(A01, A00, B5, B2, BE, BB, C3, M5); \
+      PERM_ELT8(A02, A01, B6, B3, BF, BC, C2, M6); \
+      PERM_ELT8(A03, A02, B7, B4, B0, BD, C1, M7); \
+      PERM_ELT8(A04, A03, B8, B5, B1, BE, C0, M8); \
+      PERM_ELT8(A05, A04, B9, B6, B2, BF, CF, M9); \
+      PERM_ELT8(A06, A05, BA, B7, B3, B0, CE, MA); \
+      PERM_ELT8(A07, A06, BB, B8, B4, B1, CD, MB); \
+      PERM_ELT8(A08, A07, BC, B9, B5, B2, CC, MC); \
+      PERM_ELT8(A09, A08, BD, BA, B6, B3, CB, MD); \
+      PERM_ELT8(A0A, A09, BE, BB, B7, B4, CA, ME); \
+      PERM_ELT8(A0B, A0A, BF, BC, B8, B5, C9, MF); \
+   } while (0)
+
+#define APPLY_P8 \
+do { \
+    B0 = mm256_ror_32( B0, 15 ); \
+    B1 = mm256_ror_32( B1, 15 ); \
+    B2 = mm256_ror_32( B2, 15 ); \
+    B3 = mm256_ror_32( B3, 15 ); \
+    B4 = mm256_ror_32( B4, 15 ); \
+    B5 = mm256_ror_32( B5, 15 ); \
+    B6 = mm256_ror_32( B6, 15 ); \
+    B7 = mm256_ror_32( B7, 15 ); \
+    B8 = mm256_ror_32( B8, 15 ); \
+    B9 = mm256_ror_32( B9, 15 ); \
+    BA = mm256_ror_32( BA, 15 ); \
+    BB = mm256_ror_32( BB, 15 ); \
+    BC = mm256_ror_32( BC, 15 ); \
+    BD = mm256_ror_32( BD, 15 ); \
+    BE = mm256_ror_32( BE, 15 ); \
+    BF = mm256_ror_32( BF, 15 ); \
+    PERM_STEP_0_8; \
+    PERM_STEP_1_8; \
+    PERM_STEP_2_8; \
+    A0B = _mm256_add_epi32( A0B, C6 ); \
+    A0A = _mm256_add_epi32( A0A, C5 ); \
+    A09 = _mm256_add_epi32( A09, C4 ); \
+    A08 = _mm256_add_epi32( A08, C3 ); \
+    A07 = _mm256_add_epi32( A07, C2 ); \
+    A06 = _mm256_add_epi32( A06, C1 ); \
+    A05 = _mm256_add_epi32( A05, C0 ); \
+    A04 = _mm256_add_epi32( A04, CF ); \
+    A03 = _mm256_add_epi32( A03, CE ); \
+    A02 = _mm256_add_epi32( A02, CD ); \
+    A01 = _mm256_add_epi32( A01, CC ); \
+    A00 = _mm256_add_epi32( A00, CB ); \
+    A0B = _mm256_add_epi32( A0B, CA ); \
+    A0A = _mm256_add_epi32( A0A, C9 ); \
+    A09 = _mm256_add_epi32( A09, C8 ); \
+    A08 = _mm256_add_epi32( A08, C7 ); \
+    A07 = _mm256_add_epi32( A07, C6 ); \
+    A06 = _mm256_add_epi32( A06, C5 ); \
+    A05 = _mm256_add_epi32( A05, C4 ); \
+    A04 = _mm256_add_epi32( A04, C3 ); \
+    A03 = _mm256_add_epi32( A03, C2 ); \
+    A02 = _mm256_add_epi32( A02, C1 ); \
+    A01 = _mm256_add_epi32( A01, C0 ); \
+    A00 = _mm256_add_epi32( A00, CF ); \
+    A0B = _mm256_add_epi32( A0B, CE ); \
+    A0A = _mm256_add_epi32( A0A, CD ); \
+    A09 = _mm256_add_epi32( A09, CC ); \
+    A08 = _mm256_add_epi32( A08, CB ); \
+    A07 = _mm256_add_epi32( A07, CA ); \
+    A06 = _mm256_add_epi32( A06, C9 ); \
+    A05 = _mm256_add_epi32( A05, C8 ); \
+    A04 = _mm256_add_epi32( A04, C7 ); \
+    A03 = _mm256_add_epi32( A03, C6 ); \
+    A02 = _mm256_add_epi32( A02, C5 ); \
+    A01 = _mm256_add_epi32( A01, C4 ); \
+    A00 = _mm256_add_epi32( A00, C3 ); \
+} while (0)
+
+#define INCR_W8   do { \
+      if ((Wlow = T32(Wlow + 1)) == 0) \
+         Whigh = T32(Whigh + 1); \
+   } while (0)
+
+static void
+shabal_8way_init( void *cc, unsigned size )
+{
+   shabal_8way_context *sc = (shabal_8way_context*)cc;
+
+   if ( size == 512 )
+   { // copy immediate constants directly to working registers later.
+       sc->state_loaded = false;
+   }
+   else
+   {  // No users
+       sc->state_loaded = true;
+       sc->A[ 0] = m256_const1_64( 0x52F8455252F84552 );
+       sc->A[ 1] = m256_const1_64( 0xE54B7999E54B7999 );
+       sc->A[ 2] = m256_const1_64( 0x2D8EE3EC2D8EE3EC );
+       sc->A[ 3] = m256_const1_64( 0xB9645191B9645191 );
+       sc->A[ 4] = m256_const1_64( 0xE0078B86E0078B86 );
+       sc->A[ 5] = m256_const1_64( 0xBB7C44C9BB7C44C9 );
+       sc->A[ 6] = m256_const1_64( 0xD2B5C1CAD2B5C1CA );
+       sc->A[ 7] = m256_const1_64( 0xB0D2EB8CB0D2EB8C );
+       sc->A[ 8] = m256_const1_64( 0x14CE5A4514CE5A45 );
+       sc->A[ 9] = m256_const1_64( 0x22AF50DC22AF50DC );
+       sc->A[10] = m256_const1_64( 0xEFFDBC6BEFFDBC6B );
+       sc->A[11] = m256_const1_64( 0xEB21B74AEB21B74A );
+
+       sc->B[ 0] = m256_const1_64( 0xB555C6EEB555C6EE );
+       sc->B[ 1] = m256_const1_64( 0x3E7105963E710596 );
+       sc->B[ 2] = m256_const1_64( 0xA72A652FA72A652F );
+       sc->B[ 3] = m256_const1_64( 0x9301515F9301515F );
+       sc->B[ 4] = m256_const1_64( 0xDA28C1FADA28C1FA );
+       sc->B[ 5] = m256_const1_64( 0x696FD868696FD868 );
+       sc->B[ 6] = m256_const1_64( 0x9CB6BF729CB6BF72 );
+       sc->B[ 7] = m256_const1_64( 0x0AFE40020AFE4002 );
+       sc->B[ 8] = m256_const1_64( 0xA6E03615A6E03615 );
+       sc->B[ 9] = m256_const1_64( 0x5138C1D45138C1D4 );
+       sc->B[10] = m256_const1_64( 0xBE216306BE216306 );
+       sc->B[11] = m256_const1_64( 0xB38B8890B38B8890 );
+       sc->B[12] = m256_const1_64( 0x3EA8B96B3EA8B96B );
+       sc->B[13] = m256_const1_64( 0x3299ACE43299ACE4 );
+       sc->B[14] = m256_const1_64( 0x30924DD430924DD4 );
+       sc->B[15] = m256_const1_64( 0x55CB34A555CB34A5 );
+
+       sc->C[ 0] = m256_const1_64( 0xB405F031B405F031 );
+       sc->C[ 1] = m256_const1_64( 0xC4233EBAC4233EBA );
+       sc->C[ 2] = m256_const1_64( 0xB3733979B3733979 );
+       sc->C[ 3] = m256_const1_64( 0xC0DD9D55C0DD9D55 );
+       sc->C[ 4] = m256_const1_64( 0xC51C28AEC51C28AE );
+       sc->C[ 5] = m256_const1_64( 0xA327B8E1A327B8E1 );
+       sc->C[ 6] = m256_const1_64( 0x56C5616756C56167 );
+       sc->C[ 7] = m256_const1_64( 0xED614433ED614433 );
+       sc->C[ 8] = m256_const1_64( 0x88B59D6088B59D60 );
+       sc->C[ 9] = m256_const1_64( 0x60E2CEBA60E2CEBA );
+       sc->C[10] = m256_const1_64( 0x758B4B8B758B4B8B );
+       sc->C[11] = m256_const1_64( 0x83E82A7F83E82A7F );
+       sc->C[12] = m256_const1_64( 0xBC968828BC968828 );
+       sc->C[13] = m256_const1_64( 0xE6E00BF7E6E00BF7 );
+       sc->C[14] = m256_const1_64( 0xBA839E55BA839E55 );
+       sc->C[15] = m256_const1_64( 0x9B491C609B491C60 );
+   }
+    sc->Wlow = 1;
+    sc->Whigh = 0;
+    sc->ptr = 0;
+}
+
+static void
+shabal_8way_core( void *cc, const unsigned char *data, size_t len )
+{
+   shabal_8way_context *sc = (shabal_8way_context*)cc;
+    __m256i *buf;
+    __m256i *vdata = (__m256i*)data;
+   const int buf_size = 64;
+   size_t ptr;
+   DECL_STATE8
+
+   buf = sc->buf;
+   ptr = sc->ptr;
+
+   if ( len < (buf_size - ptr ) )
+   {
+      memcpy_256( buf + (ptr>>2), vdata, len>>2 );
+      ptr += len;
+      sc->ptr = ptr;
+      return;
+   }
+
+   READ_STATE8( sc );
+
+   while ( len > 0 )
+   {
+      size_t clen;
+      clen = buf_size - ptr;
+      if ( clen > len )
+         clen = len;
+      memcpy_256( buf + (ptr>>2), vdata, clen>>2 );
+
+      ptr += clen;
+      vdata += clen>>2;
+      len -= clen;
+      if ( ptr == buf_size )
+      {
+         DECODE_BLOCK8;
+         INPUT_BLOCK_ADD8;
+         XOR_W8;
+         APPLY_P8;
+         INPUT_BLOCK_SUB8;
+         SWAP_BC8;
+         INCR_W8;
+         ptr = 0;
+      }
+   }
+   WRITE_STATE8(sc);
+   sc->ptr = ptr;
+}
+
+static void
+shabal_8way_close( void *cc, unsigned ub, unsigned n, void *dst,
+                   unsigned size_words )
+{
+   shabal_8way_context *sc = (shabal_8way_context*)cc;
+    __m256i *buf;
+   const int buf_size = 64;
+   size_t ptr;
+   int i;
+   unsigned z, zz;
+   DECL_STATE8
+
+   buf = sc->buf;
+   ptr = sc->ptr;
+   z = 0x80 >> n;
+   zz = ((ub & -z) | z) & 0xFF;
+   buf[ptr>>2] = _mm256_set1_epi32( zz );
+   memset_zero_256( buf + (ptr>>2) + 1, ( (buf_size - ptr) >> 2 ) - 1 );
+   READ_STATE8(sc);
+   DECODE_BLOCK8;
+   INPUT_BLOCK_ADD8;
+   XOR_W8;
+   APPLY_P8;
+
+   for ( i = 0; i < 3; i ++ )
+   {
+      SWAP_BC8;
+      XOR_W8;
+      APPLY_P8;
+   }
+
+   __m256i *d = (__m256i*)dst;
+   if ( size_words == 16 )   // 512
+   {
+      d[ 0] = B0; d[ 1] = B1; d[ 2] = B2; d[ 3] = B3;
+      d[ 4] = B4; d[ 5] = B5; d[ 6] = B6; d[ 7] = B7;
+      d[ 8] = B8; d[ 9] = B9; d[10] = BA; d[11] = BB;
+      d[12] = BC; d[13] = BD; d[14] = BE; d[15] = BF;
+   }
+   else    // 256
+   {
+      d[ 0] = B8; d[ 1] = B9; d[ 2] = BA; d[ 3] = BB;
+      d[ 4] = BC; d[ 5] = BD; d[ 6] = BE; d[ 7] = BF;
+   }
+}
+
+void
+shabal256_8way_init( void *cc )
+{
+   shabal_8way_init(cc, 256);
+}
+
+void
+shabal256_8way_update( void *cc, const void *data, size_t len )
+{
+   shabal_8way_core( cc, data, len );
+}
+
+void
+shabal256_8way_close( void *cc, void *dst )
+{
+   shabal_8way_close(cc, 0, 0, dst, 8);
+}
+
+void
+shabal256_8way_addbits_and_close( void *cc, unsigned ub, unsigned n,
+                                  void *dst )
+{
+   shabal_8way_close(cc, ub, n, dst, 8);
+}
+
+void
+shabal512_8way_init(void *cc)
+{
+   shabal_8way_init(cc, 512);
+}
+
+void
+shabal512_8way_update(void *cc, const void *data, size_t len)
+{
+   shabal_8way_core(cc, data, len);
+}
+
+void
+shabal512_8way_close(void *cc, void *dst)
+{
+   shabal_8way_close(cc, 0, 0, dst, 16);
+}
+
+void
+shabal512_8way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+{
+   shabal_8way_close(cc, ub, n, dst, 16);
+}
+
+
+#endif  // AVX2
+
 /*
  * We copy the state into local variables, so that the compiler knows
  * that it can optimize them at will.
@@ -290,6 +883,8 @@ do { \
    A00 = _mm_xor_si128( A00, _mm_set1_epi32( Wlow ) ); \
    A01 = _mm_xor_si128( A01, _mm_set1_epi32( Whigh ) ); \
 } while (0)
+
+
 /*
 #define SWAP(v1, v2)   do { \
 		sph_u32 tmp = (v1); \
@@ -297,26 +892,39 @@ do { \
 		(v2) = tmp; \
 	} while (0)
 */
+
 #define SWAP_BC \
 do { \
-    mm128_swap128_256( B0, C0 ); \
-    mm128_swap128_256( B1, C1 ); \
-    mm128_swap128_256( B2, C2 ); \
-    mm128_swap128_256( B3, C3 ); \
-    mm128_swap128_256( B4, C4 ); \
-    mm128_swap128_256( B5, C5 ); \
-    mm128_swap128_256( B6, C6 ); \
-    mm128_swap128_256( B7, C7 ); \
-    mm128_swap128_256( B8, C8 ); \
-    mm128_swap128_256( B9, C9 ); \
-    mm128_swap128_256( BA, CA ); \
-    mm128_swap128_256( BB, CB ); \
-    mm128_swap128_256( BC, CC ); \
-    mm128_swap128_256( BD, CD ); \
-    mm128_swap128_256( BE, CE ); \
-    mm128_swap128_256( BF, CF ); \
+    mm128_swap256_128( B0, C0 ); \
+    mm128_swap256_128( B1, C1 ); \
+    mm128_swap256_128( B2, C2 ); \
+    mm128_swap256_128( B3, C3 ); \
+    mm128_swap256_128( B4, C4 ); \
+    mm128_swap256_128( B5, C5 ); \
+    mm128_swap256_128( B6, C6 ); \
+    mm128_swap256_128( B7, C7 ); \
+    mm128_swap256_128( B8, C8 ); \
+    mm128_swap256_128( B9, C9 ); \
+    mm128_swap256_128( BA, CA ); \
+    mm128_swap256_128( BB, CB ); \
+    mm128_swap256_128( BC, CC ); \
+    mm128_swap256_128( BD, CD ); \
+    mm128_swap256_128( BE, CE ); \
+    mm128_swap256_128( BF, CF ); \
 } while (0)
 
+/*
+#define PERM_ELT(xa0, xa1, xb0, xb1, xb2, xb3, xc, xm) \
+do { \
+  __m128i t1 = _mm_mullo_epi32(  mm_rol_32( xa1, 15 ),\
+                                   _mm_set1_epi32(5UL) ) \
+  __m128i t2 = _mm_xor_si128( xa0, xc ); \
+  xb0 = mm_not( _mm_xor_si256( xa0, mm_rol_32( xb0, 1 ) ) ); \
+  xa0 = mm_xor4( xm, xb1, _mm_andnot_si128( xb3, xb2 ), \
+              _mm_xor_si128( t2, \
+                      _mm_mullo_epi32( t1, _mm_set1_epi32(5UL) ) ) ) \
+*/
+
 #define PERM_ELT(xa0, xa1, xb0, xb1, xb2, xb3, xc, xm) \
 do { \
    xa0 = _mm_xor_si128( xm, _mm_xor_si128( xb1, _mm_xor_si128(  \
@@ -706,7 +1314,7 @@ shabal256_4way_init( void *cc )
 }
 
 void
-shabal256_4way( void *cc, const void *data, size_t len )
+shabal256_4way_update( void *cc, const void *data, size_t len )
 {
 	shabal_4way_core( cc, data, len );
 }
@@ -731,7 +1339,7 @@ shabal512_4way_init(void *cc)
 }
 
 void
-shabal512_4way(void *cc, const void *data, size_t len)
+shabal512_4way_update(void *cc, const void *data, size_t len)
 {
 	shabal_4way_core(cc, data, len);
 }

algo/shabal/shabal-hash-4way.h

@@ -36,7 +36,7 @@
 #ifndef SHABAL_HASH_4WAY_H__
 #define SHABAL_HASH_4WAY_H__ 1
 
-#ifdef __AVX2__
+#ifdef __SSE4_1__
 
 #include <stddef.h>
 #include "algo/sha/sph_types.h"
@@ -50,6 +50,34 @@ extern "C"{
 
 #define SPH_SIZE_shabal512   512
 
+#if defined(__AVX2__)
+
+typedef struct {
+   __m256i buf[16];
+   __m256i A[12], B[16], C[16];
+   sph_u32 Whigh, Wlow;
+   size_t ptr;
+   bool state_loaded;
+} shabal_8way_context __attribute__ ((aligned (64)));
+
+typedef shabal_8way_context shabal256_8way_context;
+typedef shabal_8way_context shabal512_8way_context;
+
+void shabal256_8way_init( void *cc );
+void shabal256_8way_update( void *cc, const void *data, size_t len );
+void shabal256_8way_close( void *cc, void *dst );
+void shabal256_8way_addbits_and_close( void *cc, unsigned ub, unsigned n,
+                                       void *dst );
+
+void shabal512_8way_init( void *cc );
+void shabal512_8way_update( void *cc, const void *data, size_t len );
+void shabal512_8way_close( void *cc, void *dst );
+void shabal512_8way_addbits_and_close( void *cc, unsigned ub, unsigned n,
+                                       void *dst );
+
+
+#endif
+
 typedef struct {
 	__m128i buf[16] __attribute__ ((aligned (64)));
 	__m128i A[12], B[16], C[16];
@@ -62,13 +90,14 @@ typedef shabal_4way_context shabal256_4way_context;
 typedef shabal_4way_context shabal512_4way_context;
 
 void shabal256_4way_init( void *cc );
-void shabal256_4way( void *cc, const void *data, size_t len );
+void shabal256_4way_update( void *cc, const void *data, size_t len );
 void shabal256_4way_close( void *cc, void *dst );
 void shabal256_4way_addbits_and_close(	void *cc, unsigned ub, unsigned n,
                                        void *dst );
 
 void shabal512_4way_init( void *cc );
-void shabal512_4way( void *cc, const void *data, size_t len );
+void shabal512_4way_update( void *cc, const void *data, size_t len );
+#define shabal512_4way shabal512_4way_update
 void shabal512_4way_close( void *cc, void *dst );
 void shabal512_4way_addbits_and_close( void *cc, unsigned ub, unsigned n,
                                        void *dst );

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

@@ -3,8 +3,15 @@
 
 #include <stdio.h>
 
+// This implementation is deprecated, superseded by VAES in Icelake
+// which provides HW based 4 way aes.
+// It was created for AVX2 to eliminate interleaving between the 
+// preceding and following function.
+// This code can be removed when current users have reverted to one way.
+
 #if defined(__AVX2__)
 
+
 static const uint32_t IV512[] =
 {
         0x72FCCDD8, 0x79CA4727, 0x128A077B, 0x40D55AEC,
@@ -13,9 +20,10 @@ 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 )
+   _mm256_blend_epi32( mm256_ror128_32( a ), \
+                       mm256_ror128_32( b ), 0x88 )
 
 static void
 c512_2way( shavite512_2way_context *ctx, const void *msg )
@@ -59,7 +67,7 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
    {
       // round 1, 5, 9
 
-     k00 = _mm256_xor_si256( k13, mm256_ror1x32_128(
+     k00 = _mm256_xor_si256( k13, mm256_ror128_32(
                                   mm256_aesenc_2x128( k00, zero ) ) );
 
      if ( r == 0 )
@@ -69,7 +77,7 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
 
      x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k00 ), zero );
      k01 = _mm256_xor_si256( k00,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k01, zero ) ) );
+		     mm256_ror128_32( mm256_aesenc_2x128( k01, zero ) ) );
 
      if ( r == 1 )
         k01 = _mm256_xor_si256( k01, _mm256_set_epi32(
@@ -78,25 +86,25 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
 
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
      k02 = _mm256_xor_si256( k01,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k02, zero ) ) );
+		     mm256_ror128_32( mm256_aesenc_2x128( k02, zero ) ) );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
      k03 = _mm256_xor_si256( k02,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k03, zero ) ) );
+		     mm256_ror128_32( mm256_aesenc_2x128( k03, zero ) ) );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
      p3 = _mm256_xor_si256( p3, x );
 
      k10 = _mm256_xor_si256( k03,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k10, zero ) ) );
+		     mm256_ror128_32( mm256_aesenc_2x128( k10, zero ) ) );
      x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k10 ), zero );
      k11 = _mm256_xor_si256( k10,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k11, zero ) ) );
+		     mm256_ror128_32( mm256_aesenc_2x128( k11, zero ) ) );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
      k12 = _mm256_xor_si256( k11,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k12, zero ) ) );
+		     mm256_ror128_32( mm256_aesenc_2x128( k12, zero ) ) );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
      k13 = _mm256_xor_si256( k12,
-		     mm256_ror1x32_128( mm256_aesenc_2x128( k13, zero ) ) );
+		     mm256_ror128_32( mm256_aesenc_2x128( k13, zero ) ) );
 
      if ( r == 2 )
         k13 = _mm256_xor_si256( k13, _mm256_set_epi32(
@@ -132,31 +140,31 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
 
      // round 3, 7, 11
 
-     k00 = _mm256_xor_si256( mm256_ror1x32_128(
+     k00 = _mm256_xor_si256( mm256_ror128_32(
                                      mm256_aesenc_2x128( k00, zero ) ), k13 );
      x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k00 ), zero );
-     k01 = _mm256_xor_si256( mm256_ror1x32_128(
+     k01 = _mm256_xor_si256( mm256_ror128_32(
                                      mm256_aesenc_2x128( k01, zero ) ), k00 );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
-     k02 = _mm256_xor_si256( mm256_ror1x32_128(
+     k02 = _mm256_xor_si256( mm256_ror128_32(
                                      mm256_aesenc_2x128( k02, zero ) ), k01 );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
-     k03 = _mm256_xor_si256( mm256_ror1x32_128(
+     k03 = _mm256_xor_si256( mm256_ror128_32(
                                      mm256_aesenc_2x128( k03, zero ) ), k02 );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
      p1 = _mm256_xor_si256( p1, x );
 
-     k10 = _mm256_xor_si256( mm256_ror1x32_128(
+     k10 = _mm256_xor_si256( mm256_ror128_32(
                                      mm256_aesenc_2x128( k10, zero ) ), k03 );
      x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k10 ), zero );
-     k11 = _mm256_xor_si256( mm256_ror1x32_128(
+     k11 = _mm256_xor_si256( mm256_ror128_32(
                                      mm256_aesenc_2x128( k11, zero ) ), k10 );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
-     k12 = _mm256_xor_si256( mm256_ror1x32_128(
+     k12 = _mm256_xor_si256( mm256_ror128_32(
                                      mm256_aesenc_2x128( k12, zero ) ), k11 );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
-     k13 = _mm256_xor_si256( mm256_ror1x32_128(
+     k13 = _mm256_xor_si256( mm256_ror128_32(
                                      mm256_aesenc_2x128( k13, zero ) ), k12 );
      x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );
 
@@ -190,35 +198,35 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
 
    // round 13
 
-   k00 = _mm256_xor_si256( mm256_ror1x32_128(
+   k00 = _mm256_xor_si256( mm256_ror128_32(
 			             mm256_aesenc_2x128( k00, zero ) ), k13  );
    x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k00 ), zero );
-   k01 = _mm256_xor_si256( mm256_ror1x32_128(
+   k01 = _mm256_xor_si256( mm256_ror128_32(
 			             mm256_aesenc_2x128( k01, zero ) ), k00 );
    x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
-   k02 = _mm256_xor_si256( mm256_ror1x32_128(
+   k02 = _mm256_xor_si256( mm256_ror128_32(
 			             mm256_aesenc_2x128( k02, zero ) ), k01 );
    x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
-   k03 = _mm256_xor_si256( mm256_ror1x32_128(
+   k03 = _mm256_xor_si256( mm256_ror128_32(
 			             mm256_aesenc_2x128( k03, zero ) ), k02 );
    x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );
 
    p3 = _mm256_xor_si256( p3, x );
 
-   k10 = _mm256_xor_si256( mm256_ror1x32_128(
+   k10 = _mm256_xor_si256( mm256_ror128_32(
 			             mm256_aesenc_2x128( k10, zero ) ), k03 );
    x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k10 ), zero );
-   k11 = _mm256_xor_si256( mm256_ror1x32_128(
+   k11 = _mm256_xor_si256( mm256_ror128_32(
 			             mm256_aesenc_2x128( k11, zero ) ), k10 );
    x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
 
-   k12 = mm256_ror1x32_128( mm256_aesenc_2x128( k12, zero ) );
+   k12 = mm256_ror128_32( mm256_aesenc_2x128( k12, zero ) );
    k12 = _mm256_xor_si256( k12, _mm256_xor_si256( k11, _mm256_set_epi32(
 	       ~ctx->count2, ctx->count3, ctx->count0, ctx->count1,
 	       ~ctx->count2, ctx->count3, ctx->count0, ctx->count1 ) ) );
 
    x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
-   k13 = _mm256_xor_si256( mm256_ror1x32_128(
+   k13 = _mm256_xor_si256( mm256_ror128_32(
 			             mm256_aesenc_2x128( k13, zero ) ), k12 );
    x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );
 
@@ -232,18 +240,14 @@ 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 +255,7 @@ 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 +292,7 @@ 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 +306,7 @@ 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 +320,12 @@ 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 +386,21 @@ 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);
 

algo/shavite/sph-shavite-aesni.c

@@ -100,9 +100,20 @@ c512( sph_shavite_big_context *sc, const void *msg )
    p3 = h[3];   
 
    // round
+
+//  working proof of concept   
+/*
+   __m512i K = m512_const1_128( m[0] );
+   __m512i X = _mm512_xor_si512( m512_const1_128( p1 ), K );
+   X = _mm512_aesenc_epi128( X, m512_zero );
+   k00 = _mm512_castsi512_si128( K );
+   x = _mm512_castsi512_si128( X );
+*/
+
    k00 = m[0];
    x = _mm_xor_si128( p1, k00 );
    x = _mm_aesenc_si128( x, zero );
+
    k01 = m[1];
    x = _mm_xor_si128( x, k01 );
    x = _mm_aesenc_si128( x, zero );

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

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

@@ -7,15 +7,37 @@
 
 #include "simd-utils.h"
 
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 typedef struct {
-  uint32_t A[ 32*2 ] __attribute__((aligned(64)));
-  uint8_t buffer[ 128*2 ] __attribute__((aligned(64)));
+  uint32_t A[ 32*4 ];
+  uint8_t buffer[ 128*4 ];
+  uint64_t count;
+  unsigned int hashbitlen;
+  unsigned int blocksize;
+  unsigned int n_feistels;
+
+} simd_4way_context __attribute__((aligned(128)));
+
+int simd_4way_init( simd_4way_context *state, int hashbitlen );
+int simd_4way_update( simd_4way_context *state, const void *data,
+                      int databitlen );
+int simd_4way_close( simd_4way_context *state, void *hashval );
+int simd_4way_update_close( simd_4way_context *state, void *hashval,
+                            const void *data, int databitlen );
+
+#endif
+
+typedef struct {
+  uint32_t A[ 32*2 ];
+  uint8_t buffer[ 128*2 ];
   uint64_t count;
   unsigned int hashbitlen;
   unsigned int blocksize;
   unsigned int n_feistels;
   
-} simd_2way_context;
+} simd_2way_context __attribute__((aligned(128)));
 
 int simd_2way_init( simd_2way_context *state, int hashbitlen );
 int simd_2way_update( simd_2way_context *state, const void *data,

algo/skein/skein-4way.c

@@ -2,17 +2,140 @@
 #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[8*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[16*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 )
 {
-     uint64_t vhash64[16*4] __attribute__ ((aligned (64)));
+     uint64_t vhash64[8*4] __attribute__ ((aligned (128)));
      skein512_4way_context ctx_skein;
 #if defined(__SHA__)
      uint32_t hash0[16] __attribute__ ((aligned (64)));
@@ -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;
 }
 

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;
+};
+
+

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

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);
 }