v3.8.3.3

v3.8.3.2
v3.8.3.1
2025-09-17 23:44:27 +00:00 · 2018-02-25 14:15:07 -05:00 · 2018-02-24 14:36:19 -05:00 · 2018-02-23 15:45:32 -05:00 · 2018-02-23 12:39:15 -05:00 · 2018-02-17 13:52:24 -05:00
238 changed files with 16860 additions and 6739 deletions
--- a/Makefile.am
+++ b/Makefile.am
@@ -22,7 +22,6 @@ cpuminer_SOURCES = \
  api.c \
  sysinfos.c \
  algo-gate-api.c\
-  crypto/blake2s.c \
  crypto/oaes_lib.c \
  crypto/c_keccak.c \
  crypto/c_groestl.c \
@@ -45,9 +44,15 @@ cpuminer_SOURCES = \
  algo/blake/blake-4way.c \
  algo/blake/sph_blake2b.c \
  algo/blake/blake2b.c \
+  algo/blake/sph-blake2s.c \
+  algo/blake/blake2s-hash-4way.c \
  algo/blake/blake2s.c \
+  algo/blake/blake2s-gate.c \
+  algo/blake/blake2s-4way.c \
+  algo/blake/blakecoin-gate.c \
  algo/blake/mod_blakecoin.c \
  algo/blake/blakecoin.c \
+  algo/blake/blakecoin-4way.c \
  algo/blake/decred-gate.c \
  algo/blake/decred.c \
  algo/blake/decred-4way.c \
@@ -68,12 +73,16 @@ cpuminer_SOURCES = \
  algo/gost/sph_gost.c \
  algo/groestl/sph_groestl.c \
  algo/groestl/groestl.c \
+  algo/groestl/myrgr-gate.c \
+  algo/groestl/myrgr-4way.c \
  algo/groestl/myr-groestl.c \
  algo/groestl/aes_ni/hash-groestl.c \
  algo/groestl/aes_ni/hash-groestl256.c \
  algo/fugue/sph_fugue.c \
  algo/hamsi/sph_hamsi.c \
-  algo/haval/haval.c\
+  algo/hamsi/hamsi-hash-4way.c \
+  algo/haval/haval.c \
+  algo/haval/haval-hash-4way.c \
  algo/heavy/sph_hefty1.c \
  algo/heavy/heavy.c \
  algo/heavy/bastion.c \
@@ -93,19 +102,26 @@ cpuminer_SOURCES = \
  algo/keccak/keccak-4way.c\
  algo/keccak/keccak-gate.c \
  algo/keccak/sse2/keccak.c \
-  algo/lbry.c \
  algo/luffa/sph_luffa.c \
  algo/luffa/luffa.c \
-  algo/luffa/sse2/luffa_for_sse2.c \
+  algo/luffa/luffa_for_sse2.c \
+  algo/luffa/luffa-hash-2way.c \
  algo/lyra2/lyra2.c \
  algo/lyra2/sponge.c \
+  algo/lyra2/lyra2rev2-gate.c \
  algo/lyra2/lyra2rev2.c \
+  algo/lyra2/lyra2rev2-4way.c \
  algo/lyra2/lyra2re.c \
  algo/lyra2/lyra2z-gate.c \
  algo/lyra2/lyra2z.c \
  algo/lyra2/lyra2z-4way.c \
  algo/lyra2/lyra2z330.c \
+  algo/lyra2/lyra2h-gate.c \
  algo/lyra2/lyra2h.c \
+  algo/lyra2/lyra2h-4way.c \
+  algo/lyra2/allium-gate.c \
+  algo/lyra2/allium-4way.c \
+  algo/lyra2/allium.c \
  algo/m7m.c \
  algo/neoscrypt/neoscrypt.c \
  algo/nist5/nist5-gate.c \
@@ -113,14 +129,28 @@ cpuminer_SOURCES = \
  algo/nist5/nist5.c \
  algo/nist5/zr5.c \
  algo/pluck.c \
+  algo/quark/quark-gate.c \
  algo/quark/quark.c \
+  algo/quark/quark-4way.c \
+  algo/quark/anime-gate.c \
+  algo/quark/anime.c \
+  algo/quark/anime-4way.c \
+  algo/qubit/qubit-gate.c \
  algo/qubit/qubit.c \
+  algo/qubit/qubit-2way.c \
+  algo/qubit/deep-gate.c \
+  algo/qubit/deep-2way.c \
  algo/qubit/deep.c \
  algo/ripemd/sph_ripemd.c \
+  algo/ripemd/ripemd-hash-4way.c \
+  algo/ripemd/lbry-gate.c \
+  algo/ripemd/lbry.c \
+  algo/ripemd/lbry-4way.c \
  algo/scrypt.c \
  algo/scryptjane/scrypt-jane.c \
  algo/sha/sph_sha2.c \
  algo/sha/sph_sha2big.c \
+  algo/sha/sha2-hash-4way.c \
  algo/sha/sha2.c \
  algo/sha/sha256t.c \
  algo/shabal/sph_shabal.c \
@@ -129,8 +159,9 @@ cpuminer_SOURCES = \
  algo/shavite/sph-shavite-aesni.c \
  algo/shavite/shavite.c \
  algo/simd/sph_simd.c \
-  algo/simd/sse2/nist.c \
-  algo/simd/sse2/vector.c \
+  algo/simd/nist.c \
+  algo/simd/vector.c \
+  algo/simd/simd-hash-2way.c \
  algo/skein/sph_skein.c \
  algo/skein/skein-hash-4way.c \
  algo/skein/skein.c \
@@ -140,9 +171,8 @@ cpuminer_SOURCES = \
  algo/skein/skein2-4way.c \
  algo/skein/skein2-gate.c \
  algo/sm3/sm3.c \
+  algo/sm3/sm3-hash-4way.c \
  algo/tiger/sph_tiger.c \
-  algo/timetravel.c \
-  algo/timetravel10.c \
  algo/whirlpool/sph_whirlpool.c \
  algo/whirlpool/whirlpool-hash-4way.c \
  algo/whirlpool/whirlpool-gate.c \
@@ -161,8 +191,19 @@ cpuminer_SOURCES = \
  algo/x11/tribus-gate.c \
  algo/x11/tribus.c \
  algo/x11/tribus-4way.c \
+  algo/x11/timetravel-gate.c \
+  algo/x11/timetravel.c \
+  algo/x11/timetravel-4way.c \
+  algo/x11/timetravel10-gate.c \
+  algo/x11/timetravel10.c \
+  algo/x11/timetravel10-4way.c \
  algo/x11/fresh.c \
  algo/x11/x11evo.c \
+  algo/x11/x11evo-4way.c \
+  algo/x11/x11evo-gate.c \
+  algo/x12/x12-gate.c \
+  algo/x12/x12.c \
+  algo/x12/x12-4way.c \
  algo/x13/x13-gate.c \
  algo/x13/x13.c \
  algo/x13/x13-4way.c \
@@ -195,6 +236,9 @@ cpuminer_SOURCES = \
  algo/x17/xevan-gate.c \
  algo/x17/xevan.c \
  algo/x17/xevan-4way.c \
+  algo/x17/x16r-gate.c \
+  algo/x17/x16r.c \
+  algo/x17/x16r-4way.c \
  algo/x17/hmq1725.c \
  algo/yescrypt/yescrypt.c \
  algo/yescrypt/sha256_Y.c \
--- a/README.md
+++ b/README.md
@@ -13,9 +13,34 @@ mailto://jayddee246@gmail.com

 See file RELEASE_NOTES for change log and compile instructions.

+Requirements
+------------
+
+1. A x86_64 architecture CPU with a minimum of SSE2 support. This includes
+Intel Core2 and newer and AMD equivalents. In order to take advantage of AES_NI
+optimizations a CPU with AES_NI is required. This includes Intel Westbridge
+and newer and AMD equivalents. Further optimizations are available on some
+algoritms for CPUs with AVX and AVX2, Sandybridge and Haswell respectively.
+
+Older CPUs are supported by cpuminer-multi by TPruvot but at reduced
+performance.
+
+ARM CPUs are not supported.
+
+2. 64 bit Linux OS. Ubuntu and Fedora based distributions, including Mint and
+Centos are known to work and have all dependencies in their repositories.
+Others may work but may require more effort.
+64 bit Windows OS is supported with mingw_w64 and msys or pre-built binaries.
+
+MacOS, OSx is not supported.
+
+3. Stratum pool. Some algos may work wallet mining using getwork or GBT. YMMV.
+
 Supported Algorithms
 --------------------

+                          allium       Garlicoin
+                          anime        Animecoin
                          argon2
                          axiom        Shabal-256 MemoHash
                          bastion
@@ -74,40 +99,19 @@ Supported Algorithms
                          x11          Dash
                          x11evo       Revolvercoin
                          x11gost      sib (SibCoin)
+                          x12          Galaxie Cash (GCH)
                          x13          X13
                          x13sm3       hsr (Hshare)
                          x14          X14
                          x15          X15
+                          x16r         Ravencoin
                          x17
                          xevan        Bitsend
                          yescrypt     Globalboost-Y (BSTY)
-                          yescryptr8   BitZeny (ZNY)\n\
+                          yescryptr8   BitZeny (ZNY)
                          yescryptr16  Yenten (YTN)
                          zr5          Ziftr

-Requirements
------------
-
-1. A x86_64 architecture CPU with a minimum of SSE2 support. This includes
-Intel Core2 and newer and AMD equivalents. In order to take advantage of AES_NI
-optimizations a CPU with AES_NI is required. This includes Intel Westbridge
-and newer and AMD equivalents. Further optimizations are available on some
-algoritms for CPUs with AVX and AVX2, Sandybridge and Haswell respectively.
-
-Older CPUs are supported by cpuminer-multi by TPruvot but at reduced
-performance.
-
-ARM CPUs are not supported.
-
-2. 64 bit Linux OS. Ubuntu and Fedora based distributions, including Mint and
-Centos are known to work and have all dependencies in their repositories.
-Others may work but may require more effort.
-64 bit Windows OS is supported with mingw_w64 and msys or pre-built binaries.
-
-MacOS, OSx is not supported.
-
-3. Stratum pool. Some algos may work wallet mining using getwork.
-
 Errata
 ------

@@ -136,10 +140,13 @@ output from the miner showing the startup and any errors.
 Donations
 ---------

-I do not do this for money but I have a donation address if users
-are so inclined.
+cpuminer-opt has no fees of any kind but donations are accepted.

-bitcoin:12tdvfF7KmAsihBXQXynT6E6th2c2pByTT?label=donations
+ BTC: 12tdvfF7KmAsihBXQXynT6E6th2c2pByTT
+ ETH: 0x72122edabcae9d3f57eab0729305a425f6fef6d0
+ LTC: LdUwoHJnux9r9EKqFWNvAi45kQompHk6e8
+ BCH: 1QKYkB6atn4P7RFozyziAXLEnurwnUM1cQ
+ BTG: GVUyECtRHeC5D58z9F3nGGfVQndwnsPnHQ

 Happy mining!

--- a/README.txt
+++ b/README.txt
@@ -19,19 +19,18 @@ Users are recommended to use an unoptimized miner such as cpuminer-multi.

 Exe name                Compile flags              Arch name

-cpuminer-sse2.exe      "-march=core2"              Core2   
-cpuminer-sse42.exe     "-march=corei7"             Nehalem
+cpuminer-sse2.exe      "-march=core2"              Core2, Nehalem   
 cpuminer-aes-sse42.exe "-maes -msse4.2"            Westmere
-cpuminer-avx.exe       "-march=corei7-avx"         Sandybridge, Ivybridge
+cpuminer-aes-avx.exe   "-march=corei7-avx"         Sandybridge, Ivybridge
 cpuminer-avx2.exe      "-march=core-avx2"          Haswell...
-cpuminer-avx-sha       "-march=corei7-avx -msha"   Ryzen...
-cpuminer-4way.exe      "-march=core-avx2 -DFOUR_WAY"       same as avx2
-cpuminer-4way-sha.exe  "-march=core-avx2 -msha -DFOUR_WAY" same as avx2-sha
+cpuminer-avx2-sha.exe  "-march=core-avx2 -msha"    Ryzen

-4way requires a CPU with AES and AVX2. It is still under development and
-only a few algos are supported. See change log in RELEASE_NOTES in source
-package for supported algos.
+If you like this software feel free to donate:
+
+BTC: 12tdvfF7KmAsihBXQXynT6E6th2c2pByTT
+ETH: 0x72122edabcae9d3f57eab0729305a425f6fef6d0
+LTC: LdUwoHJnux9r9EKqFWNvAi45kQompHk6e8
+BCH: 1QKYkB6atn4P7RFozyziAXLEnurwnUM1cQ
+BTG: GVUyECtRHeC5D58z9F3nGGfVQndwnsPnHQ

-Ryzen CPus perform better with AVX than AVX2 therefore an avx-sha build
-is provided. Four way still uses AVX2. 

--- a/90
+++ b/90
@@ -90,22 +90,17 @@ Additional optional compile flags, add the following to CFLAGS to activate:

 SPH may give slightly better performance on algos that use sha256 when using
 openssl 1.0.1 or older. Openssl 1.0.2 adds AVX2 and 1.1 adds SHA and perform
-better than SPH.
+better than SPH. This option is ignored when 4-way is used, even for CPUs
+with SHA.

-DFOUR_WAY
-
-4 way will give much better performance on supported algos with CPUs
-that have AVX2 and should only be used on CPUs with AVX2. 4 way algo
-support will be added incrementally, see change log below for supported algos.
- 
 Start mining.

 ./cpuminer -a algo -o url -u username -p password

 Windows

-The following in how the Windows binary releases are built. It's old and
-not very good but it works, for me anyway.
+Precompiled Windows binaries are built on a Linux host using Mingw
+with a more recent compiler than the following Windows hosted procedure.

 Building on Windows prerequisites:

@@ -137,10 +132,10 @@ or similar Windows program.
 In msys shell cd to miner directory.
 cd /c/path/to/cpuminer-opt

-Run winbuild.sh to build on Windows or execute the following commands.
+Run build.sh to build on Windows or execute the following commands.

 ./autogen.sh
-CFLAGS="-O3 -march=native -Wall" CXXFLAGS="$CFLAGS -std=gnu++11 -fpermissive" ./configure --with-curl
+CFLAGS="-O3 -march=native -Wall" ./configure --with-curl
 make

 Start mining
@@ -149,9 +144,9 @@ cpuminer.exe -a algo -o url -u user -p password

 The following tips may be useful for older AMD CPUs.

-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.
+AMD CPUs older than Steamroller, 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.

 Some users with AMD CPUs without AES_NI have reported problems compiling
@@ -165,6 +160,73 @@ Support for even older x86_64 without AES_NI or SSE2 is not availble.
 Change Log
 ----------

+v3.8.3.3
+
+Integrated getblocktemplate with algo_gate.
+Added support for hodl gbt (untested).
+Reworked some recent quick fixes.
+
+v3.8.3.2
+
+Reverted gbt changes from v3.8.0 that broke getwork.
+Reverted scaled hash rate for API, added HS term in addition to KHS. 
+Added blocks solved to console display and API.
+
+v3.8.3.1
+
+Fixed regression in v3.8.3 that broke several algos.
+
+v3.8.3
+
+More restoration of lost lyra2 hash.
+8 way AVX2 and 4way AVX optimization for blakecoin, vanilla & blake2s.
+8 way AVX2 for lbry.
+Scaled hashrate for API output.
+A couple of GBT fixes.
+
+v3.8.2.1
+
+Fixed low difficulty rejects with allium.
+Fixed qubit AVX2.
+Restored lyra2z lost hash.
+Fixed build.sh
+
+v3.8.2
+
+Fixed and faster myr-gr.
+Added x12 algo (Galaxie Cash), allium algo (Garlicoin).
+Faster lyra2rev2, lbry, skein.
+Large reduction in compiler warnings.
+
+v3.8.1.1
+
+Fixed Windows AVX2 crash.
+
+v3.8.1
+
+Fixes x16r on CPUs with only SSE2.
+More Optimizations for X algos, qubit & deep.
+Corrected algo optimizations for scrypt and yescrypt, no new optimizations.
+
+v3.8.0.1
+
+Fixed x16r AVX2 low hash rate.
+
+v3.8.0
+
+4way no longer a seperate feature, included in AVX2.
+Added x16r algo for Ravencoin, anime algo for Animecoin.
+More 4way optimizations for X13 and up.
+Tweaked CPU affinity to better support more than 64 CPUs.
+Fixed compile problem on some old AMD CPUs.
+
+v3.7.10
+
+4way optimizations for lyra2rev2, lyra2h, quark, timetravel8, timetravel10
+   x11evo, blakecoin.
+Faster x13sm3 (hsr).
+Added share difficulty to accepted message.
+
 v3.7.9

 Partial 4way optimizations for veltor, skunk, polytimos, lyra2z.
--- a/algo-gate-api.c
+++ b/algo-gate-api.c
@@ -16,7 +16,7 @@
 #include <memory.h>
 #include <unistd.h>
 #include <openssl/sha.h>
-#include "miner.h"
+//#include "miner.h"
 #include "algo-gate-api.h"

 // Define null and standard functions.
@@ -119,9 +119,11 @@ void init_algo_gate( algo_gate_t* gate )
   gate->gen_merkle_root         = (void*)&sha256d_gen_merkle_root;
   gate->stratum_gen_work        = (void*)&std_stratum_gen_work;
   gate->build_stratum_request   = (void*)&std_le_build_stratum_request;
+   gate->malloc_txs_request      = (void*)&std_malloc_txs_request;
   gate->set_target              = (void*)&std_set_target;
   gate->work_decode             = (void*)&std_le_work_decode;
   gate->submit_getwork_result   = (void*)&std_le_submit_getwork_result;
+   gate->build_block_header      = (void*)&std_build_block_header;
   gate->build_extraheader       = (void*)&std_build_extraheader;
   gate->set_work_data_endian    = (void*)&do_nothing;
   gate->calc_network_diff       = (void*)&std_calc_network_diff;
@@ -155,6 +157,8 @@ bool register_algo_gate( int algo, algo_gate_t *gate )

   switch (algo)
   {
+     case ALGO_ALLIUM:       register_allium_algo      ( gate ); break;
+     case ALGO_ANIME:        register_anime_algo       ( gate ); break;
     case ALGO_ARGON2:       register_argon2_algo      ( gate ); break;
     case ALGO_AXIOM:        register_axiom_algo       ( gate ); break;
     case ALGO_BASTION:      register_bastion_algo     ( gate ); break;
@@ -212,10 +216,12 @@ bool register_algo_gate( int algo, algo_gate_t *gate )
     case ALGO_X11:          register_x11_algo         ( gate ); break;
     case ALGO_X11EVO:       register_x11evo_algo      ( gate ); break;
     case ALGO_X11GOST:      register_x11gost_algo     ( gate ); break;
+     case ALGO_X12:          register_x12_algo         ( gate ); break;
     case ALGO_X13:          register_x13_algo         ( gate ); break;
     case ALGO_X13SM3:       register_x13sm3_algo      ( gate ); break;
     case ALGO_X14:          register_x14_algo         ( gate ); break;
     case ALGO_X15:          register_x15_algo         ( gate ); break;
+     case ALGO_X16R:         register_x16r_algo        ( gate ); break;
     case ALGO_X17:          register_x17_algo         ( gate ); break;
     case ALGO_XEVAN:        register_xevan_algo       ( gate ); break;
     case ALGO_YESCRYPT:     register_yescrypt_algo    ( gate ); break;
@@ -296,6 +302,7 @@ const char* const algo_alias_map[][2] =
  { "lyra2",             "lyra2re"      },
  { "lyra2v2",           "lyra2rev2"    },
  { "lyra2zoin",         "lyra2z330"    },
+  { "myrgr",             "myr-gr"       },
  { "myriad",            "myr-gr"       },
  { "neo",               "neoscrypt"    },
  { "phi",               "phi1612"      },
--- a/algo-gate-api.h
+++ b/algo-gate-api.h
@@ -1,7 +1,6 @@
 #include <stdlib.h>
 #include <stdbool.h>
 #include <stdint.h>
-
 #include "miner.h"

 /////////////////////////////
@@ -91,7 +90,7 @@ typedef  uint32_t set_t;
 #define AVX_OPT         4
 #define AVX2_OPT        8
 #define SHA_OPT      0x10
-#define FOUR_WAY_OPT 0x20
+//#define FOUR_WAY_OPT 0x20

 // return set containing all elements from sets a & b
 inline set_t set_union ( set_t a, set_t b ) { return a | b; }
@@ -128,7 +127,10 @@ void ( *set_target)              ( struct work*, double );
 bool ( *submit_getwork_result )  ( CURL*, struct work* );
 void ( *gen_merkle_root )        ( char*, struct stratum_ctx* );
 void ( *build_extraheader )      ( struct work*, struct stratum_ctx* );
+void ( *build_block_header )     ( struct work*, uint32_t, uint32_t*,
+                                   uint32_t*, uint32_t, uint32_t );
 void ( *build_stratum_request )  ( char*, struct work*, struct stratum_ctx* );
+char* ( *malloc_txs_request )    ( struct work* );
 void ( *set_work_data_endian )   ( struct work* );
 double ( *calc_network_diff )    ( struct work* );
 bool ( *ready_to_mine )          ( struct work*, struct stratum_ctx*, int );
@@ -213,7 +215,8 @@ int64_t get_max64_0x3fffffLL();
 int64_t get_max64_0x1ffff();
 int64_t get_max64_0xffffLL();

-void std_set_target   ( struct work *work, double job_diff );
+void std_set_target(    struct work *work, double job_diff );
+void alt_set_target(    struct work* work, double job_diff );
 void scrypt_set_target( struct work *work, double job_diff );

 bool std_le_work_decode( const json_t *val, struct work *work );
@@ -228,11 +231,17 @@ void std_le_build_stratum_request( char *req, struct work *work );
 void std_be_build_stratum_request( char *req, struct work *work );
 void jr2_build_stratum_request   ( char *req, struct work *work );

+char* std_malloc_txs_request( struct work *work );
+
 // Default is do_nothing (assumed LE)
 void set_work_data_big_endian( struct work *work );

 double std_calc_network_diff( struct work *work );

+void std_build_block_header( struct work* g_work, uint32_t version,
+                             uint32_t *prevhash, uint32_t *merkle_root,
+                             uint32_t ntime, uint32_t nbits );
+
 void std_build_extraheader( struct work *work, struct stratum_ctx *sctx );

 json_t* std_longpoll_rpc_call( CURL *curl, int *err, char *lp_url );
--- a/algo/blake/blake-4way.c
+++ b/algo/blake/blake-4way.c
@@ -1,5 +1,4 @@
 #include "blake-gate.h"
-#include "sph_blake.h"
 #include "blake-hash-4way.h"
 #include <string.h>
 #include <stdint.h>
@@ -7,25 +6,16 @@

 #if defined (BLAKE_4WAY)

+blake256r14_4way_context blake_4w_ctx;
+
 void blakehash_4way(void *state, const void *input)
 {
-     uint32_t vhash[4*4] __attribute__ ((aligned (64)));
-     uint32_t hash0[4] __attribute__ ((aligned (32)));
-     uint32_t hash1[4] __attribute__ ((aligned (32)));
-     uint32_t hash2[4] __attribute__ ((aligned (32)));
-     uint32_t hash3[4] __attribute__ ((aligned (32)));
-     blake256_4way_context ctx;
-
-     blake256_4way_init( &ctx );
-     blake256_4way( &ctx, input, 16 );
-     blake256_4way_close( &ctx, vhash );
-
-     mm_deinterleave_4x32( hash0, hash1, hash2, hash3, vhash, 256 );
-
-     memcpy( state,    hash0, 32 );
-     memcpy( state+32, hash1, 32 );
-     memcpy( state+64, hash1, 32 );
-     memcpy( state+96, hash1, 32 );
+     uint32_t vhash[8*4] __attribute__ ((aligned (64)));
+     blake256r14_4way_context ctx;
+     memcpy( &ctx, &blake_4w_ctx, sizeof ctx );
+     blake256r14_4way( &ctx, input + (64<<2), 16 );
+     blake256r14_4way_close( &ctx, vhash );
+     mm_deinterleave_4x32( state, state+32, state+64, state+96, vhash, 256 );
 }

 int scanhash_blake_4way( int thr_id, struct work *work, uint32_t max_nonce,
@@ -36,24 +26,23 @@ int scanhash_blake_4way( int thr_id, struct work *work, uint32_t max_nonce,
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
   const uint32_t first_nonce = pdata[19];
-//   uint32_t HTarget = ptarget[7];
+   uint32_t HTarget = ptarget[7];
   uint32_t _ALIGN(32) edata[20];
   uint32_t n = first_nonce;
   uint32_t *nonces = work->nonces;
-   bool *found = work->nfound;
   int num_found = 0;

-//   if (opt_benchmark)
-//      HTarget = 0x7f;
+   if (opt_benchmark)
+      HTarget = 0x7f;

   // we need big endian data...
   swab32_array( edata, pdata, 20 );
-
   mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
+   blake256r14_4way_init( &blake_4w_ctx );
+   blake256r14_4way( &blake_4w_ctx, vdata, 64 );

   uint32_t *noncep = vdata + 76;   // 19*4
   do {
-      found[0] = found[1] = found[2] = found[3] = false;
      be32enc( noncep,    n   );
      be32enc( noncep +1, n+1 );
      be32enc( noncep +2, n+2 );
@@ -61,45 +50,14 @@ int scanhash_blake_4way( int thr_id, struct work *work, uint32_t max_nonce,

      blakehash_4way( hash, vdata );

-      if ( hash[7] == 0 )
+      for ( int i = 0; i < 4; i++ )
+      if (  (hash+(i<<3))[7] <= HTarget && fulltest( hash+(i<<3), ptarget ) )
      {
-         if ( fulltest( hash, ptarget ) )
-         {
-             found[0] = true;
-             num_found++;
-             nonces[0] = n;
-             pdata[19] = n;
-         }
+          pdata[19] = n+i;
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
      }
-      if ( (hash+8)[7] == 0 ) 
-      {
-         if ( fulltest( hash+8, ptarget ) ) 
-         {
-             found[1] = true;
-             num_found++;
-             nonces[1] = n+1;
-         }
-      }
-      if ( (hash+16)[7] == 0 )
-      {
-          if ( fulltest( hash+8, ptarget ) )
-          {
-              found[2] = true;
-              num_found++;
-              nonces[2] = n+2;
-          }
-      }
-      if ( (hash+24)[7] == 0 )
-      {
-         if ( fulltest( hash+8, ptarget ) )
-         {
-              found[3] = true;
-              num_found++;
-              nonces[3] = n+3;
-         }
-      }
-       n += 4;
-      *hashes_done = n - first_nonce + 1;
+      n += 4;

   } while ( (num_found == 0) && (n < max_nonce) 
             && !work_restart[thr_id].restart );
@@ -110,3 +68,77 @@ int scanhash_blake_4way( int thr_id, struct work *work, uint32_t max_nonce,

 #endif

+#if defined(BLAKE_8WAY)
+
+blake256r14_8way_context blake_8w_ctx;
+
+void blakehash_8way( void *state, const void *input )
+{
+     uint32_t vhash[8*8] __attribute__ ((aligned (64)));
+     blake256r14_8way_context ctx;
+     memcpy( &ctx, &blake_8w_ctx, sizeof ctx );
+     blake256r14_8way( &ctx, input + (64<<3), 16 );
+     blake256r14_8way_close( &ctx, vhash );
+     mm256_deinterleave_8x32( state,     state+ 32, state+ 64, state+ 96,
+                              state+128, state+160, state+192, state+224,
+                              vhash, 256 );
+}
+
+int scanhash_blake_8way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done )
+{
+   uint32_t vdata[20*8] __attribute__ ((aligned (64)));
+   uint32_t hash[8*8] __attribute__ ((aligned (32)));
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   uint32_t HTarget = ptarget[7];
+   uint32_t _ALIGN(32) edata[20];
+   uint32_t n = first_nonce;
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+
+   if (opt_benchmark)
+      HTarget = 0x7f;
+
+   // we need big endian data...
+   swab32_array( edata, pdata, 20 );
+
+   mm256_interleave_8x32( vdata, edata, edata, edata, edata,
+                                 edata, edata, edata, edata, 640 );
+
+   blake256r14_8way_init( &blake_8w_ctx );
+   blake256r14_8way( &blake_8w_ctx, vdata, 64 );
+
+   uint32_t *noncep = vdata + 152;   // 19*8
+   do {
+      be32enc( noncep,    n   );
+      be32enc( noncep +1, n+1 );
+      be32enc( noncep +2, n+2 );
+      be32enc( noncep +3, n+3 );
+      be32enc( noncep +4, n+4 );
+      be32enc( noncep +5, n+5 );
+      be32enc( noncep +6, n+6 );
+      be32enc( noncep +7, n+7 );
+      pdata[19] = n;
+
+      blakehash_8way( hash, vdata );
+
+      for ( int i = 0; i < 8; i++ )
+      if ( (hash+i)[7] <= HTarget && fulltest( hash+i, ptarget ) )
+      {
+          pdata[19] = n+i;
+          num_found++;
+          nonces[i] = n+i;
+          work_set_target_ratio( work, hash+1 );
+      }
+      n += 8;
+
+   } while ( (num_found == 0) && (n < max_nonce)
+             && !work_restart[thr_id].restart );
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#endif
--- a/algo/blake/blake-gate.c
+++ b/algo/blake/blake-gate.c
@@ -7,6 +7,7 @@ int64_t blake_get_max64 ()

 bool register_blake_algo( algo_gate_t* gate )
 {
+  gate->optimizations = AVX2_OPT;
  gate->get_max64 = (void*)&blake_get_max64;
 //#if defined (__AVX2__) && defined (FOUR_WAY)
 //   gate->optimizations = SSE2_OPT | AVX_OPT | AVX2_OPT;
@@ -14,7 +15,6 @@ bool register_blake_algo( algo_gate_t* gate )
 //  gate->hash      = (void*)&blakehash_8way;
 #if defined(BLAKE_4WAY)
  four_way_not_tested();
-  gate->optimizations = FOUR_WAY_OPT;
  gate->scanhash  = (void*)&scanhash_blake_4way;
  gate->hash      = (void*)&blakehash_4way;
 #else
--- a/algo/blake/blake-gate.h
+++ b/algo/blake/blake-gate.h
@@ -4,7 +4,7 @@
 #include "algo-gate-api.h"
 #include <stdint.h>

-#if defined(FOUR_WAY) && defined(__AVX__)
+#if defined(__AVX2__)
  #define BLAKE_4WAY
 #endif

--- a/algo/blake/blake-hash-4way.c
+++ b/algo/blake/blake-hash-4way.c
@@ -58,6 +58,8 @@ extern "C"{
 #pragma warning (disable: 4146)
 #endif

+// Blake-256
+
 static const sph_u32 IV256[8] = {
 	SPH_C32(0x6A09E667), SPH_C32(0xBB67AE85),
 	SPH_C32(0x3C6EF372), SPH_C32(0xA54FF53A),
@@ -67,6 +69,8 @@ static const sph_u32 IV256[8] = {

 #if defined (__AVX2__)

+// Blake-512
+
 static const sph_u64 IV512[8] = {
 	SPH_C64(0x6A09E667F3BCC908), SPH_C64(0xBB67AE8584CAA73B),
 	SPH_C64(0x3C6EF372FE94F82B), SPH_C64(0xA54FF53A5F1D36F1),
@@ -78,6 +82,8 @@ static const sph_u64 IV512[8] = {

 #if SPH_COMPACT_BLAKE_32 || SPH_COMPACT_BLAKE_64

+// Blake-256 4 & 8 way, Blake-512 4 way
+
 static const unsigned sigma[16][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 },
@@ -273,6 +279,8 @@ static const unsigned sigma[16][16] = {
 #define Mx_(n)      Mx__(n)
 #define Mx__(n)     M ## n

+// Blake-256 4 & 8 way
+
 #define CSx(r, i)   CSx_(Z ## r ## i)
 #define CSx_(n)     CSx__(n)
 #define CSx__(n)    CS ## n
@@ -311,6 +319,8 @@ static const sph_u32 CS[16] = {

 #if defined(__AVX2__)

+// Blake-512 4 way
+
 #define CBx(r, i)   CBx_(Z ## r ## i)
 #define CBx_(n)     CBx__(n)
 #define CBx__(n)    CB ## n
@@ -365,6 +375,8 @@ do { \

 #if SPH_COMPACT_BLAKE_32

+// Blake-256 4 way
+
 #define ROUND_S_4WAY(r)   do { \
 	GS_4WAY(M[sigma[r][0x0]], M[sigma[r][0x1]], \
 		CS[sigma[r][0x0]], CS[sigma[r][0x1]], V0, V4, V8, VC); \
@@ -401,6 +413,35 @@ do { \

 #if defined (__AVX2__)

+// Blake-256 8 way
+
+#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 ); \
+   d = mm256_rotr_32( _mm256_xor_si256( d, a ), 16 ); \
+   c = _mm256_add_epi32( c, d ); \
+   b = mm256_rotr_32( _mm256_xor_si256( b, c ), 12 ); \
+   a = _mm256_add_epi32( _mm256_add_epi32( _mm256_xor_si256( \
+                 _mm256_set1_epi32( c0 ), m1 ), b ), a ); \
+   d = mm256_rotr_32( _mm256_xor_si256( d, a ), 8 ); \
+   c = _mm256_add_epi32( c, d ); \
+   b = mm256_rotr_32( _mm256_xor_si256( b, c ), 7 ); \
+} while (0)
+
+#define ROUND_S_8WAY(r)   do { \
+        GS_8WAY(Mx(r, 0), Mx(r, 1), CSx(r, 0), CSx(r, 1), V0, V4, V8, VC); \
+        GS_8WAY(Mx(r, 2), Mx(r, 3), CSx(r, 2), CSx(r, 3), V1, V5, V9, VD); \
+        GS_8WAY(Mx(r, 4), Mx(r, 5), CSx(r, 4), CSx(r, 5), V2, V6, VA, VE); \
+        GS_8WAY(Mx(r, 6), Mx(r, 7), CSx(r, 6), CSx(r, 7), V3, V7, VB, VF); \
+        GS_8WAY(Mx(r, 8), Mx(r, 9), CSx(r, 8), CSx(r, 9), V0, V5, VA, VF); \
+        GS_8WAY(Mx(r, A), Mx(r, B), CSx(r, A), CSx(r, B), V1, V6, VB, VC); \
+        GS_8WAY(Mx(r, C), Mx(r, D), CSx(r, C), CSx(r, D), V2, V7, V8, VD); \
+        GS_8WAY(Mx(r, E), Mx(r, F), CSx(r, E), CSx(r, F), V3, V4, V9, VE); \
+} while (0)
+
+// Blake-512 4 way
+
 #define GB_4WAY(m0, m1, c0, c1, a, b, c, d)   do { \
   a = _mm256_add_epi64( _mm256_add_epi64( _mm256_xor_si256( \
                 _mm256_set_epi64x( c1, c1, c1, c1 ), m0 ), b ), a ); \
@@ -452,6 +493,8 @@ do { \

 #endif

+// Blake-256 4 way
+
 #define DECL_STATE32_4WAY \
 	__m128i H0, H1, H2, H3, H4, H5, H6, H7; \
 	__m128i S0, S1, S2, S3; \
@@ -491,14 +534,10 @@ do { \
 		(state)->T1 = T1; \
 	} while (0)

-//#define BLAKE32_ROUNDS 8
-#ifndef BLAKE32_ROUNDS
-#define BLAKE32_ROUNDS 14
-#endif
-
 #if SPH_COMPACT_BLAKE_32
+// not used

-#define COMPRESS32_4WAY   do { \
+#define COMPRESS32_4WAY( rounds )   do { \
 	__m128i M[16]; \
 	__m128i V0, V1, V2, V3, V4, V5, V6, V7; \
 	__m128i V8, V9, VA, VB, VC, VD, VE, VF; \
@@ -523,23 +562,23 @@ do { \
                          , _mm_set_epi32( CS6, CS6, CS6, CS6 ) ); \
        VF = _mm_xor_si128( _mm_set_epi32( T1, T1, T1, T1 ), \
                            _mm_set_epi32( CS7, CS7, CS7, CS7 ) ); \
-	M[0x0] = mm_byteswap_32( *(buf +  0) ); \
-	M[0x1] = mm_byteswap_32( *(buf +  1) ); \
-	M[0x2] = mm_byteswap_32( *(buf +  2) ); \
-	M[0x3] = mm_byteswap_32( *(buf +  3) ); \
-	M[0x4] = mm_byteswap_32( *(buf +  4) ); \
-	M[0x5] = mm_byteswap_32( *(buf +  5) ); \
-	M[0x6] = mm_byteswap_32( *(buf +  6) ); \
-	M[0x7] = mm_byteswap_32( *(buf +  7) ); \
-	M[0x8] = mm_byteswap_32( *(buf +  8) ); \
-	M[0x9] = mm_byteswap_32( *(buf +  9) ); \
-	M[0xA] = mm_byteswap_32( *(buf + 10) ); \
-	M[0xB] = mm_byteswap_32( *(buf + 11) ); \
-	M[0xC] = mm_byteswap_32( *(buf + 12) ); \
-	M[0xD] = mm_byteswap_32( *(buf + 13) ); \
-	M[0xE] = mm_byteswap_32( *(buf + 14) ); \
-	M[0xF] = mm_byteswap_32( *(buf + 15) ); \
-	for (r = 0; r < BLAKE32_ROUNDS; r ++) \
+	M[0x0] = mm_bswap_32( *(buf +  0) ); \
+	M[0x1] = mm_bswap_32( *(buf +  1) ); \
+	M[0x2] = mm_bswap_32( *(buf +  2) ); \
+	M[0x3] = mm_bswap_32( *(buf +  3) ); \
+	M[0x4] = mm_bswap_32( *(buf +  4) ); \
+	M[0x5] = mm_bswap_32( *(buf +  5) ); \
+	M[0x6] = mm_bswap_32( *(buf +  6) ); \
+	M[0x7] = mm_bswap_32( *(buf +  7) ); \
+	M[0x8] = mm_bswap_32( *(buf +  8) ); \
+	M[0x9] = mm_bswap_32( *(buf +  9) ); \
+	M[0xA] = mm_bswap_32( *(buf + 10) ); \
+	M[0xB] = mm_bswap_32( *(buf + 11) ); \
+	M[0xC] = mm_bswap_32( *(buf + 12) ); \
+	M[0xD] = mm_bswap_32( *(buf + 13) ); \
+	M[0xE] = mm_bswap_32( *(buf + 14) ); \
+	M[0xF] = mm_bswap_32( *(buf + 15) ); \
+	for (r = 0; r < rounds; r ++) \
 		ROUND_S_4WAY(r); \
        H0 = _mm_xor_si128( _mm_xor_si128( \
                                   _mm_xor_si128( S0, V0 ), V8 ), H0 ); \
@@ -563,85 +602,193 @@ do { \

 // current impl

-#define COMPRESS32_4WAY   do { \
-	__m128i M0, M1, M2, M3, M4, M5, M6, M7; \
-	__m128i M8, M9, MA, MB, MC, MD, ME, MF; \
-	__m128i V0, V1, V2, V3, V4, V5, V6, V7; \
-	__m128i V8, V9, VA, VB, VC, VD, VE, VF; \
-	V0 = H0; \
-	V1 = H1; \
-	V2 = H2; \
-	V3 = H3; \
-	V4 = H4; \
-	V5 = H5; \
-	V6 = H6; \
-	V7 = H7; \
-        V8 = _mm_xor_si128( S0, _mm_set_epi32( CS0, CS0, CS0, CS0 ) ); \
-        V9 = _mm_xor_si128( S1, _mm_set_epi32( CS1, CS1, CS1, CS1 ) ); \
-        VA = _mm_xor_si128( S2, _mm_set_epi32( CS2, CS2, CS2, CS2 ) ); \
-        VB = _mm_xor_si128( S3, _mm_set_epi32( CS3, CS3, CS3, CS3 ) ); \
-        VC = _mm_xor_si128( _mm_set_epi32( T0, T0, T0, T0 ), \
-                            _mm_set_epi32( CS4, CS4, CS4, CS4 ) ); \
-        VD = _mm_xor_si128( _mm_set_epi32( T0, T0, T0, T0 ), \
-                            _mm_set_epi32( CS5, CS5, CS5, CS5 ) ); \
-        VE = _mm_xor_si128( _mm_set_epi32( T1, T1, T1, T1 ), \
-                            _mm_set_epi32( CS6, CS6, CS6, CS6 ) ); \
-        VF = _mm_xor_si128( _mm_set_epi32( T1, T1, T1, T1 ), \
-                            _mm_set_epi32( CS7, CS7, CS7, CS7 ) ); \
-	M0 = mm_byteswap_32( * buf ); \
-	M1 = mm_byteswap_32( *(buf+1) ); \
-	M2 = mm_byteswap_32( *(buf+2) ); \
-	M3 = mm_byteswap_32( *(buf+3) ); \
-	M4 = mm_byteswap_32( *(buf+4) ); \
-	M5 = mm_byteswap_32( *(buf+5) ); \
-	M6 = mm_byteswap_32( *(buf+6) ); \
-	M7 = mm_byteswap_32( *(buf+7) ); \
-	M8 = mm_byteswap_32( *(buf+8) ); \
-	M9 = mm_byteswap_32( *(buf+9) ); \
-	MA = mm_byteswap_32( *(buf+10) ); \
-	MB = mm_byteswap_32( *(buf+11) ); \
-	MC = mm_byteswap_32( *(buf+12) ); \
-	MD = mm_byteswap_32( *(buf+13) ); \
-	ME = mm_byteswap_32( *(buf+14) ); \
-	MF = mm_byteswap_32( *(buf+15) ); \
-	ROUND_S_4WAY(0); \
-	ROUND_S_4WAY(1); \
-	ROUND_S_4WAY(2); \
-	ROUND_S_4WAY(3); \
-	ROUND_S_4WAY(4); \
-	ROUND_S_4WAY(5); \
-	ROUND_S_4WAY(6); \
-	ROUND_S_4WAY(7); \
-	if (BLAKE32_ROUNDS == 14) { \
-	ROUND_S_4WAY(8); \
-	ROUND_S_4WAY(9); \
-	ROUND_S_4WAY(0); \
-	ROUND_S_4WAY(1); \
-	ROUND_S_4WAY(2); \
-	ROUND_S_4WAY(3); \
-	} \
-        H0 = _mm_xor_si128( _mm_xor_si128( \
-                                _mm_xor_si128( V8, V0 ), S0 ), H0 ); \
-        H1 = _mm_xor_si128( _mm_xor_si128( \
-                                _mm_xor_si128( V9, V1 ), S1 ), H1 ); \
-        H2 = _mm_xor_si128( _mm_xor_si128( \
-                                _mm_xor_si128( VA, V2 ), S2 ), H2 ); \
-        H3 = _mm_xor_si128( _mm_xor_si128( \
-                                _mm_xor_si128( VB, V3 ), S3 ), H3 ); \
-        H4 = _mm_xor_si128( _mm_xor_si128( \
-                                _mm_xor_si128( VC, V4 ), S0 ), H4 ); \
-        H5 = _mm_xor_si128( _mm_xor_si128( \
-                                _mm_xor_si128( VD, V5 ), S1 ), H5 ); \
-        H6 = _mm_xor_si128( _mm_xor_si128( \
-                                _mm_xor_si128( VE, V6 ), S2 ), H6 ); \
-        H7 = _mm_xor_si128( _mm_xor_si128( \
-                                _mm_xor_si128( VF, V7 ), S3 ), H7 ); \
-	} while (0)
+#define COMPRESS32_4WAY( rounds ) \
+do { \
+   __m128i M0, M1, M2, M3, M4, M5, M6, M7; \
+   __m128i M8, M9, MA, MB, MC, MD, ME, MF; \
+   __m128i V0, V1, V2, V3, V4, V5, V6, V7; \
+   __m128i V8, V9, VA, VB, VC, VD, VE, VF; \
+   V0 = H0; \
+   V1 = H1; \
+   V2 = H2; \
+   V3 = H3; \
+   V4 = H4; \
+   V5 = H5; \
+   V6 = H6; \
+   V7 = H7; \
+   V8 = _mm_xor_si128( S0, _mm_set_epi32( CS0, CS0, CS0, CS0 ) ); \
+   V9 = _mm_xor_si128( S1, _mm_set_epi32( CS1, CS1, CS1, CS1 ) ); \
+   VA = _mm_xor_si128( S2, _mm_set_epi32( CS2, CS2, CS2, CS2 ) ); \
+   VB = _mm_xor_si128( S3, _mm_set_epi32( CS3, CS3, CS3, CS3 ) ); \
+   VC = _mm_xor_si128( _mm_set1_epi32( T0 ), _mm_set1_epi32( CS4 ) ); \
+   VD = _mm_xor_si128( _mm_set1_epi32( T0 ), _mm_set1_epi32( CS5 ) ); \
+   VE = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS6 ) ); \
+   VF = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS7 ) ); \
+   M0 = mm_bswap_32( * buf ); \
+   M1 = mm_bswap_32( *(buf+1) ); \
+   M2 = mm_bswap_32( *(buf+2) ); \
+   M3 = mm_bswap_32( *(buf+3) ); \
+   M4 = mm_bswap_32( *(buf+4) ); \
+   M5 = mm_bswap_32( *(buf+5) ); \
+   M6 = mm_bswap_32( *(buf+6) ); \
+   M7 = mm_bswap_32( *(buf+7) ); \
+   M8 = mm_bswap_32( *(buf+8) ); \
+   M9 = mm_bswap_32( *(buf+9) ); \
+   MA = mm_bswap_32( *(buf+10) ); \
+   MB = mm_bswap_32( *(buf+11) ); \
+   MC = mm_bswap_32( *(buf+12) ); \
+   MD = mm_bswap_32( *(buf+13) ); \
+   ME = mm_bswap_32( *(buf+14) ); \
+   MF = mm_bswap_32( *(buf+15) ); \
+   ROUND_S_4WAY(0); \
+   ROUND_S_4WAY(1); \
+   ROUND_S_4WAY(2); \
+   ROUND_S_4WAY(3); \
+   ROUND_S_4WAY(4); \
+   ROUND_S_4WAY(5); \
+   ROUND_S_4WAY(6); \
+   ROUND_S_4WAY(7); \
+   if (rounds == 14) \
+   { \
+      ROUND_S_4WAY(8); \
+      ROUND_S_4WAY(9); \
+      ROUND_S_4WAY(0); \
+      ROUND_S_4WAY(1); \
+      ROUND_S_4WAY(2); \
+      ROUND_S_4WAY(3); \
+   } \
+   H0 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( V8, V0 ), S0 ), H0 ); \
+   H1 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( V9, V1 ), S1 ), H1 ); \
+   H2 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VA, V2 ), S2 ), H2 ); \
+   H3 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VB, V3 ), S3 ), H3 ); \
+   H4 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VC, V4 ), S0 ), H4 ); \
+   H5 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VD, V5 ), S1 ), H5 ); \
+   H6 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VE, V6 ), S2 ), H6 ); \
+   H7 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VF, V7 ), S3 ), H7 ); \
+} while (0)

 #endif

 #if defined (__AVX2__)

+// Blake-256 8 way
+
+#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) \
+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_STATE32_8WAY(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)
+
+#define COMPRESS32_8WAY( rounds ) \
+do { \
+   __m256i M0, M1, M2, M3, M4, M5, M6, M7; \
+   __m256i M8, M9, MA, MB, MC, MD, ME, MF; \
+   __m256i V0, V1, V2, V3, V4, V5, V6, V7; \
+   __m256i V8, V9, VA, VB, VC, VD, VE, VF; \
+   V0 = H0; \
+   V1 = H1; \
+   V2 = H2; \
+   V3 = H3; \
+   V4 = H4; \
+   V5 = H5; \
+   V6 = H6; \
+   V7 = H7; \
+   V8 = _mm256_xor_si256( S0, _mm256_set1_epi32( CS0 ) ); \
+   V9 = _mm256_xor_si256( S1, _mm256_set1_epi32( CS1 ) ); \
+   VA = _mm256_xor_si256( S2, _mm256_set1_epi32( CS2 ) ); \
+   VB = _mm256_xor_si256( S3, _mm256_set1_epi32( CS3 ) ); \
+   VC = _mm256_xor_si256( _mm256_set1_epi32( T0 ), _mm256_set1_epi32( CS4 ) ); \
+   VD = _mm256_xor_si256( _mm256_set1_epi32( T0 ), _mm256_set1_epi32( CS5 ) ); \
+   VE = _mm256_xor_si256( _mm256_set1_epi32( T1 ), _mm256_set1_epi32( CS6 ) ); \
+   VF = _mm256_xor_si256( _mm256_set1_epi32( T1 ), _mm256_set1_epi32( CS7 ) ); \
+   M0 = mm256_bswap_32( * buf ); \
+   M1 = mm256_bswap_32( *(buf+1) ); \
+   M2 = mm256_bswap_32( *(buf+2) ); \
+   M3 = mm256_bswap_32( *(buf+3) ); \
+   M4 = mm256_bswap_32( *(buf+4) ); \
+   M5 = mm256_bswap_32( *(buf+5) ); \
+   M6 = mm256_bswap_32( *(buf+6) ); \
+   M7 = mm256_bswap_32( *(buf+7) ); \
+   M8 = mm256_bswap_32( *(buf+8) ); \
+   M9 = mm256_bswap_32( *(buf+9) ); \
+   MA = mm256_bswap_32( *(buf+10) ); \
+   MB = mm256_bswap_32( *(buf+11) ); \
+   MC = mm256_bswap_32( *(buf+12) ); \
+   MD = mm256_bswap_32( *(buf+13) ); \
+   ME = mm256_bswap_32( *(buf+14) ); \
+   MF = mm256_bswap_32( *(buf+15) ); \
+   ROUND_S_8WAY(0); \
+   ROUND_S_8WAY(1); \
+   ROUND_S_8WAY(2); \
+   ROUND_S_8WAY(3); \
+   ROUND_S_8WAY(4); \
+   ROUND_S_8WAY(5); \
+   ROUND_S_8WAY(6); \
+   ROUND_S_8WAY(7); \
+   if (rounds == 14) \
+   { \
+      ROUND_S_8WAY(8); \
+      ROUND_S_8WAY(9); \
+      ROUND_S_8WAY(0); \
+      ROUND_S_8WAY(1); \
+      ROUND_S_8WAY(2); \
+      ROUND_S_8WAY(3); \
+   } \
+   H0 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( V8, V0 ), \
+                                                              S0 ), H0 ); \
+   H1 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( V9, V1 ), \
+                                                              S1 ), H1 ); \
+   H2 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VA, V2 ), \
+                                                              S2 ), H2 ); \
+   H3 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VB, V3 ), \
+                                                              S3 ), H3 ); \
+   H4 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VC, V4 ), \
+                                                              S0 ), H4 ); \
+   H5 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VD, V5 ), \
+                                                              S1 ), H5 ); \
+   H6 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VE, V6 ), \
+                                                              S2 ), H6 ); \
+   H7 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VF, V7 ), \
+                                                              S3 ), H7 ); \
+} while (0)
+
+// Blake-512 4 way
+
 #define DECL_STATE64_4WAY \
 	__m256i H0, H1, H2, H3, H4, H5, H6, H7; \
        __m256i S0, S1, S2, S3; \
@@ -709,22 +856,22 @@ do { \
                               _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_byteswap_64( *(buf+0) ); \
-	M[0x1] = mm256_byteswap_64( *(buf+1) ); \
-	M[0x2] = mm256_byteswap_64( *(buf+2) ); \
-	M[0x3] = mm256_byteswap_64( *(buf+3) ); \
-	M[0x4] = mm256_byteswap_64( *(buf+4) ); \
-	M[0x5] = mm256_byteswap_64( *(buf+5) ); \
-	M[0x6] = mm256_byteswap_64( *(buf+6) ); \
-	M[0x7] = mm256_byteswap_64( *(buf+7) ); \
-	M[0x8] = mm256_byteswap_64( *(buf+8) ); \
-	M[0x9] = mm256_byteswap_64( *(buf+9) ); \
-	M[0xA] = mm256_byteswap_64( *(buf+10) ); \
-	M[0xB] = mm256_byteswap_64( *(buf+11) ); \
-	M[0xC] = mm256_byteswap_64( *(buf+12) ); \
-	M[0xD] = mm256_byteswap_64( *(buf+13) ); \
-	M[0xE] = mm256_byteswap_64( *(buf+14) ); \
-	M[0xF] = mm256_byteswap_64( *(buf+15) ); \
+	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( \
@@ -774,22 +921,22 @@ do { \
                            _mm256_set_epi64x( CB6, CB6, CB6, CB6 ) );  \
     VF = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
                            _mm256_set_epi64x( CB7, CB7, CB7, CB7 ) );  \
-     M0 = mm256_byteswap_64( *(buf + 0) ); \
-     M1 = mm256_byteswap_64( *(buf + 1) ); \
-     M2 = mm256_byteswap_64( *(buf + 2) ); \
-     M3 = mm256_byteswap_64( *(buf + 3) ); \
-     M4 = mm256_byteswap_64( *(buf + 4) ); \
-     M5 = mm256_byteswap_64( *(buf + 5) ); \
-     M6 = mm256_byteswap_64( *(buf + 6) ); \
-     M7 = mm256_byteswap_64( *(buf + 7) ); \
-     M8 = mm256_byteswap_64( *(buf + 8) ); \
-     M9 = mm256_byteswap_64( *(buf + 9) ); \
-     MA = mm256_byteswap_64( *(buf + 10) ); \
-     MB = mm256_byteswap_64( *(buf + 11) ); \
-     MC = mm256_byteswap_64( *(buf + 12) ); \
-     MD = mm256_byteswap_64( *(buf + 13) ); \
-     ME = mm256_byteswap_64( *(buf + 14) ); \
-     MF = mm256_byteswap_64( *(buf + 15) ); \
+     M0 = mm256_bswap_64( *(buf + 0) ); \
+     M1 = mm256_bswap_64( *(buf + 1) ); \
+     M2 = mm256_bswap_64( *(buf + 2) ); \
+     M3 = mm256_bswap_64( *(buf + 3) ); \
+     M4 = mm256_bswap_64( *(buf + 4) ); \
+     M5 = mm256_bswap_64( *(buf + 5) ); \
+     M6 = mm256_bswap_64( *(buf + 6) ); \
+     M7 = mm256_bswap_64( *(buf + 7) ); \
+     M8 = mm256_bswap_64( *(buf + 8) ); \
+     M9 = mm256_bswap_64( *(buf + 9) ); \
+     MA = mm256_bswap_64( *(buf + 10) ); \
+     MB = mm256_bswap_64( *(buf + 11) ); \
+     MC = mm256_bswap_64( *(buf + 12) ); \
+     MD = mm256_bswap_64( *(buf + 13) ); \
+     ME = mm256_bswap_64( *(buf + 14) ); \
+     MF = mm256_bswap_64( *(buf + 15) ); \
     ROUND_B_4WAY(0); \
     ROUND_B_4WAY(1); \
     ROUND_B_4WAY(2); \
@@ -828,19 +975,22 @@ do { \

 #endif

-static const sph_u32 salt_zero_small[4] = { 0, 0, 0, 0 };
+// Blake-256 4 way
+
+static const sph_u32 salt_zero_4way_small[4] = { 0, 0, 0, 0 };

 static void
 blake32_4way_init( blake_4way_small_context *sc, const sph_u32 *iv,
-                   const sph_u32 *salt)
+                   const sph_u32 *salt, int rounds )
 {
-        int i;
-        for ( i = 0; i < 8; i++ )
-           sc->H[i] = _mm_set1_epi32( iv[i] );
-        for ( i = 0; i < 4; i++ )
-           sc->S[i] = _mm_set1_epi32( salt[i] );
-	sc->T0 = sc->T1 = 0;
-	sc->ptr = 0;
+   int i;
+   for ( i = 0; i < 8; i++ )
+      sc->H[i] = _mm_set1_epi32( iv[i] );
+   for ( i = 0; i < 4; i++ )
+      sc->S[i] = _mm_set1_epi32( salt[i] );
+   sc->T0 = sc->T1 = 0;
+   sc->ptr = 0;
+   sc->rounds = rounds;
 }

 static void
@@ -848,52 +998,51 @@ blake32_4way( blake_4way_small_context *sc, const void *data, size_t len )
 {
   __m128i *vdata = (__m128i*)data;
   __m128i *buf;
-	size_t ptr;
-        const int buf_size = 64;   // number of elements, sizeof/4
-	DECL_STATE32_4WAY
+   size_t ptr;
+   const int buf_size = 64;   // number of elements, sizeof/4
+   DECL_STATE32_4WAY
+   buf = sc->buf;
+   ptr = sc->ptr;
+   if ( len < buf_size - ptr )
+   {
+      memcpy_128( buf + (ptr>>2), vdata, len>>2 );
+      ptr += len;
+      sc->ptr = ptr;
+      return;
+   }

-	buf = sc->buf;
-	ptr = sc->ptr;
-	if ( len < buf_size - ptr )
-        {
-		memcpy_128( buf + (ptr>>2), vdata, len>>2 );
-		ptr += len;
-		sc->ptr = ptr;
-		return;
-	}
+   READ_STATE32_4WAY(sc);
+   while ( len > 0 )
+   {
+      size_t clen;

-	READ_STATE32_4WAY(sc);
-	while ( len > 0 )
-        {
-           size_t clen;
-
-	   clen = buf_size - ptr;
-	   if (clen > len)
-		clen = len;
-	   memcpy_128( buf + (ptr>>2), vdata, clen>>2 );
-	   ptr += clen;
-           vdata += (clen>>2);
-	   len -= clen;
-	   if ( ptr == buf_size )
-           {
-		if ( ( T0 = SPH_T32(T0 + 512) ) < 512 )
-			T1 = SPH_T32(T1 + 1);
-		COMPRESS32_4WAY;
-		ptr = 0;
-	   }
-	}
-	WRITE_STATE32_4WAY(sc);
-	sc->ptr = ptr;
+      clen = buf_size - ptr;
+      if ( clen > len )
+         clen = len;
+      memcpy_128( buf + (ptr>>2), vdata, clen>>2 );
+      ptr += clen;
+      vdata += (clen>>2);
+      len -= clen;
+      if ( ptr == buf_size )
+      {
+         if ( ( T0 = SPH_T32(T0 + 512) ) < 512 )
+            T1 = SPH_T32(T1 + 1);
+         COMPRESS32_4WAY( sc->rounds );
+         ptr = 0;
+      }
+   }
+   WRITE_STATE32_4WAY(sc);
+   sc->ptr = ptr;
 }

 static void
 blake32_4way_close( blake_4way_small_context *sc, unsigned ub, unsigned n,
               void *dst, size_t out_size_w32 )
 {
-   union {
+//   union {
 	__m128i buf[16];
-	sph_u32 dummy;
-   } u;
+//	sph_u32 dummy;
+//   } u;
   size_t ptr, k;
   unsigned bit_len;
   sph_u32 th, tl;
@@ -901,7 +1050,7 @@ blake32_4way_close( blake_4way_small_context *sc, unsigned ub, unsigned n,

   ptr = sc->ptr;
   bit_len = ((unsigned)ptr << 3);
-   u.buf[ptr>>2] = _mm_set1_epi32( 0x80 );
+   buf[ptr>>2] = _mm_set1_epi32( 0x80 );
   tl = sc->T0 + bit_len;
   th = sc->T1;

@@ -920,34 +1069,156 @@ blake32_4way_close( blake_4way_small_context *sc, unsigned ub, unsigned n,

   if ( ptr <= 52 )
   {
-       memset_zero_128( u.buf + (ptr>>2) + 1, (52 - ptr) >> 2 );
+       memset_zero_128( buf + (ptr>>2) + 1, (52 - ptr) >> 2 );
       if (out_size_w32 == 8)
-           u.buf[52>>2] = _mm_or_si128( u.buf[52>>2],
+           buf[52>>2] = _mm_or_si128( buf[52>>2],
                                        _mm_set1_epi32( 0x01000000UL ) );
-       *(u.buf+(56>>2)) = mm_byteswap_32( _mm_set1_epi32( th ) );
-       *(u.buf+(60>>2)) = mm_byteswap_32( _mm_set1_epi32( tl ) );
-       blake32_4way( sc, u.buf + (ptr>>2), 64 - ptr );
+       *(buf+(56>>2)) = mm_bswap_32( _mm_set1_epi32( th ) );
+       *(buf+(60>>2)) = mm_bswap_32( _mm_set1_epi32( tl ) );
+       blake32_4way( sc, buf + (ptr>>2), 64 - ptr );
   }
   else
   {
-	memset_zero_128( u.buf + (ptr>>2) + 1, (60-ptr) >> 2 );
-	blake32_4way( sc, u.buf + (ptr>>2), 64 - ptr );
+	memset_zero_128( buf + (ptr>>2) + 1, (60-ptr) >> 2 );
+	blake32_4way( sc, buf + (ptr>>2), 64 - ptr );
 	sc->T0 = SPH_C32(0xFFFFFE00UL);
 	sc->T1 = SPH_C32(0xFFFFFFFFUL);
-	memset_zero_128( u.buf, 56>>2 );
+	memset_zero_128( buf, 56>>2 );
       if (out_size_w32 == 8)
-           u.buf[52>>2] = _mm_set1_epi32( 0x01000000UL );
-        *(u.buf+(56>>2)) = mm_byteswap_32( _mm_set1_epi32( th ) );
-        *(u.buf+(60>>2)) = mm_byteswap_32( _mm_set1_epi32( tl ) );
-	blake32_4way( sc, u.buf, 64 );
+           buf[52>>2] = _mm_set1_epi32( 0x01000000UL );
+        *(buf+(56>>2)) = mm_bswap_32( _mm_set1_epi32( th ) );
+        *(buf+(60>>2)) = mm_bswap_32( _mm_set1_epi32( tl ) );
+	blake32_4way( sc, buf, 64 );
   }
   out = (__m128i*)dst;
   for ( k = 0; k < out_size_w32; k++ )
-        out[k] = mm_byteswap_32( sc->H[k] );
+        out[k] = mm_bswap_32( sc->H[k] );
 }

 #if defined (__AVX2__)

+// Blake-256 8 way
+
+static const sph_u32 salt_zero_8way_small[8] = { 0, 0, 0, 0, 0, 0, 0, 0 };
+
+static void
+blake32_8way_init( blake_8way_small_context *sc, const sph_u32 *iv,
+                   const sph_u32 *salt, int rounds )
+{
+   int i;
+   for ( i = 0; i < 8; i++ )
+      sc->H[i] = _mm256_set1_epi32( iv[i] );
+   for ( i = 0; i < 4; i++ )
+      sc->S[i] = _mm256_set1_epi32( salt[i] );
+   sc->T0 = sc->T1 = 0;
+   sc->ptr = 0;
+   sc->rounds = rounds;
+}
+
+static void
+blake32_8way( blake_8way_small_context *sc, const void *data, size_t len )
+{
+   __m256i *vdata = (__m256i*)data;
+   __m256i *buf;
+   size_t ptr;
+   const int buf_size = 64;   // number of elements, sizeof/4
+   DECL_STATE32_8WAY
+   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_STATE32_8WAY(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 )
+      {
+          if ( ( T0 = SPH_T32(T0 + 512) ) < 512 )
+                T1 = SPH_T32(T1 + 1);
+          COMPRESS32_8WAY( sc->rounds );
+          ptr = 0;
+      }
+   }
+   WRITE_STATE32_8WAY(sc);
+   sc->ptr = ptr;
+}
+
+static void
+blake32_8way_close( blake_8way_small_context *sc, unsigned ub, unsigned n,
+                    void *dst, size_t out_size_w32 )
+{
+//   union {
+        __m256i buf[16];
+//        sph_u32 dummy;
+//   } u;
+   size_t ptr, k;
+   unsigned bit_len;
+   sph_u32 th, tl;
+   __m256i *out;
+
+   ptr = sc->ptr;
+   bit_len = ((unsigned)ptr << 3);
+   buf[ptr>>2] = _mm256_set1_epi32( 0x80 );
+   tl = sc->T0 + bit_len;
+   th = sc->T1;
+
+   if ( ptr == 0 )
+   {
+        sc->T0 = SPH_C32(0xFFFFFE00UL);
+        sc->T1 = SPH_C32(0xFFFFFFFFUL);
+   }
+   else if ( sc->T0 == 0 )
+   {
+        sc->T0 = SPH_C32(0xFFFFFE00UL) + bit_len;
+        sc->T1 = SPH_T32(sc->T1 - 1);
+   }
+   else
+        sc->T0 -= 512 - bit_len;
+
+   if ( ptr <= 52 )
+   {
+       memset_zero_256( buf + (ptr>>2) + 1, (52 - ptr) >> 2 );
+       if ( out_size_w32 == 8 )
+           buf[52>>2] = _mm256_or_si256( buf[52>>2],
+                                           _mm256_set1_epi32( 0x01000000UL ) );
+       *(buf+(56>>2)) = mm256_bswap_32( _mm256_set1_epi32( th ) );
+       *(buf+(60>>2)) = mm256_bswap_32( _mm256_set1_epi32( tl ) );
+       blake32_8way( sc, buf + (ptr>>2), 64 - ptr );
+   }
+   else
+   {
+        memset_zero_256( buf + (ptr>>2) + 1, (60-ptr) >> 2 );
+        blake32_8way( sc, buf + (ptr>>2), 64 - ptr );
+        sc->T0 = SPH_C32(0xFFFFFE00UL);
+        sc->T1 = SPH_C32(0xFFFFFFFFUL);
+        memset_zero_256( buf, 56>>2 );
+       if ( out_size_w32 == 8 )
+           buf[52>>2] = _mm256_set1_epi32( 0x01000000UL );
+        *(buf+(56>>2)) = mm256_bswap_32( _mm256_set1_epi32( th ) );
+        *(buf+(60>>2)) = mm256_bswap_32( _mm256_set1_epi32( tl ) );
+        blake32_8way( sc, buf, 64 );
+   }
+   out = (__m256i*)dst;
+   for ( k = 0; k < out_size_w32; k++ )
+        out[k] = mm256_bswap_32( sc->H[k] );
+}
+
+// Blake-512 4 way
+
 static const sph_u64 salt_zero_big[4] = { 0, 0, 0, 0 };

 static void
@@ -1011,10 +1282,10 @@ static void
 blake64_4way_close( blake_4way_big_context *sc,
 	unsigned ub, unsigned n, void *dst, size_t out_size_w64)
 {
-   union {
+//   union {
      __m256i buf[16];
-      sph_u64 dummy;
-   } u;
+//      sph_u64 dummy;
+//   } u;
   size_t ptr, k;
   unsigned bit_len;
   uint64_t z, zz;
@@ -1025,7 +1296,7 @@ blake64_4way_close( blake_4way_big_context *sc,
   bit_len = ((unsigned)ptr << 3);
   z = 0x80 >> n;
   zz = ((ub & -z) | z) & 0xFF;
-   u.buf[ptr>>3] = _mm256_set_epi64x( zz, zz, zz, zz );
+   buf[ptr>>3] = _mm256_set_epi64x( zz, zz, zz, zz );
   tl = sc->T0 + bit_len;
   th = sc->T1;
   if (ptr == 0 )
@@ -1044,45 +1315,48 @@ blake64_4way_close( blake_4way_big_context *sc,
   }
   if ( ptr <= 104 )
   {
-       memset_zero_256( u.buf + (ptr>>3) + 1, (104-ptr) >> 3 );
+       memset_zero_256( buf + (ptr>>3) + 1, (104-ptr) >> 3 );
       if ( out_size_w64 == 8 )
-          u.buf[(104>>3)] = _mm256_or_si256( u.buf[(104>>3)],
+          buf[(104>>3)] = _mm256_or_si256( buf[(104>>3)],
                                 _mm256_set1_epi64x( 0x0100000000000000ULL ) );
-       *(u.buf+(112>>3)) = mm256_byteswap_64(
+       *(buf+(112>>3)) = mm256_bswap_64(
                                    _mm256_set_epi64x( th, th, th, th ) );
-       *(u.buf+(120>>3)) = mm256_byteswap_64(
+       *(buf+(120>>3)) = mm256_bswap_64(
                                    _mm256_set_epi64x( tl, tl, tl, tl ) );

-       blake64_4way( sc, u.buf + (ptr>>3), 128 - ptr );
+       blake64_4way( sc, buf + (ptr>>3), 128 - ptr );
   }
   else
  {
-       memset_zero_256( u.buf + (ptr>>3) + 1, (120 - ptr) >> 3 );
+       memset_zero_256( buf + (ptr>>3) + 1, (120 - ptr) >> 3 );

-       blake64_4way( sc, u.buf + (ptr>>3), 128 - ptr );
+       blake64_4way( sc, buf + (ptr>>3), 128 - ptr );
       sc->T0 = SPH_C64(0xFFFFFFFFFFFFFC00ULL);
       sc->T1 = SPH_C64(0xFFFFFFFFFFFFFFFFULL);
-       memset_zero_256( u.buf, 112>>3 ); 
+       memset_zero_256( buf, 112>>3 ); 
       if ( out_size_w64 == 8 )
-           u.buf[104>>3] = _mm256_set1_epi64x( 0x0100000000000000ULL );
-       *(u.buf+(112>>3)) = mm256_byteswap_64(
+           buf[104>>3] = _mm256_set1_epi64x( 0x0100000000000000ULL );
+       *(buf+(112>>3)) = mm256_bswap_64(
                                    _mm256_set_epi64x( th, th, th, th ) );
-       *(u.buf+(120>>3)) = mm256_byteswap_64(
+       *(buf+(120>>3)) = mm256_bswap_64(
                                    _mm256_set_epi64x( tl, tl, tl, tl ) );

-       blake64_4way( sc, u.buf, 128 );
+       blake64_4way( sc, buf, 128 );
   }
   out = (__m256i*)dst;
   for ( k = 0; k < out_size_w64; k++ )
-       out[k] = mm256_byteswap_64( sc->H[k] );
+       out[k] = mm256_bswap_64( sc->H[k] );
 }

 #endif

+// Blake-256 4 way
+
+// default 14 rounds, backward copatibility
 void
 blake256_4way_init(void *cc)
 {
-	blake32_4way_init(cc, IV256, salt_zero_small);
+   blake32_4way_init( cc, IV256, salt_zero_4way_small, 14 );
 }

 void
@@ -1094,15 +1368,113 @@ blake256_4way(void *cc, const void *data, size_t len)
 void
 blake256_4way_close(void *cc, void *dst)
 {
-	blake256_4way_addbits_and_close(cc, 0, 0, dst);
+        blake32_4way_close(cc, 0, 0, dst, 8);
+}
+
+#if defined(__AVX2__)
+
+// Blake-256 8way
+
+void
+blake256_8way_init(void *cc)
+{
+   blake32_8way_init( cc, IV256, salt_zero_8way_small, 14 );
 }

 void
-blake256_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
+blake256_8way(void *cc, const void *data, size_t len)
 {
-	blake32_4way_close(cc, ub, n, dst, 8);
+        blake32_8way(cc, data, len);
 }

+void
+blake256_8way_close(void *cc, void *dst)
+{
+        blake32_8way_close(cc, 0, 0, dst, 8);
+}
+
+#endif
+
+// 14 rounds Blake, Decred
+void blake256r14_4way_init(void *cc)
+{
+   blake32_4way_init( cc, IV256, salt_zero_4way_small, 14 );
+}
+
+void
+blake256r14_4way(void *cc, const void *data, size_t len)
+{
+   blake32_4way(cc, data, len);
+}
+
+void
+blake256r14_4way_close(void *cc, void *dst)
+{
+   blake32_4way_close(cc, 0, 0, dst, 8);
+}
+
+#if defined(__AVX2__)
+
+void blake256r14_8way_init(void *cc)
+{
+   blake32_8way_init( cc, IV256, salt_zero_8way_small, 14 );
+}
+
+void
+blake256r14_8way(void *cc, const void *data, size_t len)
+{
+   blake32_8way(cc, data, len);
+}
+
+void
+blake256r14_8way_close(void *cc, void *dst)
+{
+   blake32_8way_close(cc, 0, 0, dst, 8);
+}
+
+#endif
+
+// 8 rounds Blakecoin, Vanilla
+void blake256r8_4way_init(void *cc)
+{
+   blake32_4way_init( cc, IV256, salt_zero_4way_small, 8 );
+}
+
+void
+blake256r8_4way(void *cc, const void *data, size_t len)
+{
+   blake32_4way(cc, data, len);
+}
+
+void
+blake256r8_4way_close(void *cc, void *dst)
+{
+   blake32_4way_close(cc, 0, 0, dst, 8);
+}
+
+#if defined (__AVX2__)
+
+void blake256r8_8way_init(void *cc)
+{
+   blake32_8way_init( cc, IV256, salt_zero_8way_small, 8 );
+}
+
+void
+blake256r8_8way(void *cc, const void *data, size_t len)
+{
+   blake32_8way(cc, data, len);
+}
+
+void
+blake256r8_8way_close(void *cc, void *dst)
+{
+   blake32_8way_close(cc, 0, 0, dst, 8);
+}
+
+#endif
+
+// Blake-512 4 way
+
 #if defined (__AVX2__)

 void
--- a/algo/blake/blake-hash-4way.h
+++ b/algo/blake/blake-hash-4way.h
@@ -35,7 +35,9 @@
 */

 #ifndef __BLAKE_HASH_4WAY__
-#define __BLAKE_HASH_4WAY___
+#define __BLAKE_HASH_4WAY__ 1
+
+#ifdef __AVX__

 #ifdef __cplusplus
 extern "C"{
@@ -45,47 +47,81 @@ extern "C"{
 #include "algo/sha/sph_types.h"
 #include "avxdefs.h"

-/**
- * Output size (in bits) for BLAKE-256.
- */
 #define SPH_SIZE_blake256   256

-#if SPH_64
-
-/**
- * Output size (in bits) for BLAKE-512.
- */
 #define SPH_SIZE_blake512   512

-#endif
+// With AVX only Blake-256 4 way is available.
+// With AVX2 Blake-256 8way & Blake-512 4 way are also available.
+
+// Blake-256 4 way

-#ifdef __AVX__
 typedef struct {
-        __m128i buf[16] __attribute__ ((aligned (64)));
-        __m128i H[8];
-        __m128i S[4];    
-        size_t ptr;
-	sph_u32 T0, T1;
+   __m128i buf[16] __attribute__ ((aligned (64)));
+   __m128i H[8];
+   __m128i S[4];    
+   size_t ptr;
+   sph_u32 T0, T1;
+   int rounds;   // 14 for blake, 8 for blakecoin & vanilla
 } blake_4way_small_context;

+// Default 14 rounds
 typedef blake_4way_small_context blake256_4way_context;
-
 void blake256_4way_init(void *cc);
 void blake256_4way(void *cc, const void *data, size_t len);
 void blake256_4way_close(void *cc, void *dst);
-void blake256_4way_addbits_and_close(
-        void *cc, unsigned ub, unsigned n, void *dst);

-#endif
+// 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_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_close(void *cc, void *dst);

 #ifdef __AVX2__

+// Blake-256 8 way
+
 typedef struct {
-        __m256i buf[16] __attribute__ ((aligned (64)));
-        __m256i H[8];
-        __m256i S[4];   
-        size_t ptr;
-	sph_u64 T0, T1;
+   __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
+} blake_8way_small_context;
+
+// 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_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_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_close(void *cc, void *dst);
+
+// Blake-512 4 way
+
+typedef struct {
+   __m256i buf[16] __attribute__ ((aligned (64)));
+   __m256i H[8];
+   __m256i S[4];   
+   size_t ptr;
+   sph_u64 T0, T1;
 } blake_4way_big_context;

 typedef blake_4way_big_context blake512_4way_context;
@@ -103,3 +139,5 @@ void blake512_4way_addbits_and_close(
 #endif

 #endif
+
+#endif
--- a/algo/blake/blake2s-4way.c
+++ b/algo/blake/blake2s-4way.c
@@ -0,0 +1,136 @@
+#include "blake2s-gate.h"
+#include "blake2s-hash-4way.h"
+#include <string.h>
+#include <stdint.h>
+
+#if 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 );
+
+   mm256_deinterleave_8x32( output,     output+ 32, output+ 64, output+ 96,
+                            output+128, output+160, output+192, output+224,
+                            vhash, 256 );
+}
+
+int scanhash_blake2s_8way( int thr_id, struct work *work, uint32_t max_nonce,
+                      uint64_t *hashes_done )
+{
+   uint32_t vdata[20*8] __attribute__ ((aligned (64)));
+   uint32_t hash[8*8] __attribute__ ((aligned (32)));
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   uint32_t _ALIGN(64) edata[20];
+   const uint32_t Htarg = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   uint32_t n = first_nonce;
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   uint32_t *noncep = vdata + 152;   // 19*8
+
+   swab32_array( edata, pdata, 20 );
+   mm256_interleave_8x32( vdata, edata, edata, edata, edata,
+                                 edata, edata, edata, edata, 640 );
+   blake2s_8way_init( &blake2s_8w_ctx, BLAKE2S_OUTBYTES );
+   blake2s_8way_update( &blake2s_8w_ctx, vdata, 64 );
+
+   do {
+      be32enc( noncep,    n   );
+      be32enc( noncep +1, n+1 );
+      be32enc( noncep +2, n+2 );
+      be32enc( noncep +3, n+3 );
+      be32enc( noncep +4, n+4 );
+      be32enc( noncep +5, n+5 );
+      be32enc( noncep +6, n+6 );
+      be32enc( noncep +7, n+7 );
+      pdata[19] = n;
+
+      blake2s_8way_hash( hash, vdata );
+
+
+      for ( int i = 0; i < 8; i++ )
+      if (  (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
+      {
+          pdata[19] = n+i;
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+      }
+      n += 8;
+
+   } while ( (num_found == 0) && (n < max_nonce)
+             && !work_restart[thr_id].restart );
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#elif defined(BLAKE2S_4WAY)
+
+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 );
+
+   mm_deinterleave_4x32( output, output+32, output+64, output+96, vhash, 256 );
+}
+
+int scanhash_blake2s_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                      uint64_t *hashes_done )
+{
+   uint32_t vdata[20*4] __attribute__ ((aligned (64)));
+   uint32_t hash[8*4] __attribute__ ((aligned (32)));
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   uint32_t _ALIGN(64) edata[20];
+   const uint32_t Htarg = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   uint32_t n = first_nonce;
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   uint32_t *noncep = vdata + 76;   // 19*4
+
+   swab32_array( edata, pdata, 20 );
+   mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
+   blake2s_4way_init( &blake2s_4w_ctx, BLAKE2S_OUTBYTES );
+   blake2s_4way_update( &blake2s_4w_ctx, vdata, 64 );
+
+   do {
+      be32enc( noncep,    n   );
+      be32enc( noncep +1, n+1 );
+      be32enc( noncep +2, n+2 );
+      be32enc( noncep +3, n+3 );
+      pdata[19] = n;
+
+      blake2s_4way_hash( hash, vdata );
+
+      for ( int i = 0; i < 4; i++ )
+      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
+      {
+          pdata[19] = n+i;
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+      }
+      n += 4;
+
+   } while ( (num_found == 0) && (n < max_nonce)
+             && !work_restart[thr_id].restart );
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#endif
--- a/algo/blake/blake2s-gate.c
+++ b/algo/blake/blake2s-gate.c
@@ -0,0 +1,27 @@
+#include "blake2s-gate.h"
+
+
+// changed to get_max64_0x3fffffLL in cpuminer-multi-decred
+int64_t blake2s_get_max64 ()
+{
+   return 0x7ffffLL;
+}
+
+bool register_blake2s_algo( algo_gate_t* gate )
+{
+#if defined(BLAKE2S_8WAY)
+  gate->scanhash  = (void*)&scanhash_blake2s_8way;
+  gate->hash      = (void*)&blake2s_8way_hash;
+#elif defined(BLAKE2S_4WAY)
+  gate->scanhash  = (void*)&scanhash_blake2s_4way;
+  gate->hash      = (void*)&blake2s_4way_hash;
+#else
+  gate->scanhash  = (void*)&scanhash_blake2s;
+  gate->hash      = (void*)&blake2s_hash;
+#endif
+  gate->get_max64 = (void*)&blake2s_get_max64;
+  gate->optimizations = AVX_OPT | AVX2_OPT;
+  return true;
+};
+
+
--- a/algo/blake/blake2s-gate.h
+++ b/algo/blake/blake2s-gate.h
@@ -0,0 +1,35 @@
+#ifndef __BLAKE2S_GATE_H__
+#define __BLAKE2S_GATE_H__ 1
+
+#include <stdint.h>
+#include "algo-gate-api.h"
+
+#if defined(__AVX__)
+  #define BLAKE2S_4WAY
+#endif
+#if defined(__AVX2__)
+  #define BLAKE2S_8WAY
+#endif
+
+bool register_blake2s_algo( algo_gate_t* gate );
+
+#if defined(BLAKE2S_8WAY)
+
+void blake2s_8way_hash( void *state, const void *input );
+int scanhash_blake2s_8way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+
+#elif defined (BLAKE2S_4WAY)
+
+void blake2s_4way_hash( void *state, const void *input );
+int scanhash_blake2s_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+#else
+
+void blake2s_hash( void *state, const void *input );
+int scanhash_blake2s( int thr_id, struct work *work, uint32_t max_nonce,
+                      uint64_t *hashes_done );
+
+#endif
+
+#endif
--- a/algo/blake/blake2s-hash-4way.c
+++ b/algo/blake/blake2s-hash-4way.c
@@ -0,0 +1,362 @@
+/**
+ * BLAKE2 reference source code package - reference C implementations
+ *
+ * Written in 2012 by Samuel Neves <sneves@dei.uc.pt>
+ *
+ * To the extent possible under law, the author(s) have dedicated all copyright
+ * and related and neighboring rights to this software to the public domain
+ * worldwide. This software is distributed without any warranty.
+ *
+ * You should have received a copy of the CC0 Public Domain Dedication along with
+ * this software. If not, see <http://creativecommons.org/publicdomain/zero/1.0/>.
+ */
+
+#include "blake2s-hash-4way.h"
+
+#include <stdint.h>
+#include <string.h>
+#include <stdio.h>
+
+#if defined(__AVX__)
+
+static const uint32_t blake2s_IV[8] =
+{
+	0x6A09E667UL, 0xBB67AE85UL, 0x3C6EF372UL, 0xA54FF53AUL,
+	0x510E527FUL, 0x9B05688CUL, 0x1F83D9ABUL, 0x5BE0CD19UL
+};
+
+static const uint8_t blake2s_sigma[10][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 } ,
+};
+
+// define a constant for initial param.
+
+int blake2s_4way_init( blake2s_4way_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_4way_state ) );
+   for( int i = 0; i < 8; ++i )
+      S->h[i] = _mm_set1_epi32( blake2s_IV[i] );
+
+   uint32_t *p = ( uint32_t * )( P );
+
+   /* IV XOR ParamBlock */
+   for ( size_t i = 0; i < 8; ++i )
+      S->h[i] = _mm_xor_si128( S->h[i], _mm_set1_epi32( p[i] ) );
+   return 0;
+}
+
+int blake2s_4way_compress( blake2s_4way_state *S, const __m128i* block )
+{
+   __m128i m[16];
+   __m128i v[16];
+
+   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[12] = _mm_xor_si128( _mm_set1_epi32( S->t[0] ),
+                          _mm_set1_epi32( blake2s_IV[4] ) );
+   v[13] = _mm_xor_si128( _mm_set1_epi32( S->t[1] ),
+                          _mm_set1_epi32( blake2s_IV[5] ) );
+   v[14] = _mm_xor_si128( _mm_set1_epi32( S->f[0] ),
+                          _mm_set1_epi32( blake2s_IV[6] ) );
+   v[15] = _mm_xor_si128( _mm_set1_epi32( S->f[1] ),
+                          _mm_set1_epi32( blake2s_IV[7] ) );
+
+#define G4W(r,i,a,b,c,d) \
+do { \
+   a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ blake2s_sigma[r][2*i+0] ] ); \
+   d = mm_rotr_32( _mm_xor_si128( d, a ), 16 ); \
+   c = _mm_add_epi32( c, d ); \
+   b = mm_rotr_32( _mm_xor_si128( b, c ), 12 ); \
+   a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ blake2s_sigma[r][2*i+1] ] ); \
+   d = mm_rotr_32( _mm_xor_si128( d, a ),  8 ); \
+   c = _mm_add_epi32( c, d ); \
+   b = mm_rotr_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] ); \
+} while(0)
+
+   ROUND4W( 0 );
+   ROUND4W( 1 );
+   ROUND4W( 2 );
+   ROUND4W( 3 );
+   ROUND4W( 4 );
+   ROUND4W( 5 );
+   ROUND4W( 6 );
+   ROUND4W( 7 );
+   ROUND4W( 8 );
+   ROUND4W( 9 );
+
+   for( size_t i = 0; i < 8; ++i )
+      S->h[i] = _mm_xor_si128( _mm_xor_si128( S->h[i], v[i] ), v[i + 8] );
+
+#undef G4W
+#undef ROUND4W
+   return 0;
+}
+
+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;
+
+   while( inlen > 0 )
+   {
+      size_t left = S->buflen;
+      if( inlen >= bsize - left )
+      {
+         memcpy_128( 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_4way_compress( S, buf ); 
+         S->buflen = 0;
+         input += ( bsize >> 2 );
+         inlen -= bsize;
+      }
+      else
+      {
+          memcpy_128( buf + ( left>>2 ), input, inlen>>2 );
+          S->buflen += (size_t) inlen; 
+          input += ( inlen>>2 );
+          inlen -= inlen;
+      }
+   }
+   return 0;
+}
+
+int blake2s_4way_final( blake2s_4way_state *S, void *out, uint8_t outlen )
+{
+   __m128i *buf = (__m128i*)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_128( buf + ( S->buflen>>2 ),
+                    ( BLAKE2S_BLOCKBYTES - S->buflen ) >> 2 );      
+   blake2s_4way_compress( S, buf );
+
+   for ( int i = 0; i < 8; ++i )
+      casti_m128i( out, i ) = S->h[ i ];
+   return 0;
+}
+
+#if defined(__AVX2__)
+
+int blake2s_8way_compress( blake2s_8way_state *S, const __m256i *block )
+{
+   __m256i m[16];
+   __m256i v[16];
+
+   memcpy_256( m, block, 16 );
+   memcpy_256( v, S->h, 8 );
+
+   v[ 8] = _mm256_set1_epi32( blake2s_IV[0] );
+   v[ 9] = _mm256_set1_epi32( blake2s_IV[1] );
+   v[10] = _mm256_set1_epi32( blake2s_IV[2] );
+   v[11] = _mm256_set1_epi32( blake2s_IV[3] );
+   v[12] = _mm256_xor_si256( _mm256_set1_epi32( S->t[0] ),
+                             _mm256_set1_epi32( blake2s_IV[4] ) );
+   v[13] = _mm256_xor_si256( _mm256_set1_epi32( S->t[1] ),
+                             _mm256_set1_epi32( blake2s_IV[5] ) );
+   v[14] = _mm256_xor_si256( _mm256_set1_epi32( S->f[0] ),
+                             _mm256_set1_epi32( blake2s_IV[6] ) );
+   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 ), \
+                          m[ blake2s_sigma[r][2*i+0] ] ); \
+   d = mm256_rotr_32( _mm256_xor_si256( d, a ), 16 ); \
+   c = _mm256_add_epi32( c, d ); \
+   b = mm256_rotr_32( _mm256_xor_si256( b, c ), 12 ); \
+   a = _mm256_add_epi32( _mm256_add_epi32( a, b ), \
+                         m[ blake2s_sigma[r][2*i+1] ] ); \
+   d = mm256_rotr_32( _mm256_xor_si256( d, a ),  8 ); \
+   c = _mm256_add_epi32( c, d ); \
+   b = mm256_rotr_32( _mm256_xor_si256( b, c ),  7 ); \
+} while(0)
+
+#define ROUND8W(r)  \
+do { \
+   G8W( r, 0, v[ 0], v[ 4], v[ 8], v[12] ); \
+   G8W( r, 1, v[ 1], v[ 5], v[ 9], v[13] ); \
+   G8W( r, 2, v[ 2], v[ 6], v[10], v[14] ); \
+   G8W( r, 3, v[ 3], v[ 7], v[11], v[15] ); \
+   G8W( r, 4, v[ 0], v[ 5], v[10], v[15] ); \
+   G8W( r, 5, v[ 1], v[ 6], v[11], v[12] ); \
+   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 );
+   ROUND8W( 2 );
+   ROUND8W( 3 );
+   ROUND8W( 4 );
+   ROUND8W( 5 );
+   ROUND8W( 6 );
+   ROUND8W( 7 );
+   ROUND8W( 8 );
+   ROUND8W( 9 );
+
+   for( size_t i = 0; i < 8; ++i )
+      S->h[i] = _mm256_xor_si256( _mm256_xor_si256( S->h[i], v[i] ), v[i + 8] );
+
+#undef G8W
+#undef ROUND8W
+   return 0;
+}
+
+int blake2s_8way_init( blake2s_8way_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_8way_state ) );
+   for( int i = 0; i < 8; ++i )
+      S->h[i] = _mm256_set1_epi32( blake2s_IV[i] );
+
+   uint32_t *p = ( uint32_t * )( P );
+
+   /* IV XOR ParamBlock */
+   for ( size_t i = 0; i < 8; ++i )
+      S->h[i] = _mm256_xor_si256( S->h[i], _mm256_set1_epi32( p[i] ) );
+   return 0;
+}
+
+int blake2s_8way_update( blake2s_8way_state *S, const void *in,
+                         uint64_t inlen )
+{
+  __m256i *input = (__m256i*)in;
+  __m256i *buf = (__m256i*)S->buf;
+  const int bsize = BLAKE2S_BLOCKBYTES;
+
+   while( inlen > 0 )
+   {
+      size_t left = S->buflen;
+      if( inlen >= bsize - left )
+      {
+         memcpy_256( 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_8way_compress( S, buf );
+         S->buflen = 0;
+         input += ( bsize >> 2 );
+         inlen -= bsize;
+      }
+      else
+      {
+          memcpy_256( buf + ( left>>2 ), input, inlen>>2 );
+          S->buflen += (size_t) inlen;
+          input += ( inlen>>2 );
+          inlen -= inlen;
+      }
+   }
+   return 0;
+}
+
+int blake2s_8way_final( blake2s_8way_state *S, void *out, uint8_t outlen )
+{
+   __m256i *buf = (__m256i*)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_256( buf + ( S->buflen>>2 ),
+                    ( BLAKE2S_BLOCKBYTES - S->buflen ) >> 2 );
+   blake2s_8way_compress( S, buf );
+
+   for ( int i = 0; i < 8; ++i )
+      casti_m256i( out, i ) = S->h[ i ];
+   return 0;
+}
+
+
+#endif // __AVX2__
+
+#if 0
+int blake2s( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen )
+{
+	blake2s_state S[1];
+
+	/* Verify parameters */
+	if ( NULL == in ) return -1;
+
+	if ( NULL == out ) return -1;
+
+	if ( NULL == key ) keylen = 0; /* Fail here instead if keylen != 0 and key == NULL? */
+
+	if( keylen > 0 )
+	{
+		if( blake2s_init_key( S, outlen, key, keylen ) < 0 ) return -1;
+	}
+	else
+	{
+		if( blake2s_init( S, outlen ) < 0 ) return -1;
+	}
+
+	blake2s_update( S, ( uint8_t * )in, inlen );
+	blake2s_final( S, out, outlen );
+	return 0;
+}
+#endif
+
+#endif // __AVX__
--- a/algo/blake/blake2s-hash-4way.h
+++ b/algo/blake/blake2s-hash-4way.h
@@ -0,0 +1,112 @@
+/**
+ * BLAKE2 reference source code package - reference C implementations
+ *
+ * Written in 2012 by Samuel Neves <sneves@dei.uc.pt>
+ *
+ * To the extent possible under law, the author(s) have dedicated all copyright
+ * and related and neighboring rights to this software to the public domain
+ * worldwide. This software is distributed without any warranty.
+ *
+ * You should have received a copy of the CC0 Public Domain Dedication along with
+ * this software. If not, see <http://creativecommons.org/publicdomain/zero/1.0/>.
+ */
+//#pragma once
+#ifndef __BLAKE2S_HASH_4WAY_H__
+#define __BLAKE2S_HASH_4WAY_H__ 1
+
+#if defined(__AVX__)
+
+#include "avxdefs.h"
+
+#include <stddef.h>
+#include <stdint.h>
+
+#if defined(_MSC_VER)
+#include <inttypes.h>
+#define inline __inline
+#define ALIGN(x) __declspec(align(x))
+#else
+#define ALIGN(x) __attribute__((aligned(x)))
+#endif
+
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+enum blake2s_constant
+{
+   BLAKE2S_BLOCKBYTES = 64,
+   BLAKE2S_OUTBYTES   = 32,
+   BLAKE2S_KEYBYTES   = 32,
+   BLAKE2S_SALTBYTES  = 8,
+   BLAKE2S_PERSONALBYTES = 8
+};
+
+#pragma pack(push, 1)
+typedef struct __blake2s_nway_param
+{
+   uint8_t  digest_length; // 1
+   uint8_t  key_length;    // 2
+   uint8_t  fanout;        // 3
+   uint8_t  depth;         // 4
+   uint32_t leaf_length;   // 8
+   uint8_t  node_offset[6];// 14
+   uint8_t  node_depth;    // 15
+   uint8_t  inner_length;  // 16
+   // uint8_t  reserved[0];
+   uint8_t  salt[BLAKE2S_SALTBYTES]; // 24
+   uint8_t  personal[BLAKE2S_PERSONALBYTES];  // 32
+} blake2s_nway_param;
+#pragma pack(pop)
+
+ALIGN( 64 ) typedef struct __blake2s_4way_state
+{
+   __m128i h[8];
+   uint8_t  buf[ BLAKE2S_BLOCKBYTES * 4 ];
+   uint32_t t[2];
+   uint32_t f[2];
+   size_t   buflen;
+   uint8_t  last_node;
+} blake2s_4way_state ;
+
+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 );
+
+#if defined(__AVX2__)
+
+ALIGN( 64 ) typedef struct __blake2s_8way_state
+{
+   __m256i h[8];
+   uint8_t  buf[ BLAKE2S_BLOCKBYTES * 8 ];
+   uint32_t t[2];
+   uint32_t f[2];
+   size_t   buflen;
+   uint8_t  last_node;
+} blake2s_8way_state ;
+
+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 );
+
+#endif
+
+#if 0
+	// Simple API
+//	int blake2s( uint8_t *out, const void *in, const void *key, const uint8_t outlen, const uint64_t inlen, uint8_t keylen );
+
+	// Direct Hash Mining Helpers
+	#define blake2s_salt32(out, in, inlen, key32) blake2s(out, in, key32, 32, inlen, 32) /* neoscrypt */
+	#define blake2s_simple(out, in, inlen) blake2s(out, in, NULL, 32, inlen, 0)
+#endif
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif  // __AVX__
+
+#endif
--- a/algo/blake/blake2s.c
+++ b/algo/blake/blake2s.c
@@ -1,26 +1,29 @@
-#include "algo-gate-api.h"
+#include "blake2s-gate.h"

 #include <string.h>
 #include <stdint.h>

-#include "crypto/blake2s.h"
+#include "sph-blake2s.h"

-static __thread blake2s_state s_midstate;
-static __thread blake2s_state s_ctx;
+static __thread blake2s_state blake2s_ctx;
+//static __thread blake2s_state s_ctx;
 #define MIDLEN 76

-void blake2s_hash(void *output, const void *input)
+void blake2s_hash( void *output, const void *input )
 {
-	unsigned char _ALIGN(64) hash[BLAKE2S_OUTBYTES];
-	blake2s_state blake2_ctx __attribute__ ((aligned (64)));
-
-	blake2s_init(&blake2_ctx, BLAKE2S_OUTBYTES);
-	blake2s_update(&blake2_ctx, input, 80);
-	blake2s_final(&blake2_ctx, hash, BLAKE2S_OUTBYTES);
+   unsigned char _ALIGN(64) hash[BLAKE2S_OUTBYTES];
+   blake2s_state ctx __attribute__ ((aligned (64)));
+  
+   memcpy( &ctx, &blake2s_ctx, sizeof ctx );
+   blake2s_update( &ctx, input+64, 16 );
+ 
+//	blake2s_init(&ctx, BLAKE2S_OUTBYTES);
+//	blake2s_update(&ctx, input, 80);
+	blake2s_final( &ctx, hash, BLAKE2S_OUTBYTES );

 	memcpy(output, hash, 32);
 }
-
+/*
 static void blake2s_hash_end(uint32_t *output, const uint32_t *input)
 {
 	s_ctx.buflen = MIDLEN;
@@ -28,7 +31,7 @@ static void blake2s_hash_end(uint32_t *output, const uint32_t *input)
 	blake2s_update(&s_ctx, (uint8_t*) &input[MIDLEN/4], 80 - MIDLEN);
 	blake2s_final(&s_ctx, (uint8_t*) output, BLAKE2S_OUTBYTES);
 }
-
+*/
 int scanhash_blake2s(int thr_id, struct work *work,
 	uint32_t max_nonce, uint64_t *hashes_done)
 {
@@ -46,13 +49,12 @@ int scanhash_blake2s(int thr_id, struct work *work,
        swab32_array( endiandata, pdata, 20 );

 	// midstate
-	blake2s_init(&s_midstate, BLAKE2S_OUTBYTES);
-	blake2s_update(&s_midstate, (uint8_t*) endiandata, MIDLEN);
-	memcpy(&s_ctx, &s_midstate, sizeof(blake2s_state));
+	blake2s_init( &blake2s_ctx, BLAKE2S_OUTBYTES );
+	blake2s_update( &blake2s_ctx, (uint8_t*) endiandata, 64 );

 	do {
 		be32enc(&endiandata[19], n);
-		blake2s_hash_end(hash64, endiandata);
+		blake2s_hash( hash64, endiandata );
 		if (hash64[7] < Htarg && fulltest(hash64, ptarget)) {
 			*hashes_done = n - first_nonce + 1;
 			pdata[19] = n;
@@ -67,7 +69,7 @@ int scanhash_blake2s(int thr_id, struct work *work,

 	return 0;
 }
-
+/*
 // changed to get_max64_0x3fffffLL in cpuminer-multi-decred
 int64_t blake2s_get_max64 ()
 {
@@ -81,4 +83,4 @@ bool register_blake2s_algo( algo_gate_t* gate )
  gate->get_max64 = (void*)&blake2s_get_max64;
  return true;
 };
-
+*/
--- a/algo/blake/blakecoin-4way.c
+++ b/algo/blake/blakecoin-4way.c
@@ -0,0 +1,141 @@
+#include "blakecoin-gate.h"
+#include "blake-hash-4way.h"
+#include <string.h>
+#include <stdint.h>
+#include <memory.h>
+
+#if defined (BLAKECOIN_4WAY)
+
+blake256r8_4way_context blakecoin_4w_ctx;
+
+void blakecoin_4way_hash(void *state, const void *input)
+{
+     uint32_t vhash[8*4] __attribute__ ((aligned (64)));
+     blake256r8_4way_context ctx;
+
+     memcpy( &ctx, &blakecoin_4w_ctx, sizeof ctx );
+     blake256r8_4way( &ctx, input + (64<<2), 16 );
+     blake256r8_4way_close( &ctx, vhash );
+
+     mm_deinterleave_4x32( state, state+32, state+64, state+96, vhash, 256 );
+}
+
+int scanhash_blakecoin_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done )
+{
+   uint32_t vdata[20*4] __attribute__ ((aligned (64)));
+   uint32_t hash[8*4] __attribute__ ((aligned (32)));
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   uint32_t HTarget = ptarget[7];
+   uint32_t _ALIGN(32) edata[20];
+   uint32_t n = first_nonce;
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   if ( opt_benchmark )
+      HTarget = 0x7f;
+
+   swab32_array( edata, pdata, 20 );
+   mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
+   blake256r8_4way_init( &blakecoin_4w_ctx );
+   blake256r8_4way( &blakecoin_4w_ctx, vdata, 64 );
+
+   uint32_t *noncep = vdata + 76;   // 19*4
+   do {
+      be32enc( noncep,    n   );
+      be32enc( noncep +1, n+1 );
+      be32enc( noncep +2, n+2 );
+      be32enc( noncep +3, n+3 );
+      pdata[19] = n;
+      blakecoin_4way_hash( hash, vdata );
+
+      for ( int i = 0; i < 4; i++ )
+      if (  (hash+(i<<3))[7] <= HTarget && fulltest( hash+(i<<3), ptarget ) )
+      {
+          pdata[19] = n+i;
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+      }
+      n += 4;
+
+   } while ( (num_found == 0) && (n < max_nonce) 
+             && !work_restart[thr_id].restart );
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#endif
+
+#if defined(BLAKECOIN_8WAY)
+
+blake256r8_8way_context blakecoin_8w_ctx;
+
+void blakecoin_8way_hash( void *state, const void *input )
+{
+     uint32_t vhash[8*8] __attribute__ ((aligned (64)));
+     blake256r8_8way_context ctx;
+
+     memcpy( &ctx, &blakecoin_8w_ctx, sizeof ctx );
+     blake256r8_8way( &ctx, input + (64<<3), 16 );
+     blake256r8_8way_close( &ctx, vhash );
+
+     mm256_deinterleave_8x32( state,     state+ 32, state+ 64, state+ 96,
+                              state+128, state+160, state+192, state+224,
+                              vhash, 256 );
+}
+
+int scanhash_blakecoin_8way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done )
+{
+   uint32_t vdata[20*8] __attribute__ ((aligned (64)));
+   uint32_t hash[8*8] __attribute__ ((aligned (32)));
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   uint32_t HTarget = ptarget[7];
+   uint32_t _ALIGN(32) edata[20];
+   uint32_t n = first_nonce;
+   uint32_t *nonces = work->nonces;
+   uint32_t *noncep = vdata + 152;   // 19*8
+   int num_found = 0;
+   if ( opt_benchmark )
+      HTarget = 0x7f;
+
+   // we need big endian data...
+   swab32_array( edata, pdata, 20 );
+   mm256_interleave_8x32( vdata, edata, edata, edata, edata,
+                                 edata, edata, edata, edata, 640 );
+   blake256r8_8way_init( &blakecoin_8w_ctx );
+   blake256r8_8way( &blakecoin_8w_ctx, vdata, 64 );
+
+   do {
+      be32enc( noncep,    n   );
+      be32enc( noncep +1, n+1 );
+      be32enc( noncep +2, n+2 );
+      be32enc( noncep +3, n+3 );
+      be32enc( noncep +4, n+4 );
+      be32enc( noncep +5, n+5 );
+      be32enc( noncep +6, n+6 );
+      be32enc( noncep +7, n+7 );
+      pdata[19] = n;
+      blakecoin_8way_hash( hash, vdata );
+
+      for ( int i = 0; i < 8; i++ )
+      if (  (hash+(i<<3))[7] <= HTarget && fulltest( hash+(i<<3), ptarget ) )
+      {
+          pdata[19] = n+i;
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+      }
+      n += 8;
+   } while ( (num_found == 0) && (n < max_nonce)
+             && !work_restart[thr_id].restart );
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#endif
+
--- a/algo/blake/blakecoin-gate.c
+++ b/algo/blake/blakecoin-gate.c
@@ -0,0 +1,36 @@
+#include "blakecoin-gate.h"
+#include <memory.h>
+
+// changed to get_max64_0x3fffffLL in cpuminer-multi-decred
+int64_t blakecoin_get_max64 ()
+{
+  return 0x7ffffLL;
+//  return 0x3fffffLL;
+}
+
+// vanilla uses default gen merkle root, otherwise identical to blakecoin
+bool register_vanilla_algo( algo_gate_t* gate )
+{
+#if defined(BLAKECOIN_8WAY)
+  gate->scanhash  = (void*)&scanhash_blakecoin_8way;
+  gate->hash      = (void*)&blakecoin_8way_hash;
+
+#elif defined(BLAKECOIN_4WAY)
+  gate->scanhash  = (void*)&scanhash_blakecoin_4way;
+  gate->hash      = (void*)&blakecoin_4way_hash;
+#else
+  gate->scanhash = (void*)&scanhash_blakecoin;
+  gate->hash     = (void*)&blakecoinhash;
+#endif
+  gate->optimizations = AVX_OPT | AVX2_OPT;
+  gate->get_max64 = (void*)&blakecoin_get_max64;
+  return true;
+}
+
+bool register_blakecoin_algo( algo_gate_t* gate )
+{
+  register_vanilla_algo( gate );
+  gate->gen_merkle_root = (void*)&SHA256_gen_merkle_root;
+  return true;
+}
+
--- a/algo/blake/blakecoin-gate.h
+++ b/algo/blake/blakecoin-gate.h
@@ -0,0 +1,30 @@
+#ifndef __BLAKECOIN_GATE_H__
+#define __BLAKECOIN_GATE_H__ 1
+
+#include "algo-gate-api.h"
+#include <stdint.h>
+
+#if defined(__AVX__)
+  #define BLAKECOIN_4WAY
+#endif
+#if defined(__AVX2__)
+  #define BLAKECOIN_8WAY
+#endif
+
+#if defined (BLAKECOIN_8WAY)
+void blakecoin_8way_hash(void *state, const void *input);
+int scanhash_blakecoin_8way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+#endif
+
+#if defined (BLAKECOIN_4WAY)
+void blakecoin_4way_hash(void *state, const void *input);
+int scanhash_blakecoin_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+#endif
+
+void blakecoinhash( void *state, const void *input );
+int scanhash_blakecoin( int thr_id, struct work *work, uint32_t max_nonce,
+                      uint64_t *hashes_done );
+
+#endif
--- a/algo/blake/blakecoin.c
+++ b/algo/blake/blakecoin.c
@@ -1,4 +1,4 @@
-#include "algo-gate-api.h"
+#include "blakecoin-gate.h"
 #define BLAKE32_ROUNDS 8
 #include "sph_blake.h"

@@ -98,7 +98,7 @@ void blakecoin_gen_merkle_root ( char* merkle_root, struct stratum_ctx* sctx )
 SHA256( sctx->job.coinbase, (int)sctx->job.coinbase_size, merkle_root );
 }
 */
-
+/*
 // changed to get_max64_0x3fffffLL in cpuminer-multi-decred
 int64_t blakecoin_get_max64 ()
 {
@@ -121,4 +121,4 @@ bool register_blakecoin_algo( algo_gate_t* gate )
  gate->gen_merkle_root = (void*)&SHA256_gen_merkle_root;
  return true;
 }
-
+*/
--- a/algo/blake/decred-4way.c
+++ b/algo/blake/decred-4way.c
@@ -1,5 +1,4 @@
 #include "decred-gate.h"
-#include "sph_blake.h"
 #include "blake-hash-4way.h"
 #include <string.h>
 #include <stdint.h>
@@ -9,59 +8,22 @@
 #if defined (DECRED_4WAY)

 static __thread blake256_4way_context blake_mid;
-static __thread bool ctx_midstate_done = false;

 void decred_hash_4way( void *state, const void *input )
 {
     uint32_t vhash[8*4] __attribute__ ((aligned (64)));
-     uint32_t hash0[8] __attribute__ ((aligned (32)));
-     uint32_t hash1[8] __attribute__ ((aligned (32)));
-     uint32_t hash2[8] __attribute__ ((aligned (32)));
-     uint32_t hash3[8] __attribute__ ((aligned (32)));
-     blake256_4way_context ctx __attribute__ ((aligned (64)));
-
-     sph_blake256_context ctx2 __attribute__ ((aligned (64)));
-     uint32_t hash[16] __attribute__ ((aligned (64)));
-     uint32_t sin0[45], sin1[45], sin2[45], sin3[45];
-
-     mm_deinterleave_4x32x( sin0, sin1, sin2, sin3, input, 180*8 );
-
-     void *tail = input + ( DECRED_MIDSTATE_LEN << 2 );
+//     uint32_t hash0[8] __attribute__ ((aligned (32)));
+//     uint32_t hash1[8] __attribute__ ((aligned (32)));
+//     uint32_t hash2[8] __attribute__ ((aligned (32)));
+//     uint32_t hash3[8] __attribute__ ((aligned (32)));
+     const void *tail = input + ( DECRED_MIDSTATE_LEN << 2 );
     int tail_len = 180 - DECRED_MIDSTATE_LEN; 
+     blake256_4way_context ctx __attribute__ ((aligned (64)));

     memcpy( &ctx, &blake_mid, sizeof(blake_mid) );
     blake256_4way( &ctx, tail, tail_len );
     blake256_4way_close( &ctx, vhash );
-/*
-     sph_blake256_init( &ctx2 );
-     sph_blake256( &ctx2, sin0, 180 );
-     sph_blake256_close( &ctx2, hash );
-*/
-/*
-     blake256_4way_init( &ctx );
-     blake256_4way( &ctx, input, 180 );
-     blake256_4way_close( &ctx, vhash );
-*/
-     mm_deinterleave_4x32( hash0, hash1, hash2, hash3, vhash, 256 );
-/*
-        for ( int i = 0; i < 8; i++ )
-          if ( hash[i] != hash0[i] )
-            printf(" hash mismatch, i = %u\n",i);
-
-printf("hash:  %08lx %08lx %08lx %08lx\n", *hash, *(hash+1),
-                             *(hash+2), *(hash+3) );
-printf("hash0: %08lx %08lx %08lx %08lx\n", *hash0, *(hash0+1),
-                             *(hash0+2), *(hash0+3) );
-printf("\n");
-*/
-
-     memcpy( state,    hash0, 32 );
-     memcpy( state+32, hash1, 32 );
-     memcpy( state+64, hash2, 32 );
-     memcpy( state+96, hash3, 32 );
-
-//     memcpy( state, hash, 32 );
-
+     mm_deinterleave_4x32( state, state+32, state+64, state+96, vhash, 256 );
 }

 int scanhash_decred_4way( int thr_id, struct work *work, uint32_t max_nonce,
@@ -69,28 +31,26 @@ int scanhash_decred_4way( int thr_id, struct work *work, uint32_t max_nonce,
 {
   uint32_t vdata[48*4] __attribute__ ((aligned (64)));
   uint32_t hash[8*4] __attribute__ ((aligned (32)));
-        uint32_t _ALIGN(64) edata[48];
-        uint32_t *pdata = work->data;
-        uint32_t *ptarget = work->target;
-        const uint32_t first_nonce = pdata[DECRED_NONCE_INDEX];
-        uint32_t n = first_nonce;
-        const uint32_t HTarget = opt_benchmark ? 0x7f : ptarget[7];
+   uint32_t _ALIGN(64) edata[48];
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[DECRED_NONCE_INDEX];
+   uint32_t n = first_nonce;
+   const uint32_t HTarget = opt_benchmark ? 0x7f : ptarget[7];
   uint32_t *nonces = work->nonces;
-   bool *found = work->nfound;
   int num_found = 0;

-        ctx_midstate_done = false;
-        memcpy( edata, pdata, 180 );
+   // copy to buffer guaranteed to be aligned.
+   memcpy( edata, pdata, 180 );

   // use the old way until  new way updated for size.
-   mm_interleave_4x32( vdata, edata, edata, edata, edata, 180*8 );
+   mm_interleave_4x32x( vdata, edata, edata, edata, edata, 180*8 );

   blake256_4way_init( &blake_mid );
   blake256_4way( &blake_mid, vdata, DECRED_MIDSTATE_LEN );

   uint32_t *noncep = vdata + DECRED_NONCE_INDEX * 4;
   do {
-      found[0] = found[1] = found[2] = found[3] = false;
      * noncep    = n;
      *(noncep+1) = n+1;
      *(noncep+2) = n+2;
@@ -98,55 +58,14 @@ int scanhash_decred_4way( int thr_id, struct work *work, uint32_t max_nonce,

      decred_hash_4way( hash, vdata );

-      if ( hash[7] <= HTarget && fulltest( hash, ptarget ) )
+      for ( int i = 0; i < 4; i++ )
+      if (  (hash+(i<<3))[7] <= HTarget && fulltest( hash+(i<<3), ptarget ) )
      {
-          work_set_target_ratio( work, hash );
-          found[0] = true;
-          num_found++;
-          nonces[0] = n;
-          pdata[DECRED_NONCE_INDEX] = n;
+          pdata[DECRED_NONCE_INDEX] = n+i;
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
      }
-/*
-      if ( (hash+8)[7] <= HTarget && fulltest( hash+8, ptarget ) )
-      {
-printf("found 1\n");          
-
-printf("vhash: %08lx %08lx %08lx %08lx\n", hash[8], hash[9], hash[10],hash[11] );
-printf("vhash: %08lx %08lx %08lx %08lx\n", hash[12], hash[13], hash[14],hash[15] );
-printf("shash: %08lx %08lx %08lx %08lx\n", shash[0], shash[1], shash[2],shash[3] );
-printf("shash: %08lx %08lx %08lx %08lx\n\n", shash[4], shash[5], shash[6],shash[7] );
-
-          work_set_target_ratio( work, hash+8 );
-          found[1] = true;
-          num_found++;
-          nonces[1] = n+1;
-      }
-*/
-      if ( (hash+16)[7] <= HTarget && fulltest( hash+16, ptarget ) )
-      {
-          work_set_target_ratio( work, hash+16 );
-          found[2] = true;
-          num_found++;
-          nonces[2] = n+2;
-      }
-/*
-      if ( (hash+24)[7] <= HTarget && fulltest( hash+24, ptarget ) )
-      {
-printf("found 3\n");          
-
-printf("vhash: %08lx %08lx %08lx %08lx\n", hash[0], hash[1], hash[2],hash[3] );
-printf("vhash: %08lx %08lx %08lx %08lx\n", hash[4], hash[5], hash[6],hash[7] );
-printf("shash: %08lx %08lx %08lx %08lx\n", shash[0], shash[1], shash[2],shash[3] );
-printf("shash: %08lx %08lx %08lx %08lx\n\n", shash[4], shash[5], shash[6],shash[7] );
-
-          work_set_target_ratio( work, hash+24 );
-          found[3] = true;
-          num_found++;
-          nonces[3] = n+3;
-      }
-*/
-      n += 2;
-//      n += 4;
+      n += 4;
  } while ( (num_found == 0) && (n < max_nonce) 
            && !work_restart[thr_id].restart );

--- a/algo/blake/decred-gate.c
+++ b/algo/blake/decred-gate.c
@@ -145,15 +145,13 @@ bool register_decred_algo( algo_gate_t* gate )
 {
 #if defined(DECRED_4WAY)
  four_way_not_tested();
-  gate->optimizations = FOUR_WAY_OPT;
  gate->scanhash  = (void*)&scanhash_decred_4way;
  gate->hash      = (void*)&decred_hash_4way;
 #else
-  gate->optimizations = SSE2_OPT;
  gate->scanhash  = (void*)&scanhash_decred;
  gate->hash      = (void*)&decred_hash;
 #endif
-
+  gate->optimizations = AVX2_OPT;
  gate->get_nonceptr          = (void*)&decred_get_nonceptr;
  gate->get_max64             = (void*)&get_max64_0x3fffffLL;
  gate->display_extra_data    = (void*)&decred_decode_extradata;
--- a/algo/blake/decred-gate.h
+++ b/algo/blake/decred-gate.h
@@ -18,7 +18,7 @@
 //                         uint64_t *hashes_done );
 #endif

-#if defined(FOUR_WAY) && defined(__AVX__)
+#if defined(__AVX2__)
  #define DECRED_4WAY
 #endif

--- a/algo/blake/pentablake-4way.c
+++ b/algo/blake/pentablake-4way.c
@@ -1,4 +1,7 @@
 #include "pentablake-gate.h"
+
+#if defined (__AVX2__)
+
 #include <stdlib.h>
 #include <stdint.h>
 #include <string.h>
@@ -9,8 +12,6 @@

 //#define DEBUG_ALGO

-#ifdef PENTABLAKE_4WAY
-
 extern void pentablakehash_4way( void *output, const void *input )
 {
 	unsigned char _ALIGN(32) hash[128];
@@ -110,12 +111,8 @@ int scanhash_pentablake_4way( int thr_id, struct work *work,
    const uint32_t first_nonce = pdata[19];
    const uint32_t Htarg = ptarget[7];
    uint32_t *nonces = work->nonces;
-    bool *found = work->nfound;
    int num_found = 0;
-    uint32_t *noncep0 = vdata + 73;   // 9*8 + 1
-    uint32_t *noncep1 = vdata + 75;
-    uint32_t *noncep2 = vdata + 77;
-    uint32_t *noncep3 = vdata + 79;
+    uint32_t *noncep = vdata + 73;   // 9*8 + 1

 //    uint32_t _ALIGN(32) hash64[8];
 //    uint32_t _ALIGN(32) endiandata[32];
@@ -149,47 +146,19 @@ int scanhash_pentablake_4way( int thr_id, struct work *work,
        {
           uint32_t mask = masks[m];
           do {
-              found[0] = found[1] = found[2] = found[3] = false;
-              be32enc( noncep0, n   );
-              be32enc( noncep1, n+1 );
-              be32enc( noncep2, n+2 );
-              be32enc( noncep3, n+3 );
+              be32enc( noncep,   n   );
+              be32enc( noncep+2, n+1 );
+              be32enc( noncep+4, n+2 );
+              be32enc( noncep+6, n+3 );

              pentablakehash_4way( hash, vdata );

-              // return immediately on nonce found, only one submit
-              if ( ( !(hash[7] & mask) ) && fulltest( hash, ptarget ) )
+              for ( int i = 0; i < 4; i++ )
+              if ( !( (hash+(i<<3))[7] & mask )
+                  && fulltest( hash+(i<<3), ptarget ) )
              {
-                  found[0] = true;
-                  num_found++;
-                  nonces[0] = n;
-                  pdata[19] = n;
-                  *hashes_done = n - first_nonce + 1;
-                  return 1;
-              }
-              if ( (! ((hash+8)[7] & mask) ) && fulltest( hash+8, ptarget ) )
-              {
-                  found[1] = true;
-                  num_found++;
-                  nonces[1] = n;
-                  *hashes_done = n - first_nonce + 1;
-                  return 1;
-              }
-              if ( ( !((hash+16)[7] & mask) ) && fulltest( hash+16, ptarget ) )
-              {
-                  found[2] = true;
-                  num_found++;
-                  nonces[2] = n;
-                  *hashes_done = n - first_nonce + 1;
-                  return 1;
-              }
-              if ( ( !((hash+24)[7] & mask) ) && fulltest( hash+24, ptarget ) )
-              {
-                  found[3] = true;
-                  num_found++;
-                  nonces[3] = n;
-                  *hashes_done = n - first_nonce + 1;
-                  return 1;
+                 nonces[ num_found++ ] = n+i;
+                 work_set_target_ratio( work, hash+(i<<3) );
              }
              n += 4;

--- a/algo/blake/pentablake-gate.c
+++ b/algo/blake/pentablake-gate.c
@@ -9,7 +9,7 @@ bool register_pentablake_algo( algo_gate_t* gate )
    gate->scanhash  = (void*)&scanhash_pentablake;
    gate->hash      = (void*)&pentablakehash;
 #endif
-    gate->optimizations = FOUR_WAY_OPT;
+    gate->optimizations = AVX2_OPT;
    gate->get_max64 = (void*)&get_max64_0x3ffff;
    return true;
 };
--- a/algo/blake/pentablake-gate.h
+++ b/algo/blake/pentablake-gate.h
@@ -4,7 +4,7 @@
 #include "algo-gate-api.h"
 #include <stdint.h>

-#if defined(FOUR_WAY) && defined(__AVX__)
+#if defined(__AVX2__)
  #define PENTABLAKE_4WAY
 #endif

--- a/algo/blake/sph-blake2s.c
+++ b/algo/blake/sph-blake2s.c
@@ -16,7 +16,7 @@
 #include <stdio.h>

 #include "algo/sha/sph_types.h"
-#include "crypto/blake2s.h"
+#include "sph-blake2s.h"

 static const uint32_t blake2s_IV[8] =
 {
--- a/algo/blake/sph-blake2s.h
+++ b/algo/blake/sph-blake2s.h
--- a/algo/bmw/bmw-hash-4way.c
+++ b/algo/bmw/bmw-hash-4way.c
@@ -41,19 +41,13 @@
 extern "C"{
 #endif

-//#include "sph_bmw.h"
-
-//#if SPH_SMALL_FOOTPRINT && !defined SPH_SMALL_FOOTPRINT_BMW
-#define SPH_SMALL_FOOTPRINT_BMW   1
-//#endif
-
 #ifdef _MSC_VER
 #pragma warning (disable: 4146)
 #endif

-//#undef SPH_ROTL64
-//#define SPH_ROTL64(x,n)  (((x) << (n)) | ((x) >> (64 - (n))))
-//#define SPH_ROTL64(x,n)  mm256_rotl_64(x,n)
+#define LPAR   (
+
+// BMW256

 static const sph_u32 IV256[] = {
 	SPH_C32(0x40414243), SPH_C32(0x44454647),
@@ -66,8 +60,7 @@ static const sph_u32 IV256[] = {
 	SPH_C32(0x78797A7B), SPH_C32(0x7C7D7E7F)
 };

-#if SPH_64
-
+// BMW512
 static const sph_u64 IV512[] = {
 	SPH_C64(0x8081828384858687), SPH_C64(0x88898A8B8C8D8E8F),
 	SPH_C64(0x9091929394959697), SPH_C64(0x98999A9B9C9D9E9F),
@@ -79,74 +72,113 @@ static const sph_u64 IV512[] = {
 	SPH_C64(0xF0F1F2F3F4F5F6F7), SPH_C64(0xF8F9FAFBFCFDFEFF)
 };

-#endif
+// BMW256

-#define XCAT(x, y)    XCAT_(x, y)
-#define XCAT_(x, y)   x ## y
+#define ss0(x) \
+   _mm_xor_si128( _mm_xor_si128( _mm_srli_epi32( (x), 1), \
+                                 _mm_slli_epi32( (x), 3) ), \
+                  _mm_xor_si128( mm_rotl_32( (x),  4), \
+                                 mm_rotl_32( (x), 19) ) )

-#define LPAR   (
+#define ss1(x) \
+   _mm_xor_si128( _mm_xor_si128( _mm_srli_epi32( (x), 1), \
+                                 _mm_slli_epi32( (x), 2) ), \
+                  _mm_xor_si128( mm_rotl_32( (x),  8), \
+                                 mm_rotl_32( (x), 23) ) )

-/*
-#define ss0(x)    (((x) >> 1) ^ SPH_T32((x) << 3) \
-                  ^ SPH_ROTL32(x,  4) ^ SPH_ROTL32(x, 19))
-#define ss1(x)    (((x) >> 1) ^ SPH_T32((x) << 2) \
-                  ^ SPH_ROTL32(x,  8) ^ SPH_ROTL32(x, 23))
-#define ss2(x)    (((x) >> 2) ^ SPH_T32((x) << 1) \
-                  ^ SPH_ROTL32(x, 12) ^ SPH_ROTL32(x, 25))
-#define ss3(x)    (((x) >> 2) ^ SPH_T32((x) << 2) \
-                  ^ SPH_ROTL32(x, 15) ^ SPH_ROTL32(x, 29))
-#define ss4(x)    (((x) >> 1) ^ (x))
-#define ss5(x)    (((x) >> 2) ^ (x))
-#define rs1(x)    SPH_ROTL32(x,  3)
-#define rs2(x)    SPH_ROTL32(x,  7)
-#define rs3(x)    SPH_ROTL32(x, 13)
-#define rs4(x)    SPH_ROTL32(x, 16)
-#define rs5(x)    SPH_ROTL32(x, 19)
-#define rs6(x)    SPH_ROTL32(x, 23)
-#define rs7(x)    SPH_ROTL32(x, 27)
+#define ss2(x) \
+   _mm_xor_si128( _mm_xor_si128( _mm_srli_epi32( (x), 2), \
+                                 _mm_slli_epi32( (x), 1) ), \
+                  _mm_xor_si128( mm_rotl_32( (x), 12), \
+                                 mm_rotl_32( (x), 25) ) )

-#define Ks(j)   SPH_T32((sph_u32)(j) * SPH_C32(0x05555555))
+#define ss3(x) \
+   _mm_xor_si128( _mm_xor_si128( _mm_srli_epi32( (x), 2), \
+                                 _mm_slli_epi32( (x), 2) ), \
+                  _mm_xor_si128( mm_rotl_32( (x), 15), \
+                                 mm_rotl_32( (x), 29) ) )

-#define add_elt_s(mf, hf, j0m, j1m, j3m, j4m, j7m, j10m, j11m, j16) \
-	(SPH_T32(SPH_ROTL32(mf(j0m), j1m) + SPH_ROTL32(mf(j3m), j4m) \
-		- SPH_ROTL32(mf(j10m), j11m) + Ks(j16)) ^ hf(j7m))
+#define ss4(x) \
+  _mm_xor_si128( (x), _mm_srli_epi32( (x), 1 ) )

-#define expand1s_inner(qf, mf, hf, i16, \
-		i0, i1, i2, i3, i4, i5, i6, i7, i8, \
-		i9, i10, i11, i12, i13, i14, i15, \
-		i0m, i1m, i3m, i4m, i7m, i10m, i11m) \
-	SPH_T32(ss1(qf(i0)) + ss2(qf(i1)) + ss3(qf(i2)) + ss0(qf(i3)) \
-		+ ss1(qf(i4)) + ss2(qf(i5)) + ss3(qf(i6)) + ss0(qf(i7)) \
-		+ ss1(qf(i8)) + ss2(qf(i9)) + ss3(qf(i10)) + ss0(qf(i11)) \
-		+ ss1(qf(i12)) + ss2(qf(i13)) + ss3(qf(i14)) + ss0(qf(i15)) \
-		+ add_elt_s(mf, hf, i0m, i1m, i3m, i4m, i7m, i10m, i11m, i16))
+#define ss5(x) \
+  _mm_xor_si128( (x), _mm_srli_epi32( (x), 2 ) )

-#define expand1s(qf, mf, hf, i16) \
-	expand1s_(qf, mf, hf, i16, I16_ ## i16, M16_ ## i16)
-#define expand1s_(qf, mf, hf, i16, ix, iy) \
-	expand1s_inner LPAR qf, mf, hf, i16, ix, iy)
+#define rs1(x)    mm_rotl_32( x,  3 ) 
+#define rs2(x)    mm_rotl_32( x,  7 ) 
+#define rs3(x)    mm_rotl_32( x, 13 ) 
+#define rs4(x)    mm_rotl_32( x, 16 ) 
+#define rs5(x)    mm_rotl_32( x, 19 ) 
+#define rs6(x)    mm_rotl_32( x, 23 ) 
+#define rs7(x)    mm_rotl_32( x, 27 ) 

-#define expand2s_inner(qf, mf, hf, i16, \
-		i0, i1, i2, i3, i4, i5, i6, i7, i8, \
-		i9, i10, i11, i12, i13, i14, i15, \
-		i0m, i1m, i3m, i4m, i7m, i10m, i11m) \
-	SPH_T32(qf(i0) + rs1(qf(i1)) + qf(i2) + rs2(qf(i3)) \
-		+ qf(i4) + rs3(qf(i5)) + qf(i6) + rs4(qf(i7)) \
-		+ qf(i8) + rs5(qf(i9)) + qf(i10) + rs6(qf(i11)) \
-		+ qf(i12) + rs7(qf(i13)) + ss4(qf(i14)) + ss5(qf(i15)) \
-		+ add_elt_s(mf, hf, i0m, i1m, i3m, i4m, i7m, i10m, i11m, i16))
+#define rol_off_32( M, j, off ) \
+   mm_rotl_32( M[ ( (j) + (off) ) & 0xF ] , \
+                ( ( (j) + (off) ) & 0xF ) + 1 )

-#define expand2s(qf, mf, hf, i16) \
-	expand2s_(qf, mf, hf, i16, I16_ ## i16, M16_ ## i16)
-#define expand2s_(qf, mf, hf, i16, ix, iy) \
-	expand2s_inner LPAR qf, mf, hf, i16, ix, iy)
-*/
-#if SPH_64
+#define add_elt_s( M, H, j ) \
+   _mm_xor_si128( \
+       _mm_add_epi32( \
+             _mm_sub_epi32( _mm_add_epi32( rol_off_32( M, j, 0 ), \
+                                           rol_off_32( M, j, 3 ) ), \
+                            rol_off_32( M, j, 10 ) ), \
+       _mm_set1_epi32( ( (j)+16 ) * SPH_C32(0x05555555UL) ) ), \
+   H[ ( (j)+7 ) & 0xF ] )
+
+
+#define expand1s( qt, M, H, i ) \
+   _mm_add_epi32( \
+      _mm_add_epi32( \
+         _mm_add_epi32( \
+             _mm_add_epi32( \
+                _mm_add_epi32( ss1( qt[ (i)-16 ] ), \
+                               ss2( qt[ (i)-15 ] ) ), \
+                _mm_add_epi32( ss3( qt[ (i)-14 ] ), \
+                               ss0( qt[ (i)-13 ] ) ) ), \
+             _mm_add_epi32( \
+                _mm_add_epi32( ss1( qt[ (i)-12 ] ), \
+                               ss2( qt[ (i)-11 ] ) ), \
+                _mm_add_epi32( ss3( qt[ (i)-10 ] ), \
+                               ss0( qt[ (i)- 9 ] ) ) ) ), \
+         _mm_add_epi32( \
+             _mm_add_epi32( \
+                _mm_add_epi32( ss1( qt[ (i)- 8 ] ), \
+                               ss2( qt[ (i)- 7 ] ) ), \
+                _mm_add_epi32( ss3( qt[ (i)- 6 ] ), \
+                               ss0( qt[ (i)- 5 ] ) ) ), \
+             _mm_add_epi32( \
+                _mm_add_epi32( ss1( qt[ (i)- 4 ] ), \
+                               ss2( qt[ (i)- 3 ] ) ), \
+                _mm_add_epi32( ss3( qt[ (i)- 2 ] ), \
+                               ss0( qt[ (i)- 1 ] ) ) ) ) ), \
+      add_elt_s( M, H, (i)-16 ) )
+
+#define expand2s( qt, M, H, i) \
+   _mm_add_epi32( \
+      _mm_add_epi32( \
+         _mm_add_epi32( \
+             _mm_add_epi32( \
+                _mm_add_epi32( qt[ (i)-16 ], rs1( qt[ (i)-15 ] ) ), \
+                _mm_add_epi32( qt[ (i)-14 ], rs2( qt[ (i)-13 ] ) ) ), \
+             _mm_add_epi32( \
+                _mm_add_epi32( qt[ (i)-12 ], rs3( qt[ (i)-11 ] ) ), \
+                _mm_add_epi32( qt[ (i)-10 ], rs4( qt[ (i)- 9 ] ) ) ) ), \
+         _mm_add_epi32( \
+             _mm_add_epi32( \
+                _mm_add_epi32( qt[ (i)- 8 ], rs5( qt[ (i)- 7 ] ) ), \
+                _mm_add_epi32( qt[ (i)- 6 ], rs6( qt[ (i)- 5 ] ) ) ), \
+             _mm_add_epi32( \
+                _mm_add_epi32( qt[ (i)- 4 ], rs7( qt[ (i)- 3 ] ) ), \
+                _mm_add_epi32( ss4( qt[ (i)- 2 ] ), \
+                               ss5( qt[ (i)- 1 ] ) ) ) ) ), \
+      add_elt_s( M, H, (i)-16 ) )
+
+// BMW512

 #define sb0(x) \
   _mm256_xor_si256( _mm256_xor_si256( _mm256_srli_epi64( (x), 1), \
                                       _mm256_slli_epi64( (x), 3) ), \
-                     _mm256_xor_si256( mm256_rotl_64( (x), 4), \
+                     _mm256_xor_si256( mm256_rotl_64( (x),  4), \
                                       mm256_rotl_64( (x), 37) ) )

 #define sb1(x) \
@@ -181,18 +213,18 @@ static const sph_u64 IV512[] = {
 #define rb6(x)    mm256_rotl_64( x, 43 ) 
 #define rb7(x)    mm256_rotl_64( x, 53 ) 

-#define rol_off( M, j, off ) \
-   mm256_rotl_64( M[ ( (j) + (off) ) & 15 ] , \
-                   ( ( (j) + (off) ) & 15 ) + 1 )
+#define rol_off_64( M, j, off ) \
+   mm256_rotl_64( M[ ( (j) + (off) ) & 0xF ] , \
+                   ( ( (j) + (off) ) & 0xF ) + 1 )

 #define add_elt_b( M, H, j ) \
   _mm256_xor_si256( \
      _mm256_add_epi64( \
-            _mm256_sub_epi64( _mm256_add_epi64( rol_off( M, j, 0 ), \
-                                                rol_off( M, j, 3 ) ), \
-                             rol_off( M, j, 10 ) ), \
+            _mm256_sub_epi64( _mm256_add_epi64( rol_off_64( M, j, 0 ), \
+                                                rol_off_64( M, j, 3 ) ), \
+                             rol_off_64( M, j, 10 ) ), \
            _mm256_set1_epi64x( ( (j) + 16 ) * 0x0555555555555555ULL ) ), \
-       H[ ( (j)+7 ) & 15 ] )
+       H[ ( (j)+7 ) & 0xF ] )
          
 #define expand1b( qt, M, H, i ) \
   _mm256_add_epi64( \
@@ -241,132 +273,301 @@ static const sph_u64 IV512[] = {
                                  sb5( qt[ (i)- 1 ] ) ) ) ) ), \
      add_elt_b( M, H, (i)-16 ) )

-#endif
+// BMW256

-/*
-#define MAKE_W( i0, op01, i1, op12, i2, op23, i3, op34, i4) \
-        ((M(i0) ^ H(i0)) op01 (M(i1) ^ H(i1)) op12 (M(i2) ^ H(i2)) \
-        op23 (M(i3) ^ H(i3)) op34 (M(i4) ^ H(i4)))
-*/
+#define Ws0 \
+   _mm_add_epi32( \
+       _mm_add_epi32( \
+          _mm_add_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 5], H[ 5] ), \
+                            _mm_xor_si128( M[ 7], H[ 7] ) ), \
+             _mm_xor_si128( M[10], H[10] ) ), \
+          _mm_xor_si128( M[13], H[13] ) ), \
+       _mm_xor_si128( M[14], H[14] ) )

-/*
-#define Ws0    MAKE_W(SPH_T32,  5, -,  7, +, 10, +, 13, +, 14)
-#define Ws1    MAKE_W(SPH_T32,  6, -,  8, +, 11, +, 14, -, 15)
-#define Ws2    MAKE_W(SPH_T32,  0, +,  7, +,  9, -, 12, +, 15)
-#define Ws3    MAKE_W(SPH_T32,  0, -,  1, +,  8, -, 10, +, 13)
-#define Ws4    MAKE_W(SPH_T32,  1, +,  2, +,  9, -, 11, -, 14)
-#define Ws5    MAKE_W(SPH_T32,  3, -,  2, +, 10, -, 12, +, 15)
-#define Ws6    MAKE_W(SPH_T32,  4, -,  0, -,  3, -, 11, +, 13)
-#define Ws7    MAKE_W(SPH_T32,  1, -,  4, -,  5, -, 12, -, 14)
-#define Ws8    MAKE_W(SPH_T32,  2, -,  5, -,  6, +, 13, -, 15)
-#define Ws9    MAKE_W(SPH_T32,  0, -,  3, +,  6, -,  7, +, 14)
-#define Ws10   MAKE_W(SPH_T32,  8, -,  1, -,  4, -,  7, +, 15)
-#define Ws11   MAKE_W(SPH_T32,  8, -,  0, -,  2, -,  5, +,  9)
-#define Ws12   MAKE_W(SPH_T32,  1, +,  3, -,  6, -,  9, +, 10)
-#define Ws13   MAKE_W(SPH_T32,  2, +,  4, +,  7, +, 10, +, 11)
-#define Ws14   MAKE_W(SPH_T32,  3, -,  5, +,  8, -, 11, -, 12)
-#define Ws15   MAKE_W(SPH_T32, 12, -,  4, -,  6, -,  9, +, 13)
+#define Ws1 \
+   _mm_sub_epi32( \
+       _mm_add_epi32( \
+          _mm_add_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 6], H[ 6] ), \
+                            _mm_xor_si128( M[ 8], H[ 8] ) ), \
+             _mm_xor_si128( M[11], H[11] ) ), \
+          _mm_xor_si128( M[14], H[14] ) ), \
+       _mm_xor_si128( M[15], H[15] ) )

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

-#define MAKE_Qas   do { \
-		unsigned u; \
-		sph_u32 Ws[16]; \
-		Ws[ 0] = Ws0; \
-		Ws[ 1] = Ws1; \
-		Ws[ 2] = Ws2; \
-		Ws[ 3] = Ws3; \
-		Ws[ 4] = Ws4; \
-		Ws[ 5] = Ws5; \
-		Ws[ 6] = Ws6; \
-		Ws[ 7] = Ws7; \
-		Ws[ 8] = Ws8; \
-		Ws[ 9] = Ws9; \
-		Ws[10] = Ws10; \
-		Ws[11] = Ws11; \
-		Ws[12] = Ws12; \
-		Ws[13] = Ws13; \
-		Ws[14] = Ws14; \
-		Ws[15] = Ws15; \
-		for (u = 0; u < 15; u += 5) { \
-			qt[u + 0] = SPH_T32(ss0(Ws[u + 0]) + H(u + 1)); \
-			qt[u + 1] = SPH_T32(ss1(Ws[u + 1]) + H(u + 2)); \
-			qt[u + 2] = SPH_T32(ss2(Ws[u + 2]) + H(u + 3)); \
-			qt[u + 3] = SPH_T32(ss3(Ws[u + 3]) + H(u + 4)); \
-			qt[u + 4] = SPH_T32(ss4(Ws[u + 4]) + H(u + 5)); \
-		} \
-		qt[15] = SPH_T32(ss0(Ws[15]) + H(0)); \
-	} while (0)
+#define Ws3 \
+   _mm_add_epi32( \
+       _mm_sub_epi32( \
+          _mm_add_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 0], H[ 0] ), \
+                            _mm_xor_si128( M[ 1], H[ 1] ) ), \
+             _mm_xor_si128( M[ 8], H[ 8] ) ), \
+          _mm_xor_si128( M[10], H[10] ) ), \
+       _mm_xor_si128( M[13], H[13] ) )

-#define MAKE_Qbs   do { \
-		qt[16] = expand1s(Qs, M, H, 16); \
-		qt[17] = expand1s(Qs, M, H, 17); \
-		qt[18] = expand2s(Qs, M, H, 18); \
-		qt[19] = expand2s(Qs, M, H, 19); \
-		qt[20] = expand2s(Qs, M, H, 20); \
-		qt[21] = expand2s(Qs, M, H, 21); \
-		qt[22] = expand2s(Qs, M, H, 22); \
-		qt[23] = expand2s(Qs, M, H, 23); \
-		qt[24] = expand2s(Qs, M, H, 24); \
-		qt[25] = expand2s(Qs, M, H, 25); \
-		qt[26] = expand2s(Qs, M, H, 26); \
-		qt[27] = expand2s(Qs, M, H, 27); \
-		qt[28] = expand2s(Qs, M, H, 28); \
-		qt[29] = expand2s(Qs, M, H, 29); \
-		qt[30] = expand2s(Qs, M, H, 30); \
-		qt[31] = expand2s(Qs, M, H, 31); \
-	} while (0)
+#define Ws4 \
+   _mm_sub_epi32( \
+       _mm_sub_epi32( \
+          _mm_add_epi32( \
+             _mm_add_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
+                            _mm_xor_si128( M[ 2], H[ 2] ) ), \
+             _mm_xor_si128( M[ 9], H[ 9] ) ), \
+          _mm_xor_si128( M[11], H[11] ) ), \
+       _mm_xor_si128( M[14], H[14] ) )

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

-#define MAKE_Qas   do { \
-		qt[ 0] = SPH_T32(ss0(Ws0 ) + H( 1)); \
-		qt[ 1] = SPH_T32(ss1(Ws1 ) + H( 2)); \
-		qt[ 2] = SPH_T32(ss2(Ws2 ) + H( 3)); \
-		qt[ 3] = SPH_T32(ss3(Ws3 ) + H( 4)); \
-		qt[ 4] = SPH_T32(ss4(Ws4 ) + H( 5)); \
-		qt[ 5] = SPH_T32(ss0(Ws5 ) + H( 6)); \
-		qt[ 6] = SPH_T32(ss1(Ws6 ) + H( 7)); \
-		qt[ 7] = SPH_T32(ss2(Ws7 ) + H( 8)); \
-		qt[ 8] = SPH_T32(ss3(Ws8 ) + H( 9)); \
-		qt[ 9] = SPH_T32(ss4(Ws9 ) + H(10)); \
-		qt[10] = SPH_T32(ss0(Ws10) + H(11)); \
-		qt[11] = SPH_T32(ss1(Ws11) + H(12)); \
-		qt[12] = SPH_T32(ss2(Ws12) + H(13)); \
-		qt[13] = SPH_T32(ss3(Ws13) + H(14)); \
-		qt[14] = SPH_T32(ss4(Ws14) + H(15)); \
-		qt[15] = SPH_T32(ss0(Ws15) + H( 0)); \
-	} while (0)
+#define Ws6 \
+   _mm_add_epi32( \
+       _mm_sub_epi32( \
+          _mm_sub_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 4], H[ 4] ), \
+                            _mm_xor_si128( M[ 0], H[ 0] ) ), \
+             _mm_xor_si128( M[ 3], H[ 3] ) ), \
+          _mm_xor_si128( M[11], H[11] ) ), \
+       _mm_xor_si128( M[13], H[13] ) )

-#define MAKE_Qbs   do { \
-		qt[16] = expand1s(Qs, M, H, 16); \
-		qt[17] = expand1s(Qs, M, H, 17); \
-		qt[18] = expand2s(Qs, M, H, 18); \
-		qt[19] = expand2s(Qs, M, H, 19); \
-		qt[20] = expand2s(Qs, M, H, 20); \
-		qt[21] = expand2s(Qs, M, H, 21); \
-		qt[22] = expand2s(Qs, M, H, 22); \
-		qt[23] = expand2s(Qs, M, H, 23); \
-		qt[24] = expand2s(Qs, M, H, 24); \
-		qt[25] = expand2s(Qs, M, H, 25); \
-		qt[26] = expand2s(Qs, M, H, 26); \
-		qt[27] = expand2s(Qs, M, H, 27); \
-		qt[28] = expand2s(Qs, M, H, 28); \
-		qt[29] = expand2s(Qs, M, H, 29); \
-		qt[30] = expand2s(Qs, M, H, 30); \
-		qt[31] = expand2s(Qs, M, H, 31); \
-	} while (0)
+#define Ws7 \
+   _mm_sub_epi32( \
+       _mm_sub_epi32( \
+          _mm_sub_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
+                            _mm_xor_si128( M[ 4], H[ 4] ) ), \
+             _mm_xor_si128( M[ 5], H[ 5] ) ), \
+          _mm_xor_si128( M[12], H[12] ) ), \
+       _mm_xor_si128( M[14], H[14] ) )

-#endif
+#define Ws8 \
+   _mm_sub_epi32( \
+       _mm_add_epi32( \
+          _mm_sub_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 2], H[ 2] ), \
+                            _mm_xor_si128( M[ 5], H[ 5] ) ), \
+             _mm_xor_si128( M[ 6], H[ 6] ) ), \
+          _mm_xor_si128( M[13], H[13] ) ), \
+       _mm_xor_si128( M[15], H[15] ) )

-#define MAKE_Qs   do { \
-		MAKE_Qas; \
-		MAKE_Qbs; \
-	} while (0)
+#define Ws9 \
+   _mm_add_epi32( \
+       _mm_sub_epi32( \
+          _mm_add_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 0], H[ 0] ), \
+                            _mm_xor_si128( M[ 3], H[ 3] ) ), \
+             _mm_xor_si128( M[ 6], H[ 6] ) ), \
+          _mm_xor_si128( M[ 7], H[ 7] ) ), \
+       _mm_xor_si128( M[14], H[14] ) )

-#define Qs(j)   (qt[j])
-*/
-#if SPH_64
+#define Ws10 \
+   _mm_add_epi32( \
+       _mm_sub_epi32( \
+          _mm_sub_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 8], H[ 8] ), \
+                            _mm_xor_si128( M[ 1], H[ 1] ) ), \
+             _mm_xor_si128( M[ 4], H[ 4] ) ), \
+          _mm_xor_si128( M[ 7], H[ 7] ) ), \
+       _mm_xor_si128( M[15], H[15] ) )
+
+#define Ws11 \
+   _mm_add_epi32( \
+       _mm_sub_epi32( \
+          _mm_sub_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 8], H[ 8] ), \
+                            _mm_xor_si128( M[ 0], H[ 0] ) ), \
+             _mm_xor_si128( M[ 2], H[ 2] ) ), \
+          _mm_xor_si128( M[ 5], H[ 5] ) ), \
+       _mm_xor_si128( M[ 9], H[ 9] ) )
+
+#define Ws12 \
+   _mm_add_epi32( \
+       _mm_sub_epi32( \
+          _mm_sub_epi32( \
+             _mm_add_epi32( _mm_xor_si128( M[ 1], H[ 1] ), \
+                            _mm_xor_si128( M[ 3], H[ 3] ) ), \
+             _mm_xor_si128( M[ 6], H[ 6] ) ), \
+          _mm_xor_si128( M[ 9], H[ 9] ) ), \
+       _mm_xor_si128( M[10], H[10] ) )
+
+#define Ws13 \
+   _mm_add_epi32( \
+       _mm_add_epi32( \
+          _mm_add_epi32( \
+             _mm_add_epi32( _mm_xor_si128( M[ 2], H[ 2] ), \
+                            _mm_xor_si128( M[ 4], H[ 4] ) ), \
+             _mm_xor_si128( M[ 7], H[ 7] ) ), \
+          _mm_xor_si128( M[10], H[10] ) ), \
+       _mm_xor_si128( M[11], H[11] ) )
+
+#define Ws14 \
+   _mm_sub_epi32( \
+       _mm_sub_epi32( \
+          _mm_add_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[ 3], H[ 3] ), \
+                               _mm_xor_si128( M[ 5], H[ 5] ) ), \
+             _mm_xor_si128( M[ 8], H[ 8] ) ), \
+          _mm_xor_si128( M[11], H[11] ) ), \
+       _mm_xor_si128( M[12], H[12] ) )
+
+#define Ws15 \
+   _mm_add_epi32( \
+       _mm_sub_epi32( \
+          _mm_sub_epi32( \
+             _mm_sub_epi32( _mm_xor_si128( M[12], H[12] ), \
+                            _mm_xor_si128( M[ 4], H[ 4] ) ), \
+             _mm_xor_si128( M[ 6], H[ 6] ) ), \
+          _mm_xor_si128( M[ 9], H[ 9] ) ), \
+       _mm_xor_si128( M[13], H[13] ) )
+
+
+void compress_small( const __m128i *M, const __m128i H[16], __m128i dH[16] )
+{
+   __m128i qt[32], xl, xh; \
+
+   qt[ 0] = _mm_add_epi32( ss0( Ws0 ), H[ 1] );
+   qt[ 1] = _mm_add_epi32( ss1( Ws1 ), H[ 2] );
+   qt[ 2] = _mm_add_epi32( ss2( Ws2 ), H[ 3] );
+   qt[ 3] = _mm_add_epi32( ss3( Ws3 ), H[ 4] );
+   qt[ 4] = _mm_add_epi32( ss4( Ws4 ), H[ 5] );
+   qt[ 5] = _mm_add_epi32( ss0( Ws5 ), H[ 6] );
+   qt[ 6] = _mm_add_epi32( ss1( Ws6 ), H[ 7] );
+   qt[ 7] = _mm_add_epi32( ss2( Ws7 ), H[ 8] );
+   qt[ 8] = _mm_add_epi32( ss3( Ws8 ), H[ 9] );
+   qt[ 9] = _mm_add_epi32( ss4( Ws9 ), H[10] );
+   qt[10] = _mm_add_epi32( ss0( Ws10), H[11] );
+   qt[11] = _mm_add_epi32( ss1( Ws11), H[12] );
+   qt[12] = _mm_add_epi32( ss2( Ws12), H[13] );
+   qt[13] = _mm_add_epi32( ss3( Ws13), H[14] );
+   qt[14] = _mm_add_epi32( ss4( Ws14), H[15] );
+   qt[15] = _mm_add_epi32( ss0( Ws15), H[ 0] );
+   qt[16] = expand1s( qt, M, H, 16 );
+   qt[17] = expand1s( qt, M, H, 17 );
+   qt[18] = expand2s( qt, M, H, 18 );
+   qt[19] = expand2s( qt, M, H, 19 );
+   qt[20] = expand2s( qt, M, H, 20 );
+   qt[21] = expand2s( qt, M, H, 21 );
+   qt[22] = expand2s( qt, M, H, 22 );
+   qt[23] = expand2s( qt, M, H, 23 );
+   qt[24] = expand2s( qt, M, H, 24 );
+   qt[25] = expand2s( qt, M, H, 25 );
+   qt[26] = expand2s( qt, M, H, 26 );
+   qt[27] = expand2s( qt, M, H, 27 );
+   qt[28] = expand2s( qt, M, H, 28 );
+   qt[29] = expand2s( qt, M, H, 29 );
+   qt[30] = expand2s( qt, M, H, 30 );
+   qt[31] = expand2s( qt, M, H, 31 );
+
+   xl = _mm_xor_si128(
+              _mm_xor_si128( _mm_xor_si128( qt[16], qt[17] ),
+                             _mm_xor_si128( qt[18], qt[19] ) ),
+              _mm_xor_si128( _mm_xor_si128( qt[20], qt[21] ),
+                             _mm_xor_si128( qt[22], qt[23] ) ) );
+   xh = _mm_xor_si128( xl,
+             _mm_xor_si128(
+                 _mm_xor_si128( _mm_xor_si128( qt[24], qt[25] ),
+                                   _mm_xor_si128( qt[26], qt[27] ) ),
+                 _mm_xor_si128( _mm_xor_si128( qt[28], qt[29] ),
+                                   _mm_xor_si128( qt[30], qt[31] ) )));
+
+   dH[ 0] = _mm_add_epi32(
+                 _mm_xor_si128( M[0],
+                      _mm_xor_si128( _mm_slli_epi32( xh, 5 ),
+                                     _mm_srli_epi32( qt[16], 5 ) ) ),
+                 _mm_xor_si128( _mm_xor_si128( xl, qt[24] ), qt[ 0] ));
+   dH[ 1] = _mm_add_epi32(
+                 _mm_xor_si128( M[1],
+                      _mm_xor_si128( _mm_srli_epi32( xh, 7 ),
+                                     _mm_slli_epi32( qt[17], 8 ) ) ),
+                 _mm_xor_si128( _mm_xor_si128( xl, qt[25] ), qt[ 1] ));
+   dH[ 2] = _mm_add_epi32(
+                 _mm_xor_si128( M[2],
+                      _mm_xor_si128( _mm_srli_epi32( xh, 5 ),
+                                     _mm_slli_epi32( qt[18], 5 ) ) ),
+                 _mm_xor_si128( _mm_xor_si128( xl, qt[26] ), qt[ 2] ));
+   dH[ 3] = _mm_add_epi32(
+                 _mm_xor_si128( M[3],
+                      _mm_xor_si128( _mm_srli_epi32( xh, 1 ),
+                                     _mm_slli_epi32( qt[19], 5 ) ) ),
+                 _mm_xor_si128( _mm_xor_si128( xl, qt[27] ), qt[ 3] ));
+   dH[ 4] = _mm_add_epi32(
+                 _mm_xor_si128( M[4],
+                      _mm_xor_si128( _mm_srli_epi32( xh, 3 ),
+                                     _mm_slli_epi32( qt[20], 0 ) ) ),
+                 _mm_xor_si128( _mm_xor_si128( xl, qt[28] ), qt[ 4] ));
+   dH[ 5] = _mm_add_epi32(
+                 _mm_xor_si128( M[5],
+                      _mm_xor_si128( _mm_slli_epi32( xh, 6 ),
+                                     _mm_srli_epi32( qt[21], 6 ) ) ),
+                 _mm_xor_si128( _mm_xor_si128( xl, qt[29] ), qt[ 5] ));
+   dH[ 6] = _mm_add_epi32(
+                 _mm_xor_si128( M[6],
+                      _mm_xor_si128( _mm_srli_epi32( xh, 4 ),
+                                     _mm_slli_epi32( qt[22], 6 ) ) ),
+                 _mm_xor_si128( _mm_xor_si128( xl, qt[30] ), qt[ 6] ));
+   dH[ 7] = _mm_add_epi32(
+                 _mm_xor_si128( M[7],
+                      _mm_xor_si128( _mm_srli_epi32( xh, 11 ),
+                                     _mm_slli_epi32( qt[23], 2 ) ) ),
+                 _mm_xor_si128( _mm_xor_si128( xl, qt[31] ), qt[ 7] ));
+   dH[ 8] = _mm_add_epi32( _mm_add_epi32(
+                 mm_rotl_32( dH[4], 9 ),
+                 _mm_xor_si128( _mm_xor_si128( xh, qt[24] ), M[ 8] )),
+                 _mm_xor_si128( _mm_slli_epi32( xl, 8 ),
+                                _mm_xor_si128( qt[23], qt[ 8] ) ) );
+   dH[ 9] = _mm_add_epi32( _mm_add_epi32(
+                 mm_rotl_32( dH[5], 10 ),
+                 _mm_xor_si128( _mm_xor_si128( xh, qt[25] ), M[ 9] )),
+                 _mm_xor_si128( _mm_srli_epi32( xl, 6 ),
+                                _mm_xor_si128( qt[16], qt[ 9] ) ) );
+   dH[10] = _mm_add_epi32( _mm_add_epi32(
+                 mm_rotl_32( dH[6], 11 ),
+                 _mm_xor_si128( _mm_xor_si128( xh, qt[26] ), M[10] )),
+                 _mm_xor_si128( _mm_slli_epi32( xl, 6 ),
+                                _mm_xor_si128( qt[17], qt[10] ) ) );
+   dH[11] = _mm_add_epi32( _mm_add_epi32(
+                 mm_rotl_32( dH[7], 12 ),
+                 _mm_xor_si128( _mm_xor_si128( xh, qt[27] ), M[11] )),
+                 _mm_xor_si128( _mm_slli_epi32( xl, 4 ),
+                                _mm_xor_si128( qt[18], qt[11] ) ) );
+   dH[12] = _mm_add_epi32( _mm_add_epi32(
+                 mm_rotl_32( dH[0], 13 ),
+                 _mm_xor_si128( _mm_xor_si128( xh, qt[28] ), M[12] )),
+                 _mm_xor_si128( _mm_srli_epi32( xl, 3 ),
+                                _mm_xor_si128( qt[19], qt[12] ) ) );
+   dH[13] = _mm_add_epi32( _mm_add_epi32(
+                 mm_rotl_32( dH[1], 14 ),
+                 _mm_xor_si128( _mm_xor_si128( xh, qt[29] ), M[13] )),
+                 _mm_xor_si128( _mm_srli_epi32( xl, 4 ),
+                                _mm_xor_si128( qt[20], qt[13] ) ) );
+   dH[14] = _mm_add_epi32( _mm_add_epi32(
+                 mm_rotl_32( dH[2], 15 ),
+                 _mm_xor_si128( _mm_xor_si128( xh, qt[30] ), M[14] )),
+                 _mm_xor_si128( _mm_srli_epi32( xl, 7 ),
+                                _mm_xor_si128( qt[21], qt[14] ) ) );
+   dH[15] = _mm_add_epi32( _mm_add_epi32(
+                 mm_rotl_32( dH[3], 16 ),
+                 _mm_xor_si128( _mm_xor_si128( xh, qt[31] ), M[15] )),
+                 _mm_xor_si128( _mm_srli_epi32( xl, 2 ),
+                                _mm_xor_si128( qt[22], qt[15] ) ) );
+}
+
+// BMW512

 #define Wb0 \
   _mm256_add_epi64( \
@@ -530,24 +731,24 @@ static const sph_u64 IV512[] = {

 void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
 {
-   __m256i qt[32], xl, xh; \
+   __m256i qt[32], xl, xh;

-   qt[ 0] = sb0( Wb0 ) + H[ 1]; 
-   qt[ 1] = sb1( Wb1 ) + H[ 2]; 
-   qt[ 2] = sb2( Wb2 ) + H[ 3]; 
-   qt[ 3] = sb3( Wb3 ) + H[ 4]; 
-   qt[ 4] = sb4( Wb4 ) + H[ 5]; 
-   qt[ 5] = sb0( Wb5 ) + H[ 6]; 
-   qt[ 6] = sb1( Wb6 ) + H[ 7]; 
-   qt[ 7] = sb2( Wb7 ) + H[ 8]; 
-   qt[ 8] = sb3( Wb8 ) + H[ 9]; 
-   qt[ 9] = sb4( Wb9 ) + H[10]; 
-   qt[10] = sb0( Wb10) + H[11]; 
-   qt[11] = sb1( Wb11) + H[12]; 
-   qt[12] = sb2( Wb12) + H[13]; 
-   qt[13] = sb3( Wb13) + H[14];
-   qt[14] = sb4( Wb14) + H[15]; 
-   qt[15] = sb0( Wb15) + H[ 0]; 
+   qt[ 0] = _mm256_add_epi64( sb0( Wb0 ), H[ 1] ); 
+   qt[ 1] = _mm256_add_epi64( sb1( Wb1 ), H[ 2] ); 
+   qt[ 2] = _mm256_add_epi64( sb2( Wb2 ), H[ 3] ); 
+   qt[ 3] = _mm256_add_epi64( sb3( Wb3 ), H[ 4] ); 
+   qt[ 4] = _mm256_add_epi64( sb4( Wb4 ), H[ 5] ); 
+   qt[ 5] = _mm256_add_epi64( sb0( Wb5 ), H[ 6] ); 
+   qt[ 6] = _mm256_add_epi64( sb1( Wb6 ), H[ 7] ); 
+   qt[ 7] = _mm256_add_epi64( sb2( Wb7 ), H[ 8] ); 
+   qt[ 8] = _mm256_add_epi64( sb3( Wb8 ), H[ 9] ); 
+   qt[ 9] = _mm256_add_epi64( sb4( Wb9 ), H[10] ); 
+   qt[10] = _mm256_add_epi64( sb0( Wb10), H[11] ); 
+   qt[11] = _mm256_add_epi64( sb1( Wb11), H[12] ); 
+   qt[12] = _mm256_add_epi64( sb2( Wb12), H[13] ); 
+   qt[13] = _mm256_add_epi64( sb3( Wb13), H[14] );
+   qt[14] = _mm256_add_epi64( sb4( Wb14), H[15] ); 
+   qt[15] = _mm256_add_epi64( sb0( Wb15), H[ 0] ); 
   qt[16] = expand1b( qt, M, H, 16 ); 
   qt[17] = expand1b( qt, M, H, 17 ); 
   qt[18] = expand2b( qt, M, H, 18 ); 
@@ -564,6 +765,7 @@ void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
   qt[29] = expand2b( qt, M, H, 29 ); 
   qt[30] = expand2b( qt, M, H, 30 ); 
   qt[31] = expand2b( qt, M, H, 31 ); 
+
   xl = _mm256_xor_si256( 
              _mm256_xor_si256( _mm256_xor_si256( qt[16], qt[17] ), 
                                _mm256_xor_si256( qt[18], qt[19] ) ), 
@@ -575,6 +777,7 @@ void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
                                   _mm256_xor_si256( qt[26], qt[27] ) ),
                 _mm256_xor_si256( _mm256_xor_si256( qt[28], qt[29] ),
                                   _mm256_xor_si256( qt[30], qt[31] ) )));
+
   dH[ 0] = _mm256_add_epi64(
                 _mm256_xor_si256( M[0],
                      _mm256_xor_si256( _mm256_slli_epi64( xh, 5 ),
@@ -657,137 +860,135 @@ void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
                                   _mm256_xor_si256( qt[22], qt[15] ) ) );
 } 

-#endif  // 64
+// BMW256

-//#define FOLDs   FOLD(sph_u32, MAKE_Qs, SPH_ROTL32, M, Qs, dH)
-
-
-/*
-static void
-compress_small(const unsigned char *data, const sph_u32 h[16], sph_u32 dh[16])
+static const uint32_t final_s[16][4] =
 {
-#define M(x)    sph_dec32le_aligned(data + 4 * (x))
-#define H(x)    (h[x])
-#define dH(x)   (dh[x])
-
-	FOLDs;
-
-#undef M
-#undef H
-#undef dH
-}
-
-static const sph_u32 final_s[16] = {
-	SPH_C32(0xaaaaaaa0), SPH_C32(0xaaaaaaa1), SPH_C32(0xaaaaaaa2),
-	SPH_C32(0xaaaaaaa3), SPH_C32(0xaaaaaaa4), SPH_C32(0xaaaaaaa5),
-	SPH_C32(0xaaaaaaa6), SPH_C32(0xaaaaaaa7), SPH_C32(0xaaaaaaa8),
-	SPH_C32(0xaaaaaaa9), SPH_C32(0xaaaaaaaa), SPH_C32(0xaaaaaaab),
-	SPH_C32(0xaaaaaaac), SPH_C32(0xaaaaaaad), SPH_C32(0xaaaaaaae),
-	SPH_C32(0xaaaaaaaf)
+   { 0xaaaaaaa0, 0xaaaaaaa0, 0xaaaaaaa0, 0xaaaaaaa0 },
+   { 0xaaaaaaa1, 0xaaaaaaa1, 0xaaaaaaa1, 0xaaaaaaa1 },
+   { 0xaaaaaaa2, 0xaaaaaaa2, 0xaaaaaaa2, 0xaaaaaaa2 },
+   { 0xaaaaaaa3, 0xaaaaaaa3, 0xaaaaaaa3, 0xaaaaaaa3 },
+   { 0xaaaaaaa4, 0xaaaaaaa4, 0xaaaaaaa4, 0xaaaaaaa4 },
+   { 0xaaaaaaa5, 0xaaaaaaa5, 0xaaaaaaa5, 0xaaaaaaa5 },
+   { 0xaaaaaaa6, 0xaaaaaaa6, 0xaaaaaaa6, 0xaaaaaaa6 },
+   { 0xaaaaaaa7, 0xaaaaaaa7, 0xaaaaaaa7, 0xaaaaaaa7 },
+   { 0xaaaaaaa8, 0xaaaaaaa8, 0xaaaaaaa8, 0xaaaaaaa8 },
+   { 0xaaaaaaa9, 0xaaaaaaa9, 0xaaaaaaa9, 0xaaaaaaa9 },
+   { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa },
+   { 0xaaaaaaab, 0xaaaaaaab, 0xaaaaaaab, 0xaaaaaaab },
+   { 0xaaaaaaac, 0xaaaaaaac, 0xaaaaaaac, 0xaaaaaaac },
+   { 0xaaaaaaad, 0xaaaaaaad, 0xaaaaaaad, 0xaaaaaaad },
+   { 0xaaaaaaae, 0xaaaaaaae, 0xaaaaaaae, 0xaaaaaaae },
+   { 0xaaaaaaaf, 0xaaaaaaaf, 0xaaaaaaaf, 0xaaaaaaaf }
 };
-
+/*
+static const __m128i final_s[16] =
+{
+   { 0xaaaaaaa0aaaaaaa0, 0xaaaaaaa0aaaaaaa0 },
+   { 0xaaaaaaa1aaaaaaa1, 0xaaaaaaa1aaaaaaa1 },
+   { 0xaaaaaaa2aaaaaaa2, 0xaaaaaaa2aaaaaaa2 },
+   { 0xaaaaaaa3aaaaaaa3, 0xaaaaaaa3aaaaaaa3 },
+   { 0xaaaaaaa4aaaaaaa4, 0xaaaaaaa4aaaaaaa4 },
+   { 0xaaaaaaa5aaaaaaa5, 0xaaaaaaa5aaaaaaa5 },
+   { 0xaaaaaaa6aaaaaaa6, 0xaaaaaaa6aaaaaaa6 },
+   { 0xaaaaaaa7aaaaaaa7, 0xaaaaaaa7aaaaaaa7 },
+   { 0xaaaaaaa8aaaaaaa8, 0xaaaaaaa8aaaaaaa8 },
+   { 0xaaaaaaa9aaaaaaa9, 0xaaaaaaa9aaaaaaa9 },
+   { 0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa },
+   { 0xaaaaaaabaaaaaaab, 0xaaaaaaabaaaaaaab },
+   { 0xaaaaaaacaaaaaaac, 0xaaaaaaacaaaaaaac },
+   { 0xaaaaaaadaaaaaaad, 0xaaaaaaadaaaaaaad },
+   { 0xaaaaaaaeaaaaaaae, 0xaaaaaaaeaaaaaaae },
+   { 0xaaaaaaafaaaaaaaf, 0xaaaaaaafaaaaaaaf }
+};
+*/
 static void
 bmw32_4way_init(bmw_4way_small_context *sc, const sph_u32 *iv)
 {
-	memcpy(sc->H, iv, sizeof sc->H);
-	sc->ptr = 0;
-#if SPH_64
-	sc->bit_count = 0;
-#else
-	sc->bit_count_high = 0;
-	sc->bit_count_low = 0;
-#endif
+   for ( int i = 0; i < 16; i++ )
+      sc->H[i] = _mm_set1_epi32( iv[i] );
+   sc->ptr = 0;
+   sc->bit_count = 0;
 }

 static void
 bmw32_4way(bmw_4way_small_context *sc, const void *data, size_t len)
 {
-	unsigned char *buf;
-	size_t ptr;
-	sph_u32 htmp[16];
-	sph_u32 *h1, *h2;
-#if !SPH_64
-	sph_u32 tmp;
-#endif
+   __m128i *vdata = (__m128i*)data;
+   __m128i *buf;
+   __m128i htmp[16];
+   __m128i *h1, *h2;
+   size_t ptr;
+   const int buf_size = 64;  // bytes of one lane, compatible with len

-#if SPH_64
-	sc->bit_count += (sph_u64)len << 3;
-#else
-	tmp = sc->bit_count_low;
-	sc->bit_count_low = SPH_T32(tmp + ((sph_u32)len << 3));
-	if (sc->bit_count_low < tmp)
-		sc->bit_count_high ++;
-	sc->bit_count_high += len >> 29;
-#endif
-	buf = sc->buf;
-	ptr = sc->ptr;
-	h1 = sc->H;
-	h2 = htmp;
-	while (len > 0) {
-		size_t clen;
+   sc->bit_count += (sph_u32)len << 3;
+   buf = sc->buf;
+   ptr = sc->ptr;
+   h1 = sc->H;
+   h2 = htmp;

-		clen = (sizeof sc->buf) - ptr;
-		if (clen > len)
-			clen = len;
-		memcpy(buf + ptr, data, clen);
-		data = (const unsigned char *)data + clen;
-		len -= clen;
-		ptr += clen;
-		if (ptr == sizeof sc->buf) {
-			sph_u32 *ht;
-
-			compress_small(buf, h1, h2);
-			ht = h1;
-			h1 = h2;
-			h2 = ht;
-			ptr = 0;
-		}
-	}
-	sc->ptr = ptr;
-	if (h1 != sc->H)
-		memcpy(sc->H, h1, sizeof sc->H);
+   while ( len > 0 )
+   {
+      size_t clen;
+      clen = buf_size - ptr;
+      if ( clen > len )
+         clen = len;
+      memcpy_128( buf + (ptr>>2), vdata, clen >> 2 );
+      vdata += ( clen >> 2 );
+      len -= clen;
+      ptr += clen;
+      if ( ptr == buf_size )
+      {
+         __m128i *ht;
+         compress_small( buf, h1, h2 );
+         ht = h1;
+         h1 = h2;
+         h2 = ht;
+         ptr = 0;
+      }
+   }
+   sc->ptr = ptr;
+   if ( h1 != sc->H )
+        memcpy_128( sc->H, h1, 16 );
 }

 static void
 bmw32_4way_close(bmw_4way_small_context *sc, unsigned ub, unsigned n,
 	void *dst, size_t out_size_w32)
 {
-	unsigned char *buf, *out;
-	size_t ptr, u, v;
-	unsigned z;
-	sph_u32 h1[16], h2[16], *h;
+   __m128i *buf;
+   __m128i h1[16], h2[16], *h;
+   size_t ptr, u, v;
+   const int buf_size = 64;  // bytes of one lane, compatible with len

-	buf = sc->buf;
-	ptr = sc->ptr;
-	z = 0x80 >> n;
-	buf[ptr ++] = ((ub & -z) | z) & 0xFF;
-	h = sc->H;
-	if (ptr > (sizeof sc->buf) - 8) {
-		memset(buf + ptr, 0, (sizeof sc->buf) - ptr);
-		compress_small(buf, h, h1);
-		ptr = 0;
-		h = h1;
-	}
-	memset(buf + ptr, 0, (sizeof sc->buf) - 8 - ptr);
-#if SPH_64
-	sph_enc64le_aligned(buf + (sizeof sc->buf) - 8,
-		SPH_T64(sc->bit_count + n));
-#else
-	sph_enc32le_aligned(buf + (sizeof sc->buf) - 8,
-		sc->bit_count_low + n);
-	sph_enc32le_aligned(buf + (sizeof sc->buf) - 4,
-		SPH_T32(sc->bit_count_high));
-#endif
-	compress_small(buf, h, h2);
-	for (u = 0; u < 16; u ++)
-		sph_enc32le_aligned(buf + 4 * u, h2[u]);
-	compress_small(buf, final_s, h1);
-	out = dst;
-	for (u = 0, v = 16 - out_size_w32; u < out_size_w32; u ++, v ++)
-		sph_enc32le(out + 4 * u, h1[v]);
+   buf = sc->buf;
+   ptr = sc->ptr;
+   buf[ ptr>>2 ] = _mm_set1_epi32( 0x80 );
+   ptr += 4;
+   h = sc->H;
+
+   // assume bit_count fits in 32 bits 
+   if ( ptr > buf_size - 4 )
+   {
+      memset_zero_128( buf + (ptr>>2), (buf_size - ptr) >> 2 );
+      compress_small( buf, h, h1 );
+      ptr = 0;
+      h = h1;
+   }
+   memset_zero_128( buf + (ptr>>2), (buf_size - 8 - ptr) >> 2 );
+   buf[ (buf_size - 8) >> 2 ] = _mm_set1_epi32( sc->bit_count + n );
+   buf[ (buf_size - 4) >> 2 ] = m128_zero;
+   compress_small( buf, h, h2 );
+
+   for ( u = 0; u < 16; u ++ )
+      buf[u] = h2[u];
+   compress_small( buf, (__m128i*)final_s, h1 );
+
+   for (u = 0, v = 16 - out_size_w32; u < out_size_w32; u ++, v ++)
+      casti_m128i( dst, u ) = h1[v];
 }
-*/
-#if SPH_64
+
+// BMW512

 static const __m256i final_b[16] =
 {
@@ -908,33 +1109,33 @@ bmw64_4way_close(bmw_4way_big_context *sc, unsigned ub, unsigned n,
      casti_m256i(dst,u) = h1[v];
 }

-#endif
+// BMW256

 void
 bmw256_4way_init(void *cc)
 {
-//	bmw32_4way_init(cc, IV256);
+	bmw32_4way_init(cc, IV256);
 }

 void
 bmw256_4way(void *cc, const void *data, size_t len)
 {
-//	bmw32_4way(cc, data, len);
+	bmw32_4way(cc, data, len);
 }

 void
 bmw256_4way_close(void *cc, void *dst)
 {
-//	bmw256_4way_addbits_and_close(cc, 0, 0, dst);
+	bmw256_4way_addbits_and_close(cc, 0, 0, dst);
 }

 void
 bmw256_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
 {
-//	bmw32_4way_close(cc, ub, n, dst, 8);
+	bmw32_4way_close(cc, ub, n, dst, 8);
 }

-#if SPH_64
+// BMW512

 void
 bmw512_4way_init(void *cc)
@@ -960,10 +1161,8 @@ bmw512_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
 	bmw64_4way_close(cc, ub, n, dst, 8);
 }

-#endif
-
 #ifdef __cplusplus
 }
 #endif

-#endif
+#endif  // __AVX2__
--- a/algo/bmw/bmw-hash-4way.h
+++ b/algo/bmw/bmw-hash-4way.h
@@ -46,94 +46,37 @@ extern "C"{
 #include "algo/sha/sph_types.h"
 #include "avxdefs.h"

-/**
- * Output size (in bits) for BMW-224.
- */
-#define SPH_SIZE_bmw224   224
-
-/**
- * Output size (in bits) for BMW-256.
- */
 #define SPH_SIZE_bmw256   256

-#if SPH_64
-
-/**
- * Output size (in bits) for BMW-384.
- */
-#define SPH_SIZE_bmw384   384
-
-/**
- * Output size (in bits) for BMW-512.
- */
 #define SPH_SIZE_bmw512   512

-#endif
-
-/**
- * This structure is a context for BMW-224 and BMW-256 computations:
- * it contains the intermediate values and some data from the last
- * entered block. Once a BMW computation has been performed, the
- * context can be reused for another computation.
- *
- * The contents of this structure are private. A running BMW
- * computation can be cloned by copying the context (e.g. with a simple
- * <code>memcpy()</code>).
- */
 typedef struct {
-#ifndef DOXYGEN_IGNORE
-	unsigned char buf[64];    /* first field, for alignment */
-	size_t ptr;
-	sph_u32 H[16];
-#if SPH_64
-	sph_u64 bit_count;
-#else
-	sph_u32 bit_count_high, bit_count_low;
-#endif
-#endif
+   __m128i buf[64];
+   __m128i H[16];
+   size_t ptr;
+   sph_u32 bit_count;  // assume bit_count fits in 32 bits
 } bmw_4way_small_context;

 typedef bmw_4way_small_context bmw256_4way_context;

-#if SPH_64
-
-/**
- * This structure is a context for BMW-384 and BMW-512 computations:
- * it contains the intermediate values and some data from the last
- * entered block. Once a BMW computation has been performed, the
- * context can be reused for another computation.
- *
- * The contents of this structure are private. A running BMW
- * computation can be cloned by copying the context (e.g. with a simple
- * <code>memcpy()</code>).
- */
 typedef struct {
-#ifndef DOXYGEN_IGNORE
   __m256i buf[16];
   __m256i H[16];
-
-//	unsigned char buf[128];    /* first field, for alignment */
-	size_t ptr;
-//	sph_u64 H[16];
-	sph_u64 bit_count;
-#endif
+   size_t ptr;
+   sph_u64 bit_count;
 } bmw_4way_big_context;

 typedef bmw_4way_big_context bmw512_4way_context;

-#endif
-
 void bmw256_4way_init(void *cc);

 void bmw256_4way(void *cc, const void *data, size_t len);

 void bmw256_4way_close(void *cc, void *dst);

-void bmw256_addbits_and_close(
+void bmw256_4way_addbits_and_close(
 	void *cc, unsigned ub, unsigned n, void *dst);

-#if SPH_64
-
 void bmw512_4way_init(void *cc);

 void bmw512_4way(void *cc, const void *data, size_t len);
@@ -150,5 +93,3 @@ void bmw512_4way_addbits_and_close(
 #endif

 #endif
-
-#endif
--- a/algo/bmw/bmw.test
+++ b/algo/bmw/bmw.test
--- a/algo/bmw/sse2/bmw.c
+++ b/algo/bmw/sse2/bmw.c
@@ -477,7 +477,7 @@ do { \
        for (u = 0; u < 16; u ++) \
        sph_enc64le_aligned(data + 8 * u, h2[u]); \
        dh = h1; \
-        h = final_b; \
+        h = (sph_u64*)final_b; \
    } \
    /* end wrapped for break loop */ \
    out = dst; \
--- a/algo/cryptonight/cryptonight-aesni.c
+++ b/algo/cryptonight/cryptonight-aesni.c
@@ -3,7 +3,8 @@
 #include "cryptonight.h"
 #include "miner.h"
 #include "crypto/c_keccak.h"
-#include "avxdefs.h"
+#include <immintrin.h>
+//#include "avxdefs.h"

 void aesni_parallel_noxor(uint8_t *long_state, uint8_t *text, uint8_t *ExpandedKey);
 void aesni_parallel_xor(uint8_t *text, uint8_t *ExpandedKey, uint8_t *long_state);
--- a/algo/cubehash/sse2/cubehash_sse2.c
+++ b/algo/cubehash/sse2/cubehash_sse2.c
@@ -10,6 +10,10 @@
 #endif
 #include "cubehash_sse2.h"
 #include "algo/sha/sha3-defs.h"
+#include <stdbool.h>
+#include <unistd.h>
+#include <memory.h>
+#include "avxdefs.h"

 static void transform( cubehashParam *sp )
 {
@@ -125,6 +129,18 @@ static void transform( cubehashParam *sp )
 #endif
 }  // transform

+// Cubehash context initializing is very expensive.
+// Cache the intial value for faster reinitializing.
+cubehashParam cube_ctx_cache __attribute__ ((aligned (64)));
+
+int cubehashReinit( cubehashParam *sp )
+{
+   memcpy( sp, &cube_ctx_cache, sizeof(cubehashParam) );
+   return SUCCESS;
+
+}
+
+// Initialize the cache then copy to sp.
 int cubehashInit(cubehashParam *sp, int hashbitlen, int rounds, int blockbytes)
 {
    int i;
@@ -135,24 +151,26 @@ int cubehashInit(cubehashParam *sp, int hashbitlen, int rounds, int blockbytes)

    /* Sanity checks */
    if ( rounds <= 0 || rounds > 32 )
-         rounds = CUBEHASH_ROUNDS;
+       rounds = CUBEHASH_ROUNDS;
    if ( blockbytes <= 0 || blockbytes >= 256)
-         blockbytes = CUBEHASH_BLOCKBYTES;
+       blockbytes = CUBEHASH_BLOCKBYTES;

    // all sizes of __m128i
-    sp->hashlen   = hashbitlen/128;
-    sp->blocksize = blockbytes/16;
-    sp->rounds    = rounds;
-    sp->pos       = 0;
+    cube_ctx_cache.hashlen   = hashbitlen/128;
+    cube_ctx_cache.blocksize = blockbytes/16;
+    cube_ctx_cache.rounds    = rounds;
+    cube_ctx_cache.pos       = 0;

    for ( i = 0; i < 8; ++i )
-         sp->x[i] = _mm_set_epi32(0, 0, 0, 0);
+       cube_ctx_cache.x[i] = _mm_setzero_si128();;

-    sp->x[0] = _mm_set_epi32( 0, rounds, blockbytes, hashbitlen / 8 );
+    cube_ctx_cache.x[0] = _mm_set_epi32( 0, rounds, blockbytes,
+                                         hashbitlen / 8 );

    for ( i = 0; i < 10; ++i )
-         transform(sp);
-//    sp->pos = 0;
+       transform( &cube_ctx_cache );
+
+    memcpy( sp, &cube_ctx_cache, sizeof(cubehashParam) );
    return SUCCESS;
 }

--- a/algo/cubehash/sse2/cubehash_sse2.h
+++ b/algo/cubehash/sse2/cubehash_sse2.h
@@ -29,6 +29,8 @@ extern "C" {
 #endif

 int cubehashInit(cubehashParam* sp, int hashbitlen, int rounds, int blockbytes);
+// reinitialize context with same parameters, much faster.
+int cubehashReinit( cubehashParam* sp );

 int cubehashUpdate(cubehashParam* sp, const byte *data, size_t size);

--- a/algo/echo/aes_ni/architectures
+++ b/algo/echo/aes_ni/architectures
@@ -1,2 +0,0 @@
-amd64
-x86
--- a/algo/echo/aes_ni/hash.c
+++ b/algo/echo/aes_ni/hash.c
@@ -14,18 +14,20 @@
 * Institute of Applied Mathematics, Middle East Technical University, Turkey.
 *
 */
+#if defined(__AES__)

 #include <memory.h>
 #include "miner.h"
 #include "hash_api.h"
-#include "vperm.h"
-
+//#include "vperm.h"
+#include <immintrin.h>
+/*
 #ifndef NO_AES_NI
 #include <wmmintrin.h>
 #else
 #include <tmmintrin.h>
 #endif
-
+*/

 MYALIGN const unsigned int _k_s0F[] = {0x0F0F0F0F, 0x0F0F0F0F, 0x0F0F0F0F, 0x0F0F0F0F};
 MYALIGN const unsigned int _k_ipt[] = {0x5A2A7000, 0xC2B2E898, 0x52227808, 0xCABAE090, 0x317C4D00, 0x4C01307D, 0xB0FDCC81, 0xCD80B1FC};
@@ -246,7 +248,8 @@ void DumpState(__m128i *ps)
 void Compress(hashState_echo *ctx, const unsigned char *pmsg, unsigned int uBlockCount)
 {
 	unsigned int r, b, i, j;
-	__m128i t1, t2, t3, t4, s1, s2, s3, k1, ktemp;
+//      __m128i t1, t2, t3, t4, s1, s2, s3, k1, ktemp;
+	__m128i t1, t2, s2, k1;
 	__m128i _state[4][4], _state2[4][4], _statebackup[4][4]; 


@@ -396,7 +399,7 @@ HashReturn init_echo(hashState_echo *ctx, int nHashSize)
 {
 	int i, j;

-	ctx->k = _mm_xor_si128(ctx->k, ctx->k);
+        ctx->k = _mm_setzero_si128(); 
 	ctx->processed_bits = 0;
 	ctx->uBufferBytes = 0;

@@ -742,4 +745,4 @@ HashReturn hash_echo(int hashbitlen, const BitSequence *data, DataLength databit
 	return SUCCESS;
 }

-
+#endif
--- a/algo/echo/aes_ni/implementors
+++ b/algo/echo/aes_ni/implementors
@@ -1 +0,0 @@
-Çağdaş Çalık
--- a/algo/echo/aes_ni/vperm.h
+++ b/algo/echo/aes_ni/vperm.h
@@ -1,120 +0,0 @@
-/*
- * file        : vperm.h
- * version     : 1.0.208
- * date        : 14.12.2010
- * 
- * vperm implementation of AES s-box 
- *
- * Credits: Adapted from Mike Hamburg's AES implementation, http://crypto.stanford.edu/vpaes/
- *
- * Cagdas Calik
- * ccalik@metu.edu.tr
- * Institute of Applied Mathematics, Middle East Technical University, Turkey.
- *
- */
-
-#ifndef VPERM_H
-#define VPERM_H
-
-#include "algo/sha/sha3_common.h"
-#include <tmmintrin.h>
-
-/*
-extern const unsigned int _k_s0F[];
-extern const unsigned int _k_ipt[];
-extern const unsigned int _k_opt[];
-extern const unsigned int _k_inv[];
-extern const unsigned int _k_sb1[];
-extern const unsigned int _k_sb2[];
-extern const unsigned int _k_sb3[];
-extern const unsigned int _k_sb4[];
-extern const unsigned int _k_sb5[];
-extern const unsigned int _k_sb7[];
-extern const unsigned int _k_sbo[];
-extern const unsigned int _k_h63[];
-extern const unsigned int _k_hc6[];
-extern const unsigned int _k_h5b[];
-extern const unsigned int _k_h4e[];
-extern const unsigned int _k_h0e[];
-extern const unsigned int _k_h15[];
-extern const unsigned int _k_aesmix1[];
-extern const unsigned int _k_aesmix2[];
-extern const unsigned int _k_aesmix3[];
-extern const unsigned int _k_aesmix4[];
-*/
-
-// input: x, table
-// output: x
-#define TRANSFORM(x, table, t1, t2)\
-	t1 = _mm_andnot_si128(M128(_k_s0F), x);\
-	t1 = _mm_srli_epi32(t1, 4);\
-	x  = _mm_and_si128(x, M128(_k_s0F));\
-	t1 = _mm_shuffle_epi8(*((__m128i*)table + 1), t1);\
-	x  = _mm_shuffle_epi8(*((__m128i*)table + 0), x);\
-	x  = _mm_xor_si128(x, t1)
-
-#if 0
-// compiled erroneously with 32-bit msc compiler
-	t2 = _mm_shuffle_epi8(table[0], x);\
-	x  = _mm_shuffle_epi8(table[1], t1);\
-	x  = _mm_xor_si128(x, t2)
-#endif
-
-// input: x
-// output: t2, t3
-#define SUBSTITUTE_VPERM_CORE(x, t1, t2, t3, t4)\
-	t1 = _mm_andnot_si128(M128(_k_s0F), x);\
-	t1 = _mm_srli_epi32(t1, 4);\
-	x  = _mm_and_si128(x, M128(_k_s0F));\
-	t2 = _mm_shuffle_epi8(*((__m128i*)_k_inv + 1), x);\
-	x  = _mm_xor_si128(x, t1);\
-	t3 = _mm_shuffle_epi8(*((__m128i*)_k_inv + 0), t1);\
-	t3 = _mm_xor_si128(t3, t2);\
-	t4 = _mm_shuffle_epi8(*((__m128i*)_k_inv + 0), x);\
-	t4 = _mm_xor_si128(t4, t2);\
-	t2 = _mm_shuffle_epi8(*((__m128i*)_k_inv + 0), t3);\
-	t2 = _mm_xor_si128(t2, x);\
-	t3 = _mm_shuffle_epi8(*((__m128i*)_k_inv + 0), t4);\
-	t3 = _mm_xor_si128(t3, t1);\
-
-
-// input: x1, x2, table
-// output: y
-#define VPERM_LOOKUP(x1, x2, table, y, t)\
-	t = _mm_shuffle_epi8(*((__m128i*)table + 0), x1);\
-	y = _mm_shuffle_epi8(*((__m128i*)table + 1), x2);\
-	y = _mm_xor_si128(y, t)
-
-
-// input: x
-// output: x
-#define SUBSTITUTE_VPERM(x, t1, t2, t3, t4)  \
-	TRANSFORM(x, _k_ipt, t1, t2);\
-	SUBSTITUTE_VPERM_CORE(x, t1, t2, t3, t4);\
-	VPERM_LOOKUP(t2, t3, _k_sbo, x, t1);\
-	x = _mm_xor_si128(x, M128(_k_h63))
-
-
-// input: x
-// output: x
-#define AES_ROUND_VPERM_CORE(x, t1, t2, t3, t4, s1, s2, s3) \
-	SUBSTITUTE_VPERM_CORE(x, t1, t2, t3, t4);\
-	VPERM_LOOKUP(t2, t3, _k_sb1, s1, t1);\
-	VPERM_LOOKUP(t2, t3, _k_sb2, s2, t1);\
-	s3 = _mm_xor_si128(s1, s2);\
-	x = _mm_shuffle_epi8(s2, M128(_k_aesmix1));\
-	x = _mm_xor_si128(x, _mm_shuffle_epi8(s3, M128(_k_aesmix2)));\
-	x = _mm_xor_si128(x, _mm_shuffle_epi8(s1, M128(_k_aesmix3)));\
-	x = _mm_xor_si128(x, _mm_shuffle_epi8(s1, M128(_k_aesmix4)));\
-	x = _mm_xor_si128(x, M128(_k_h5b))
-
-
-// input: x
-// output: x
-#define AES_ROUND_VPERM(x, t1, t2, t3, t4, s1, s2, s3) \
-	TRANSFORM(x, _k_ipt, t1, t2);\
-	AES_ROUND_VPERM_CORE(x, t1, t2, t3, t4, s1, s2, s3);\
-	TRANSFORM(x, _k_opt, t1, t2)
-
-#endif // VPERM_H
-
--- a/algo/groestl/myr-groestl.c
+++ b/algo/groestl/myr-groestl.c
@@ -1,4 +1,4 @@
-#include "algo-gate-api.h"
+#include "myrgr-gate.h"

 #include <stdio.h>
 #include <stdlib.h>
@@ -10,8 +10,6 @@
 #else
  #include "aes_ni/hash-groestl.h"
 #endif
-
-#include <openssl/sha.h>
 #include "algo/sha/sph_sha2.h"

 typedef struct {
@@ -20,11 +18,7 @@ typedef struct {
 #else
    hashState_groestl       groestl;
 #endif
-#ifndef USE_SPH_SHA
-   SHA256_CTX         sha;
-#else
-   sph_sha256_context sha;
-#endif
+    sph_sha256_context sha;
 } myrgr_ctx_holder;

 myrgr_ctx_holder myrgr_ctx;
@@ -36,44 +30,37 @@ void init_myrgr_ctx()
 #else
     init_groestl (&myrgr_ctx.groestl, 64 );
 #endif
-#ifndef USE_SPH_SHA
-   SHA256_Init( &myrgr_ctx.sha );
-#else
-   sph_sha256_init( &myrgr_ctx.sha );
-#endif
+     sph_sha256_init(&myrgr_ctx.sha);
 }

-void myriadhash( void *output, const void *input )
+void myriad_hash(void *output, const void *input)
 {
-     myrgr_ctx_holder ctx __attribute__ ((aligned (64)));
-     memcpy( &ctx, &myrgr_ctx, sizeof(myrgr_ctx) );
-     uint32_t hash[16] __attribute__ ((aligned (64))); 
+        myrgr_ctx_holder ctx;
+        memcpy( &ctx, &myrgr_ctx, sizeof(myrgr_ctx) );
+
+ 	uint32_t _ALIGN(32) hash[16];

 #ifdef NO_AES_NI
-     sph_groestl512(&ctx.groestl, input, 80);
-     sph_groestl512_close(&ctx.groestl, hash);
+	sph_groestl512(&ctx.groestl, input, 80);
+	sph_groestl512_close(&ctx.groestl, hash);
 #else
-     update_and_final_groestl( &ctx.groestl, (char*)input,
-                               (const char*)input, 640 );
+        update_groestl( &ctx.groestl, (char*)input, 640 );
+        final_groestl( &ctx.groestl, (char*)hash);
 #endif

-#ifndef USE_SPH_SHA
-     SHA256_Update( &ctx.sha, hash, 64 );
-     SHA256_Final( (unsigned char*) hash, &ctx.sha );
-#else
-     sph_sha256(&ctx.sha, hash, 64);
-     sph_sha256_close(&ctx.sha, hash);
-#endif
-     memcpy(output, hash, 32);
+	sph_sha256(&ctx.sha, hash, 64);
+	sph_sha256_close(&ctx.sha, hash);
+
+	memcpy(output, hash, 32);
 }

-int scanhash_myriad( int thr_id, struct work *work, uint32_t max_nonce,
-                     uint64_t *hashes_done)
+int scanhash_myriad(int thr_id, struct work *work,
+	uint32_t max_nonce, uint64_t *hashes_done)
 {
        uint32_t *pdata = work->data;
        uint32_t *ptarget = work->target;

-	uint32_t endiandata[20] __attribute__ ((aligned (64)));
+	uint32_t _ALIGN(64) endiandata[20];
 	const uint32_t first_nonce = pdata[19];
 	uint32_t nonce = first_nonce;

@@ -84,9 +71,9 @@ int scanhash_myriad( int thr_id, struct work *work, uint32_t max_nonce,

 	do {
 		const uint32_t Htarg = ptarget[7];
-		uint32_t hash[8] __attribute__ ((aligned (64)));
+		uint32_t hash[8];
 		be32enc(&endiandata[19], nonce);
-		myriadhash(hash, endiandata);
+		myriad_hash(hash, endiandata);

 		if (hash[7] <= Htarg && fulltest(hash, ptarget)) {
 			pdata[19] = nonce;
@@ -101,14 +88,15 @@ int scanhash_myriad( int thr_id, struct work *work, uint32_t max_nonce,
 	*hashes_done = pdata[19] - first_nonce + 1;
 	return 0;
 }
-
+/*
 bool register_myriad_algo( algo_gate_t* gate )
 {
-    gate->optimizations = SSE2_OPT | AES_OPT | AVX_OPT | AVX2_OPT | SHA_OPT;
+    gate->optimizations = SSE2_OPT | AES_OPT;
    init_myrgr_ctx();
    gate->scanhash = (void*)&scanhash_myriad;
    gate->hash     = (void*)&myriadhash;
+//    gate->hash_alt = (void*)&myriadhash;
    gate->get_max64 = (void*)&get_max64_0x3ffff;
    return true;
 };
-
+*/
--- a/algo/groestl/myrgr-4way.c
+++ b/algo/groestl/myrgr-4way.c
@@ -0,0 +1,108 @@
+#include "myrgr-gate.h"
+
+#if defined(MYRGR_4WAY)
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <string.h>
+
+#include "aes_ni/hash-groestl.h"
+#include "algo/sha/sha2-hash-4way.h"
+
+typedef struct {
+    hashState_groestl       groestl;
+    sha256_4way_context     sha;
+} myrgr_4way_ctx_holder;
+
+myrgr_4way_ctx_holder myrgr_4way_ctx;
+
+void init_myrgr_4way_ctx()
+{
+     init_groestl (&myrgr_4way_ctx.groestl, 64 );
+     sha256_4way_init( &myrgr_4way_ctx.sha );
+}
+
+void myriad_4way_hash( void *output, const void *input )
+{
+     uint32_t hash0[20] __attribute__ ((aligned (64)));
+     uint32_t hash1[20] __attribute__ ((aligned (64)));
+     uint32_t hash2[20] __attribute__ ((aligned (64)));
+     uint32_t hash3[20] __attribute__ ((aligned (64)));
+     uint32_t vhash[16*4] __attribute__ ((aligned (64)));
+     myrgr_4way_ctx_holder ctx;
+     memcpy( &ctx, &myrgr_4way_ctx, sizeof(myrgr_4way_ctx) );
+
+     mm_deinterleave_4x32( hash0, hash1, hash2, hash3, input, 640 );
+
+     update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 640 );
+     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     update_and_final_groestl( &ctx.groestl, (char*)hash1, (char*)hash1, 640 );
+     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     update_and_final_groestl( &ctx.groestl, (char*)hash2, (char*)hash2, 640 );
+     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 640 );
+
+     mm_interleave_4x32( vhash, hash0, hash1, hash2, hash3, 512 );
+
+     sha256_4way( &ctx.sha, vhash, 64 );
+     sha256_4way_close( &ctx.sha, vhash );
+
+     mm_deinterleave_4x32( output, output+32, output+64, output+96,
+                           vhash, 256 );
+}
+
+int scanhash_myriad_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done )
+{
+   uint32_t hash[8*4] __attribute__ ((aligned (64)));
+   uint32_t vdata[20*4] __attribute__ ((aligned (64)));
+   uint32_t _ALIGN(64) edata[20];
+   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;
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   uint32_t *noncep = vdata + 76; // 19*4
+
+/*
+        uint32_t *pdata = work->data;
+        uint32_t *ptarget = work->target;
+
+	uint32_t _ALIGN(64) endiandata[20];
+	const uint32_t first_nonce = pdata[19];
+	uint32_t nonce = first_nonce;
+*/
+   if ( opt_benchmark )
+      ( (uint32_t*)ptarget )[7] = 0x0000ff;
+
+   swab32_array( edata, pdata, 20 );
+   mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
+
+   do {
+      be32enc( noncep,   n   );
+      be32enc( noncep+1, n+1 );
+      be32enc( noncep+2, n+2 );
+      be32enc( noncep+3, n+3 );
+
+      myriad_4way_hash( hash, vdata );
+      pdata[19] = n;
+
+      for ( int i = 0; i < 4; i++ )
+      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
+      {
+          pdata[19] = n+i;
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+      }
+      n += 4;
+   } while ( (num_found == 0) && (n < max_nonce-4)
+                   && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#endif
--- a/algo/groestl/myrgr-gate.c
+++ b/algo/groestl/myrgr-gate.c
@@ -0,0 +1,18 @@
+#include "myrgr-gate.h"
+
+bool register_myriad_algo( algo_gate_t* gate )
+{
+#if defined (MYRGR_4WAY)
+  init_myrgr_4way_ctx();
+  gate->scanhash  = (void*)&scanhash_myriad_4way;
+  gate->hash      = (void*)&myriad_4way_hash;
+#else
+  init_myrgr_ctx();
+  gate->scanhash  = (void*)&scanhash_myriad;
+  gate->hash      = (void*)&myriad_hash;
+#endif
+  gate->optimizations = AES_OPT | AVX2_OPT;
+  gate->get_max64 = (void*)&get_max64_0x3ffff;
+  return true;
+};
+
--- a/algo/groestl/myrgr-gate.h
+++ b/algo/groestl/myrgr-gate.h
@@ -0,0 +1,30 @@
+#ifndef MYRGR_GATE_H__
+#define MYRGR_GATE_H__
+
+#include "algo-gate-api.h"
+#include <stdint.h>
+
+#if defined(__AVX2__) && defined(__AES__)
+  #define MYRGR_4WAY
+#endif
+
+#if defined(MYRGR_4WAY)
+
+void myriad_4way_hash( void *state, const void *input );
+
+int scanhash_myriad_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+
+void init_myrgr_4way_ctx();
+
+#endif
+
+void myriad_hash( void *state, const void *input );
+
+int scanhash_myriad( int thr_id, struct work *work, uint32_t max_nonce,
+                    uint64_t *hashes_done );
+
+void init_myrgr_ctx();
+
+#endif
+
--- a/algo/hamsi/hamsi-hash-4way.c
+++ b/algo/hamsi/hamsi-hash-4way.c
@@ -0,0 +1,935 @@
+/* $Id: hamsi.c 251 2010-10-19 14:31:51Z tp $ */
+/*
+ * Hamsi implementation.
+ *
+ * ==========================(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>
+ */
+
+#include <stddef.h>
+#include <string.h>
+
+//#include "miner.h"
+#include "hamsi-hash-4way.h"
+
+#if defined(__AVX2__)
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+/*
+ * The SPH_HAMSI_EXPAND_* define how many input bits we handle in one
+ * table lookup during message expansion (1 to 8, inclusive). If we note
+ * w the number of bits per message word (w=32 for Hamsi-224/256, w=64
+ * for Hamsi-384/512), r the size of a "row" in 32-bit words (r=8 for
+ * Hamsi-224/256, r=16 for Hamsi-384/512), and n the expansion level,
+ * then we will get t tables (where t=ceil(w/n)) of individual size
+ * 2^n*r*4 (in bytes). The last table may be shorter (e.g. with w=32 and
+ * n=5, there are 7 tables, but the last one uses only two bits on
+ * input, not five).
+ *
+ * Also, we read t rows of r words from RAM. Words in a given row are
+ * concatenated in RAM in that order, so most of the cost is about
+ * reading the first row word; comparatively, cache misses are thus
+ * less expensive with Hamsi-512 (r=16) than with Hamsi-256 (r=8).
+ *
+ * When n=1, tables are "special" in that we omit the first entry of
+ * each table (which always contains 0), so that total table size is
+ * halved.
+ *
+ * We thus have the following (size1 is the cumulative table size of
+ * Hamsi-224/256; size2 is for Hamsi-384/512; similarly, t1 and t2
+ * are for Hamsi-224/256 and Hamsi-384/512, respectively).
+ *
+ *   n      size1      size2    t1    t2
+ * ---------------------------------------
+ *   1       1024       4096    32    64
+ *   2       2048       8192    16    32
+ *   3       2688      10880    11    22
+ *   4       4096      16384     8    16
+ *   5       6272      25600     7    13
+ *   6      10368      41984     6    11
+ *   7      16896      73856     5    10
+ *   8      32768     131072     4     8
+ *
+ * So there is a trade-off: a lower n makes the tables fit better in
+ * L1 cache, but increases the number of memory accesses. The optimal
+ * value depends on the amount of available L1 cache and the relative
+ * impact of a cache miss.
+ *
+ * Experimentally, in ideal benchmark conditions (which are not necessarily
+ * realistic with regards to L1 cache contention), it seems that n=8 is
+ * the best value on "big" architectures (those with 32 kB or more of L1
+ * cache), while n=4 is better on "small" architectures. This was tested
+ * on an Intel Core2 Q6600 (both 32-bit and 64-bit mode), a PowerPC G3
+ * (32 kB L1 cache, hence "big"), and a MIPS-compatible Broadcom BCM3302
+ * (8 kB L1 cache).
+ *
+ * Note: with n=1, the 32 tables (actually implemented as one big table)
+ * are read entirely and sequentially, regardless of the input data,
+ * thus avoiding any data-dependent table access pattern.
+ */
+
+#ifdef _MSC_VER
+#pragma warning (disable: 4146)
+#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),
+	SPH_C32(0x75732032), SPH_C32(0x3434362c), SPH_C32(0x20422d33),
+	SPH_C32(0x30303120), SPH_C32(0x4c657576), SPH_C32(0x656e2d48),
+	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),
+	SPH_C32(0xaaaacccc), SPH_C32(0xf0f0ff00), SPH_C32(0xf0f0cccc),
+	SPH_C32(0xaaaaff00), SPH_C32(0xccccff00), SPH_C32(0xaaaaf0f0),
+	SPH_C32(0xaaaaf0f0), SPH_C32(0xff00cccc), SPH_C32(0xccccf0f0),
+	SPH_C32(0xff00aaaa), SPH_C32(0xccccaaaa), SPH_C32(0xff00f0f0),
+	SPH_C32(0xff00aaaa), SPH_C32(0xf0f0cccc), SPH_C32(0xf0f0ff00),
+	SPH_C32(0xccccaaaa), SPH_C32(0xf0f0ff00), SPH_C32(0xaaaacccc),
+	SPH_C32(0xaaaaff00), SPH_C32(0xf0f0cccc), SPH_C32(0xaaaaf0f0),
+	SPH_C32(0xccccff00), SPH_C32(0xff00cccc), SPH_C32(0xaaaaf0f0),
+	SPH_C32(0xff00aaaa), SPH_C32(0xccccf0f0)
+};
+
+static const sph_u32 alpha_f[] = {
+	SPH_C32(0xcaf9639c), SPH_C32(0x0ff0f9c0), SPH_C32(0x639c0ff0),
+	SPH_C32(0xcaf9f9c0), SPH_C32(0x0ff0f9c0), SPH_C32(0x639ccaf9),
+	SPH_C32(0xf9c00ff0), SPH_C32(0x639ccaf9), SPH_C32(0x639c0ff0),
+	SPH_C32(0xf9c0caf9), SPH_C32(0x0ff0caf9), SPH_C32(0xf9c0639c),
+	SPH_C32(0xf9c0639c), SPH_C32(0xcaf90ff0), SPH_C32(0x0ff0639c),
+	SPH_C32(0xcaf9f9c0), SPH_C32(0x0ff0f9c0), SPH_C32(0xcaf9639c),
+	SPH_C32(0xcaf9f9c0), SPH_C32(0x639c0ff0), SPH_C32(0x639ccaf9),
+	SPH_C32(0x0ff0f9c0), SPH_C32(0x639ccaf9), SPH_C32(0xf9c00ff0),
+	SPH_C32(0xf9c0caf9), SPH_C32(0x639c0ff0), SPH_C32(0xf9c0639c),
+	SPH_C32(0x0ff0caf9), SPH_C32(0xcaf90ff0), SPH_C32(0xf9c0639c),
+	SPH_C32(0xcaf9f9c0), SPH_C32(0x0ff0639c)
+};
+
+// imported from hamsi helper
+
+/* Note: this table lists bits within each byte from least
+   siginificant to most significant. */
+static const sph_u32 T512[64][16] = {
+	{ SPH_C32(0xef0b0270), SPH_C32(0x3afd0000), SPH_C32(0x5dae0000),
+	  SPH_C32(0x69490000), SPH_C32(0x9b0f3c06), SPH_C32(0x4405b5f9),
+	  SPH_C32(0x66140a51), SPH_C32(0x924f5d0a), SPH_C32(0xc96b0030),
+	  SPH_C32(0xe7250000), SPH_C32(0x2f840000), SPH_C32(0x264f0000),
+	  SPH_C32(0x08695bf9), SPH_C32(0x6dfcf137), SPH_C32(0x509f6984),
+	  SPH_C32(0x9e69af68) },
+	{ SPH_C32(0xc96b0030), SPH_C32(0xe7250000), SPH_C32(0x2f840000),
+	  SPH_C32(0x264f0000), SPH_C32(0x08695bf9), SPH_C32(0x6dfcf137),
+	  SPH_C32(0x509f6984), SPH_C32(0x9e69af68), SPH_C32(0x26600240),
+	  SPH_C32(0xddd80000), SPH_C32(0x722a0000), SPH_C32(0x4f060000),
+	  SPH_C32(0x936667ff), SPH_C32(0x29f944ce), SPH_C32(0x368b63d5),
+	  SPH_C32(0x0c26f262) },
+	{ SPH_C32(0x145a3c00), SPH_C32(0xb9e90000), SPH_C32(0x61270000),
+	  SPH_C32(0xf1610000), SPH_C32(0xce613d6c), SPH_C32(0xb0493d78),
+	  SPH_C32(0x47a96720), SPH_C32(0xe18e24c5), SPH_C32(0x23671400),
+	  SPH_C32(0xc8b90000), SPH_C32(0xf4c70000), SPH_C32(0xfb750000),
+	  SPH_C32(0x73cd2465), SPH_C32(0xf8a6a549), SPH_C32(0x02c40a3f),
+	  SPH_C32(0xdc24e61f) },
+	{ SPH_C32(0x23671400), SPH_C32(0xc8b90000), SPH_C32(0xf4c70000),
+	  SPH_C32(0xfb750000), SPH_C32(0x73cd2465), SPH_C32(0xf8a6a549),
+	  SPH_C32(0x02c40a3f), SPH_C32(0xdc24e61f), SPH_C32(0x373d2800),
+	  SPH_C32(0x71500000), SPH_C32(0x95e00000), SPH_C32(0x0a140000),
+	  SPH_C32(0xbdac1909), SPH_C32(0x48ef9831), SPH_C32(0x456d6d1f),
+	  SPH_C32(0x3daac2da) },
+	{ SPH_C32(0x54285c00), SPH_C32(0xeaed0000), SPH_C32(0xc5d60000),
+	  SPH_C32(0xa1c50000), SPH_C32(0xb3a26770), SPH_C32(0x94a5c4e1),
+	  SPH_C32(0x6bb0419d), SPH_C32(0x551b3782), SPH_C32(0x9cbb1800),
+	  SPH_C32(0xb0d30000), SPH_C32(0x92510000), SPH_C32(0xed930000),
+	  SPH_C32(0x593a4345), SPH_C32(0xe114d5f4), SPH_C32(0x430633da),
+	  SPH_C32(0x78cace29) },
+	{ SPH_C32(0x9cbb1800), SPH_C32(0xb0d30000), SPH_C32(0x92510000),
+	  SPH_C32(0xed930000), SPH_C32(0x593a4345), SPH_C32(0xe114d5f4),
+	  SPH_C32(0x430633da), SPH_C32(0x78cace29), SPH_C32(0xc8934400),
+	  SPH_C32(0x5a3e0000), SPH_C32(0x57870000), SPH_C32(0x4c560000),
+	  SPH_C32(0xea982435), SPH_C32(0x75b11115), SPH_C32(0x28b67247),
+	  SPH_C32(0x2dd1f9ab) },
+	{ SPH_C32(0x29449c00), SPH_C32(0x64e70000), SPH_C32(0xf24b0000),
+	  SPH_C32(0xc2f30000), SPH_C32(0x0ede4e8f), SPH_C32(0x56c23745),
+	  SPH_C32(0xf3e04259), SPH_C32(0x8d0d9ec4), SPH_C32(0x466d0c00),
+	  SPH_C32(0x08620000), SPH_C32(0xdd5d0000), SPH_C32(0xbadd0000),
+	  SPH_C32(0x6a927942), SPH_C32(0x441f2b93), SPH_C32(0x218ace6f),
+	  SPH_C32(0xbf2c0be2) },
+	{ SPH_C32(0x466d0c00), SPH_C32(0x08620000), SPH_C32(0xdd5d0000),
+	  SPH_C32(0xbadd0000), SPH_C32(0x6a927942), SPH_C32(0x441f2b93),
+	  SPH_C32(0x218ace6f), SPH_C32(0xbf2c0be2), SPH_C32(0x6f299000),
+	  SPH_C32(0x6c850000), SPH_C32(0x2f160000), SPH_C32(0x782e0000),
+	  SPH_C32(0x644c37cd), SPH_C32(0x12dd1cd6), SPH_C32(0xd26a8c36),
+	  SPH_C32(0x32219526) },
+	{ SPH_C32(0xf6800005), SPH_C32(0x3443c000), SPH_C32(0x24070000),
+	  SPH_C32(0x8f3d0000), SPH_C32(0x21373bfb), SPH_C32(0x0ab8d5ae),
+	  SPH_C32(0xcdc58b19), SPH_C32(0xd795ba31), SPH_C32(0xa67f0001),
+	  SPH_C32(0x71378000), SPH_C32(0x19fc0000), SPH_C32(0x96db0000),
+	  SPH_C32(0x3a8b6dfd), SPH_C32(0xebcaaef3), SPH_C32(0x2c6d478f),
+	  SPH_C32(0xac8e6c88) },
+	{ SPH_C32(0xa67f0001), SPH_C32(0x71378000), SPH_C32(0x19fc0000),
+	  SPH_C32(0x96db0000), SPH_C32(0x3a8b6dfd), SPH_C32(0xebcaaef3),
+	  SPH_C32(0x2c6d478f), SPH_C32(0xac8e6c88), SPH_C32(0x50ff0004),
+	  SPH_C32(0x45744000), SPH_C32(0x3dfb0000), SPH_C32(0x19e60000),
+	  SPH_C32(0x1bbc5606), SPH_C32(0xe1727b5d), SPH_C32(0xe1a8cc96),
+	  SPH_C32(0x7b1bd6b9) },
+	{ SPH_C32(0xf7750009), SPH_C32(0xcf3cc000), SPH_C32(0xc3d60000),
+	  SPH_C32(0x04920000), SPH_C32(0x029519a9), SPH_C32(0xf8e836ba),
+	  SPH_C32(0x7a87f14e), SPH_C32(0x9e16981a), SPH_C32(0xd46a0000),
+	  SPH_C32(0x8dc8c000), SPH_C32(0xa5af0000), SPH_C32(0x4a290000),
+	  SPH_C32(0xfc4e427a), SPH_C32(0xc9b4866c), SPH_C32(0x98369604),
+	  SPH_C32(0xf746c320) },
+	{ SPH_C32(0xd46a0000), SPH_C32(0x8dc8c000), SPH_C32(0xa5af0000),
+	  SPH_C32(0x4a290000), SPH_C32(0xfc4e427a), SPH_C32(0xc9b4866c),
+	  SPH_C32(0x98369604), SPH_C32(0xf746c320), SPH_C32(0x231f0009),
+	  SPH_C32(0x42f40000), SPH_C32(0x66790000), SPH_C32(0x4ebb0000),
+	  SPH_C32(0xfedb5bd3), SPH_C32(0x315cb0d6), SPH_C32(0xe2b1674a),
+	  SPH_C32(0x69505b3a) },
+	{ SPH_C32(0x774400f0), SPH_C32(0xf15a0000), SPH_C32(0xf5b20000),
+	  SPH_C32(0x34140000), SPH_C32(0x89377e8c), SPH_C32(0x5a8bec25),
+	  SPH_C32(0x0bc3cd1e), SPH_C32(0xcf3775cb), SPH_C32(0xf46c0050),
+	  SPH_C32(0x96180000), SPH_C32(0x14a50000), SPH_C32(0x031f0000),
+	  SPH_C32(0x42947eb8), SPH_C32(0x66bf7e19), SPH_C32(0x9ca470d2),
+	  SPH_C32(0x8a341574) },
+	{ SPH_C32(0xf46c0050), SPH_C32(0x96180000), SPH_C32(0x14a50000),
+	  SPH_C32(0x031f0000), SPH_C32(0x42947eb8), SPH_C32(0x66bf7e19),
+	  SPH_C32(0x9ca470d2), SPH_C32(0x8a341574), SPH_C32(0x832800a0),
+	  SPH_C32(0x67420000), SPH_C32(0xe1170000), SPH_C32(0x370b0000),
+	  SPH_C32(0xcba30034), SPH_C32(0x3c34923c), SPH_C32(0x9767bdcc),
+	  SPH_C32(0x450360bf) },
+	{ SPH_C32(0xe8870170), SPH_C32(0x9d720000), SPH_C32(0x12db0000),
+	  SPH_C32(0xd4220000), SPH_C32(0xf2886b27), SPH_C32(0xa921e543),
+	  SPH_C32(0x4ef8b518), SPH_C32(0x618813b1), SPH_C32(0xb4370060),
+	  SPH_C32(0x0c4c0000), SPH_C32(0x56c20000), SPH_C32(0x5cae0000),
+	  SPH_C32(0x94541f3f), SPH_C32(0x3b3ef825), SPH_C32(0x1b365f3d),
+	  SPH_C32(0xf3d45758) },
+	{ SPH_C32(0xb4370060), SPH_C32(0x0c4c0000), SPH_C32(0x56c20000),
+	  SPH_C32(0x5cae0000), SPH_C32(0x94541f3f), SPH_C32(0x3b3ef825),
+	  SPH_C32(0x1b365f3d), SPH_C32(0xf3d45758), SPH_C32(0x5cb00110),
+	  SPH_C32(0x913e0000), SPH_C32(0x44190000), SPH_C32(0x888c0000),
+	  SPH_C32(0x66dc7418), SPH_C32(0x921f1d66), SPH_C32(0x55ceea25),
+	  SPH_C32(0x925c44e9) },
+	{ SPH_C32(0x0c720000), SPH_C32(0x49e50f00), SPH_C32(0x42790000),
+	  SPH_C32(0x5cea0000), SPH_C32(0x33aa301a), SPH_C32(0x15822514),
+	  SPH_C32(0x95a34b7b), SPH_C32(0xb44b0090), SPH_C32(0xfe220000),
+	  SPH_C32(0xa7580500), SPH_C32(0x25d10000), SPH_C32(0xf7600000),
+	  SPH_C32(0x893178da), SPH_C32(0x1fd4f860), SPH_C32(0x4ed0a315),
+	  SPH_C32(0xa123ff9f) },
+	{ SPH_C32(0xfe220000), SPH_C32(0xa7580500), SPH_C32(0x25d10000),
+	  SPH_C32(0xf7600000), SPH_C32(0x893178da), SPH_C32(0x1fd4f860),
+	  SPH_C32(0x4ed0a315), SPH_C32(0xa123ff9f), SPH_C32(0xf2500000),
+	  SPH_C32(0xeebd0a00), SPH_C32(0x67a80000), SPH_C32(0xab8a0000),
+	  SPH_C32(0xba9b48c0), SPH_C32(0x0a56dd74), SPH_C32(0xdb73e86e),
+	  SPH_C32(0x1568ff0f) },
+	{ SPH_C32(0x45180000), SPH_C32(0xa5b51700), SPH_C32(0xf96a0000),
+	  SPH_C32(0x3b480000), SPH_C32(0x1ecc142c), SPH_C32(0x231395d6),
+	  SPH_C32(0x16bca6b0), SPH_C32(0xdf33f4df), SPH_C32(0xb83d0000),
+	  SPH_C32(0x16710600), SPH_C32(0x379a0000), SPH_C32(0xf5b10000),
+	  SPH_C32(0x228161ac), SPH_C32(0xae48f145), SPH_C32(0x66241616),
+	  SPH_C32(0xc5c1eb3e) },
+	{ SPH_C32(0xb83d0000), SPH_C32(0x16710600), SPH_C32(0x379a0000),
+	  SPH_C32(0xf5b10000), SPH_C32(0x228161ac), SPH_C32(0xae48f145),
+	  SPH_C32(0x66241616), SPH_C32(0xc5c1eb3e), SPH_C32(0xfd250000),
+	  SPH_C32(0xb3c41100), SPH_C32(0xcef00000), SPH_C32(0xcef90000),
+	  SPH_C32(0x3c4d7580), SPH_C32(0x8d5b6493), SPH_C32(0x7098b0a6),
+	  SPH_C32(0x1af21fe1) },
+	{ SPH_C32(0x75a40000), SPH_C32(0xc28b2700), SPH_C32(0x94a40000),
+	  SPH_C32(0x90f50000), SPH_C32(0xfb7857e0), SPH_C32(0x49ce0bae),
+	  SPH_C32(0x1767c483), SPH_C32(0xaedf667e), SPH_C32(0xd1660000),
+	  SPH_C32(0x1bbc0300), SPH_C32(0x9eec0000), SPH_C32(0xf6940000),
+	  SPH_C32(0x03024527), SPH_C32(0xcf70fcf2), SPH_C32(0xb4431b17),
+	  SPH_C32(0x857f3c2b) },
+	{ SPH_C32(0xd1660000), SPH_C32(0x1bbc0300), SPH_C32(0x9eec0000),
+	  SPH_C32(0xf6940000), SPH_C32(0x03024527), SPH_C32(0xcf70fcf2),
+	  SPH_C32(0xb4431b17), SPH_C32(0x857f3c2b), SPH_C32(0xa4c20000),
+	  SPH_C32(0xd9372400), SPH_C32(0x0a480000), SPH_C32(0x66610000),
+	  SPH_C32(0xf87a12c7), SPH_C32(0x86bef75c), SPH_C32(0xa324df94),
+	  SPH_C32(0x2ba05a55) },
+	{ SPH_C32(0x75c90003), SPH_C32(0x0e10c000), SPH_C32(0xd1200000),
+	  SPH_C32(0xbaea0000), SPH_C32(0x8bc42f3e), SPH_C32(0x8758b757),
+	  SPH_C32(0xbb28761d), SPH_C32(0x00b72e2b), SPH_C32(0xeecf0001),
+	  SPH_C32(0x6f564000), SPH_C32(0xf33e0000), SPH_C32(0xa79e0000),
+	  SPH_C32(0xbdb57219), SPH_C32(0xb711ebc5), SPH_C32(0x4a3b40ba),
+	  SPH_C32(0xfeabf254) },
+	{ SPH_C32(0xeecf0001), SPH_C32(0x6f564000), SPH_C32(0xf33e0000),
+	  SPH_C32(0xa79e0000), SPH_C32(0xbdb57219), SPH_C32(0xb711ebc5),
+	  SPH_C32(0x4a3b40ba), SPH_C32(0xfeabf254), SPH_C32(0x9b060002),
+	  SPH_C32(0x61468000), SPH_C32(0x221e0000), SPH_C32(0x1d740000),
+	  SPH_C32(0x36715d27), SPH_C32(0x30495c92), SPH_C32(0xf11336a7),
+	  SPH_C32(0xfe1cdc7f) },
+	{ SPH_C32(0x86790000), SPH_C32(0x3f390002), SPH_C32(0xe19ae000),
+	  SPH_C32(0x98560000), SPH_C32(0x9565670e), SPH_C32(0x4e88c8ea),
+	  SPH_C32(0xd3dd4944), SPH_C32(0x161ddab9), SPH_C32(0x30b70000),
+	  SPH_C32(0xe5d00000), SPH_C32(0xf4f46000), SPH_C32(0x42c40000),
+	  SPH_C32(0x63b83d6a), SPH_C32(0x78ba9460), SPH_C32(0x21afa1ea),
+	  SPH_C32(0xb0a51834) },
+	{ SPH_C32(0x30b70000), SPH_C32(0xe5d00000), SPH_C32(0xf4f46000),
+	  SPH_C32(0x42c40000), SPH_C32(0x63b83d6a), SPH_C32(0x78ba9460),
+	  SPH_C32(0x21afa1ea), SPH_C32(0xb0a51834), SPH_C32(0xb6ce0000),
+	  SPH_C32(0xdae90002), SPH_C32(0x156e8000), SPH_C32(0xda920000),
+	  SPH_C32(0xf6dd5a64), SPH_C32(0x36325c8a), SPH_C32(0xf272e8ae),
+	  SPH_C32(0xa6b8c28d) },
+	{ SPH_C32(0x14190000), SPH_C32(0x23ca003c), SPH_C32(0x50df0000),
+	  SPH_C32(0x44b60000), SPH_C32(0x1b6c67b0), SPH_C32(0x3cf3ac75),
+	  SPH_C32(0x61e610b0), SPH_C32(0xdbcadb80), SPH_C32(0xe3430000),
+	  SPH_C32(0x3a4e0014), SPH_C32(0xf2c60000), SPH_C32(0xaa4e0000),
+	  SPH_C32(0xdb1e42a6), SPH_C32(0x256bbe15), SPH_C32(0x123db156),
+	  SPH_C32(0x3a4e99d7) },
+	{ SPH_C32(0xe3430000), SPH_C32(0x3a4e0014), SPH_C32(0xf2c60000),
+	  SPH_C32(0xaa4e0000), SPH_C32(0xdb1e42a6), SPH_C32(0x256bbe15),
+	  SPH_C32(0x123db156), SPH_C32(0x3a4e99d7), SPH_C32(0xf75a0000),
+	  SPH_C32(0x19840028), SPH_C32(0xa2190000), SPH_C32(0xeef80000),
+	  SPH_C32(0xc0722516), SPH_C32(0x19981260), SPH_C32(0x73dba1e6),
+	  SPH_C32(0xe1844257) },
+	{ SPH_C32(0x54500000), SPH_C32(0x0671005c), SPH_C32(0x25ae0000),
+	  SPH_C32(0x6a1e0000), SPH_C32(0x2ea54edf), SPH_C32(0x664e8512),
+	  SPH_C32(0xbfba18c3), SPH_C32(0x7e715d17), SPH_C32(0xbc8d0000),
+	  SPH_C32(0xfc3b0018), SPH_C32(0x19830000), SPH_C32(0xd10b0000),
+	  SPH_C32(0xae1878c4), SPH_C32(0x42a69856), SPH_C32(0x0012da37),
+	  SPH_C32(0x2c3b504e) },
+	{ SPH_C32(0xbc8d0000), SPH_C32(0xfc3b0018), SPH_C32(0x19830000),
+	  SPH_C32(0xd10b0000), SPH_C32(0xae1878c4), SPH_C32(0x42a69856),
+	  SPH_C32(0x0012da37), SPH_C32(0x2c3b504e), SPH_C32(0xe8dd0000),
+	  SPH_C32(0xfa4a0044), SPH_C32(0x3c2d0000), SPH_C32(0xbb150000),
+	  SPH_C32(0x80bd361b), SPH_C32(0x24e81d44), SPH_C32(0xbfa8c2f4),
+	  SPH_C32(0x524a0d59) },
+	{ SPH_C32(0x69510000), SPH_C32(0xd4e1009c), SPH_C32(0xc3230000),
+	  SPH_C32(0xac2f0000), SPH_C32(0xe4950bae), SPH_C32(0xcea415dc),
+	  SPH_C32(0x87ec287c), SPH_C32(0xbce1a3ce), SPH_C32(0xc6730000),
+	  SPH_C32(0xaf8d000c), SPH_C32(0xa4c10000), SPH_C32(0x218d0000),
+	  SPH_C32(0x23111587), SPH_C32(0x7913512f), SPH_C32(0x1d28ac88),
+	  SPH_C32(0x378dd173) },
+	{ SPH_C32(0xc6730000), SPH_C32(0xaf8d000c), SPH_C32(0xa4c10000),
+	  SPH_C32(0x218d0000), SPH_C32(0x23111587), SPH_C32(0x7913512f),
+	  SPH_C32(0x1d28ac88), SPH_C32(0x378dd173), SPH_C32(0xaf220000),
+	  SPH_C32(0x7b6c0090), SPH_C32(0x67e20000), SPH_C32(0x8da20000),
+	  SPH_C32(0xc7841e29), SPH_C32(0xb7b744f3), SPH_C32(0x9ac484f4),
+	  SPH_C32(0x8b6c72bd) },
+	{ SPH_C32(0xcc140000), SPH_C32(0xa5630000), SPH_C32(0x5ab90780),
+	  SPH_C32(0x3b500000), SPH_C32(0x4bd013ff), SPH_C32(0x879b3418),
+	  SPH_C32(0x694348c1), SPH_C32(0xca5a87fe), SPH_C32(0x819e0000),
+	  SPH_C32(0xec570000), SPH_C32(0x66320280), SPH_C32(0x95f30000),
+	  SPH_C32(0x5da92802), SPH_C32(0x48f43cbc), SPH_C32(0xe65aa22d),
+	  SPH_C32(0x8e67b7fa) },
+	{ SPH_C32(0x819e0000), SPH_C32(0xec570000), SPH_C32(0x66320280),
+	  SPH_C32(0x95f30000), SPH_C32(0x5da92802), SPH_C32(0x48f43cbc),
+	  SPH_C32(0xe65aa22d), SPH_C32(0x8e67b7fa), SPH_C32(0x4d8a0000),
+	  SPH_C32(0x49340000), SPH_C32(0x3c8b0500), SPH_C32(0xaea30000),
+	  SPH_C32(0x16793bfd), SPH_C32(0xcf6f08a4), SPH_C32(0x8f19eaec),
+	  SPH_C32(0x443d3004) },
+	{ SPH_C32(0x78230000), SPH_C32(0x12fc0000), SPH_C32(0xa93a0b80),
+	  SPH_C32(0x90a50000), SPH_C32(0x713e2879), SPH_C32(0x7ee98924),
+	  SPH_C32(0xf08ca062), SPH_C32(0x636f8bab), SPH_C32(0x02af0000),
+	  SPH_C32(0xb7280000), SPH_C32(0xba1c0300), SPH_C32(0x56980000),
+	  SPH_C32(0xba8d45d3), SPH_C32(0x8048c667), SPH_C32(0xa95c149a),
+	  SPH_C32(0xf4f6ea7b) },
+	{ SPH_C32(0x02af0000), SPH_C32(0xb7280000), SPH_C32(0xba1c0300),
+	  SPH_C32(0x56980000), SPH_C32(0xba8d45d3), SPH_C32(0x8048c667),
+	  SPH_C32(0xa95c149a), SPH_C32(0xf4f6ea7b), SPH_C32(0x7a8c0000),
+	  SPH_C32(0xa5d40000), SPH_C32(0x13260880), SPH_C32(0xc63d0000),
+	  SPH_C32(0xcbb36daa), SPH_C32(0xfea14f43), SPH_C32(0x59d0b4f8),
+	  SPH_C32(0x979961d0) },
+	{ SPH_C32(0xac480000), SPH_C32(0x1ba60000), SPH_C32(0x45fb1380),
+	  SPH_C32(0x03430000), SPH_C32(0x5a85316a), SPH_C32(0x1fb250b6),
+	  SPH_C32(0xfe72c7fe), SPH_C32(0x91e478f6), SPH_C32(0x1e4e0000),
+	  SPH_C32(0xdecf0000), SPH_C32(0x6df80180), SPH_C32(0x77240000),
+	  SPH_C32(0xec47079e), SPH_C32(0xf4a0694e), SPH_C32(0xcda31812),
+	  SPH_C32(0x98aa496e) },
+	{ SPH_C32(0x1e4e0000), SPH_C32(0xdecf0000), SPH_C32(0x6df80180),
+	  SPH_C32(0x77240000), SPH_C32(0xec47079e), SPH_C32(0xf4a0694e),
+	  SPH_C32(0xcda31812), SPH_C32(0x98aa496e), SPH_C32(0xb2060000),
+	  SPH_C32(0xc5690000), SPH_C32(0x28031200), SPH_C32(0x74670000),
+	  SPH_C32(0xb6c236f4), SPH_C32(0xeb1239f8), SPH_C32(0x33d1dfec),
+	  SPH_C32(0x094e3198) },
+	{ SPH_C32(0xaec30000), SPH_C32(0x9c4f0001), SPH_C32(0x79d1e000),
+	  SPH_C32(0x2c150000), SPH_C32(0x45cc75b3), SPH_C32(0x6650b736),
+	  SPH_C32(0xab92f78f), SPH_C32(0xa312567b), SPH_C32(0xdb250000),
+	  SPH_C32(0x09290000), SPH_C32(0x49aac000), SPH_C32(0x81e10000),
+	  SPH_C32(0xcafe6b59), SPH_C32(0x42793431), SPH_C32(0x43566b76),
+	  SPH_C32(0xe86cba2e) },
+	{ SPH_C32(0xdb250000), SPH_C32(0x09290000), SPH_C32(0x49aac000),
+	  SPH_C32(0x81e10000), SPH_C32(0xcafe6b59), SPH_C32(0x42793431),
+	  SPH_C32(0x43566b76), SPH_C32(0xe86cba2e), SPH_C32(0x75e60000),
+	  SPH_C32(0x95660001), SPH_C32(0x307b2000), SPH_C32(0xadf40000),
+	  SPH_C32(0x8f321eea), SPH_C32(0x24298307), SPH_C32(0xe8c49cf9),
+	  SPH_C32(0x4b7eec55) },
+	{ SPH_C32(0x58430000), SPH_C32(0x807e0000), SPH_C32(0x78330001),
+	  SPH_C32(0xc66b3800), SPH_C32(0xe7375cdc), SPH_C32(0x79ad3fdd),
+	  SPH_C32(0xac73fe6f), SPH_C32(0x3a4479b1), SPH_C32(0x1d5a0000),
+	  SPH_C32(0x2b720000), SPH_C32(0x488d0000), SPH_C32(0xaf611800),
+	  SPH_C32(0x25cb2ec5), SPH_C32(0xc879bfd0), SPH_C32(0x81a20429),
+	  SPH_C32(0x1e7536a6) },
+	{ SPH_C32(0x1d5a0000), SPH_C32(0x2b720000), SPH_C32(0x488d0000),
+	  SPH_C32(0xaf611800), SPH_C32(0x25cb2ec5), SPH_C32(0xc879bfd0),
+	  SPH_C32(0x81a20429), SPH_C32(0x1e7536a6), SPH_C32(0x45190000),
+	  SPH_C32(0xab0c0000), SPH_C32(0x30be0001), SPH_C32(0x690a2000),
+	  SPH_C32(0xc2fc7219), SPH_C32(0xb1d4800d), SPH_C32(0x2dd1fa46),
+	  SPH_C32(0x24314f17) },
+	{ SPH_C32(0xa53b0000), SPH_C32(0x14260000), SPH_C32(0x4e30001e),
+	  SPH_C32(0x7cae0000), SPH_C32(0x8f9e0dd5), SPH_C32(0x78dfaa3d),
+	  SPH_C32(0xf73168d8), SPH_C32(0x0b1b4946), SPH_C32(0x07ed0000),
+	  SPH_C32(0xb2500000), SPH_C32(0x8774000a), SPH_C32(0x970d0000),
+	  SPH_C32(0x437223ae), SPH_C32(0x48c76ea4), SPH_C32(0xf4786222),
+	  SPH_C32(0x9075b1ce) },
+	{ SPH_C32(0x07ed0000), SPH_C32(0xb2500000), SPH_C32(0x8774000a),
+	  SPH_C32(0x970d0000), SPH_C32(0x437223ae), SPH_C32(0x48c76ea4),
+	  SPH_C32(0xf4786222), SPH_C32(0x9075b1ce), SPH_C32(0xa2d60000),
+	  SPH_C32(0xa6760000), SPH_C32(0xc9440014), SPH_C32(0xeba30000),
+	  SPH_C32(0xccec2e7b), SPH_C32(0x3018c499), SPH_C32(0x03490afa),
+	  SPH_C32(0x9b6ef888) },
+	{ SPH_C32(0x88980000), SPH_C32(0x1f940000), SPH_C32(0x7fcf002e),
+	  SPH_C32(0xfb4e0000), SPH_C32(0xf158079a), SPH_C32(0x61ae9167),
+	  SPH_C32(0xa895706c), SPH_C32(0xe6107494), SPH_C32(0x0bc20000),
+	  SPH_C32(0xdb630000), SPH_C32(0x7e88000c), SPH_C32(0x15860000),
+	  SPH_C32(0x91fd48f3), SPH_C32(0x7581bb43), SPH_C32(0xf460449e),
+	  SPH_C32(0xd8b61463) },
+	{ SPH_C32(0x0bc20000), SPH_C32(0xdb630000), SPH_C32(0x7e88000c),
+	  SPH_C32(0x15860000), SPH_C32(0x91fd48f3), SPH_C32(0x7581bb43),
+	  SPH_C32(0xf460449e), SPH_C32(0xd8b61463), SPH_C32(0x835a0000),
+	  SPH_C32(0xc4f70000), SPH_C32(0x01470022), SPH_C32(0xeec80000),
+	  SPH_C32(0x60a54f69), SPH_C32(0x142f2a24), SPH_C32(0x5cf534f2),
+	  SPH_C32(0x3ea660f7) },
+	{ SPH_C32(0x52500000), SPH_C32(0x29540000), SPH_C32(0x6a61004e),
+	  SPH_C32(0xf0ff0000), SPH_C32(0x9a317eec), SPH_C32(0x452341ce),
+	  SPH_C32(0xcf568fe5), SPH_C32(0x5303130f), SPH_C32(0x538d0000),
+	  SPH_C32(0xa9fc0000), SPH_C32(0x9ef70006), SPH_C32(0x56ff0000),
+	  SPH_C32(0x0ae4004e), SPH_C32(0x92c5cdf9), SPH_C32(0xa9444018),
+	  SPH_C32(0x7f975691) },
+	{ SPH_C32(0x538d0000), SPH_C32(0xa9fc0000), SPH_C32(0x9ef70006),
+	  SPH_C32(0x56ff0000), SPH_C32(0x0ae4004e), SPH_C32(0x92c5cdf9),
+	  SPH_C32(0xa9444018), SPH_C32(0x7f975691), SPH_C32(0x01dd0000),
+	  SPH_C32(0x80a80000), SPH_C32(0xf4960048), SPH_C32(0xa6000000),
+	  SPH_C32(0x90d57ea2), SPH_C32(0xd7e68c37), SPH_C32(0x6612cffd),
+	  SPH_C32(0x2c94459e) },
+	{ SPH_C32(0xe6280000), SPH_C32(0x4c4b0000), SPH_C32(0xa8550000),
+	  SPH_C32(0xd3d002e0), SPH_C32(0xd86130b8), SPH_C32(0x98a7b0da),
+	  SPH_C32(0x289506b4), SPH_C32(0xd75a4897), SPH_C32(0xf0c50000),
+	  SPH_C32(0x59230000), SPH_C32(0x45820000), SPH_C32(0xe18d00c0),
+	  SPH_C32(0x3b6d0631), SPH_C32(0xc2ed5699), SPH_C32(0xcbe0fe1c),
+	  SPH_C32(0x56a7b19f) },
+	{ SPH_C32(0xf0c50000), SPH_C32(0x59230000), SPH_C32(0x45820000),
+	  SPH_C32(0xe18d00c0), SPH_C32(0x3b6d0631), SPH_C32(0xc2ed5699),
+	  SPH_C32(0xcbe0fe1c), SPH_C32(0x56a7b19f), SPH_C32(0x16ed0000),
+	  SPH_C32(0x15680000), SPH_C32(0xedd70000), SPH_C32(0x325d0220),
+	  SPH_C32(0xe30c3689), SPH_C32(0x5a4ae643), SPH_C32(0xe375f8a8),
+	  SPH_C32(0x81fdf908) },
+	{ SPH_C32(0xb4310000), SPH_C32(0x77330000), SPH_C32(0xb15d0000),
+	  SPH_C32(0x7fd004e0), SPH_C32(0x78a26138), SPH_C32(0xd116c35d),
+	  SPH_C32(0xd256d489), SPH_C32(0x4e6f74de), SPH_C32(0xe3060000),
+	  SPH_C32(0xbdc10000), SPH_C32(0x87130000), SPH_C32(0xbff20060),
+	  SPH_C32(0x2eba0a1a), SPH_C32(0x8db53751), SPH_C32(0x73c5ab06),
+	  SPH_C32(0x5bd61539) },
+	{ SPH_C32(0xe3060000), SPH_C32(0xbdc10000), SPH_C32(0x87130000),
+	  SPH_C32(0xbff20060), SPH_C32(0x2eba0a1a), SPH_C32(0x8db53751),
+	  SPH_C32(0x73c5ab06), SPH_C32(0x5bd61539), SPH_C32(0x57370000),
+	  SPH_C32(0xcaf20000), SPH_C32(0x364e0000), SPH_C32(0xc0220480),
+	  SPH_C32(0x56186b22), SPH_C32(0x5ca3f40c), SPH_C32(0xa1937f8f),
+	  SPH_C32(0x15b961e7) },
+	{ SPH_C32(0x02f20000), SPH_C32(0xa2810000), SPH_C32(0x873f0000),
+	  SPH_C32(0xe36c7800), SPH_C32(0x1e1d74ef), SPH_C32(0x073d2bd6),
+	  SPH_C32(0xc4c23237), SPH_C32(0x7f32259e), SPH_C32(0xbadd0000),
+	  SPH_C32(0x13ad0000), SPH_C32(0xb7e70000), SPH_C32(0xf7282800),
+	  SPH_C32(0xdf45144d), SPH_C32(0x361ac33a), SPH_C32(0xea5a8d14),
+	  SPH_C32(0x2a2c18f0) },
+	{ SPH_C32(0xbadd0000), SPH_C32(0x13ad0000), SPH_C32(0xb7e70000),
+	  SPH_C32(0xf7282800), SPH_C32(0xdf45144d), SPH_C32(0x361ac33a),
+	  SPH_C32(0xea5a8d14), SPH_C32(0x2a2c18f0), SPH_C32(0xb82f0000),
+	  SPH_C32(0xb12c0000), SPH_C32(0x30d80000), SPH_C32(0x14445000),
+	  SPH_C32(0xc15860a2), SPH_C32(0x3127e8ec), SPH_C32(0x2e98bf23),
+	  SPH_C32(0x551e3d6e) },
+	{ SPH_C32(0x1e6c0000), SPH_C32(0xc4420000), SPH_C32(0x8a2e0000),
+	  SPH_C32(0xbcb6b800), SPH_C32(0x2c4413b6), SPH_C32(0x8bfdd3da),
+	  SPH_C32(0x6a0c1bc8), SPH_C32(0xb99dc2eb), SPH_C32(0x92560000),
+	  SPH_C32(0x1eda0000), SPH_C32(0xea510000), SPH_C32(0xe8b13000),
+	  SPH_C32(0xa93556a5), SPH_C32(0xebfb6199), SPH_C32(0xb15c2254),
+	  SPH_C32(0x33c5244f) },
+	{ SPH_C32(0x92560000), SPH_C32(0x1eda0000), SPH_C32(0xea510000),
+	  SPH_C32(0xe8b13000), SPH_C32(0xa93556a5), SPH_C32(0xebfb6199),
+	  SPH_C32(0xb15c2254), SPH_C32(0x33c5244f), SPH_C32(0x8c3a0000),
+	  SPH_C32(0xda980000), SPH_C32(0x607f0000), SPH_C32(0x54078800),
+	  SPH_C32(0x85714513), SPH_C32(0x6006b243), SPH_C32(0xdb50399c),
+	  SPH_C32(0x8a58e6a4) },
+	{ SPH_C32(0x033d0000), SPH_C32(0x08b30000), SPH_C32(0xf33a0000),
+	  SPH_C32(0x3ac20007), SPH_C32(0x51298a50), SPH_C32(0x6b6e661f),
+	  SPH_C32(0x0ea5cfe3), SPH_C32(0xe6da7ffe), SPH_C32(0xa8da0000),
+	  SPH_C32(0x96be0000), SPH_C32(0x5c1d0000), SPH_C32(0x07da0002),
+	  SPH_C32(0x7d669583), SPH_C32(0x1f98708a), SPH_C32(0xbb668808),
+	  SPH_C32(0xda878000) },
+	{ SPH_C32(0xa8da0000), SPH_C32(0x96be0000), SPH_C32(0x5c1d0000),
+	  SPH_C32(0x07da0002), SPH_C32(0x7d669583), SPH_C32(0x1f98708a),
+	  SPH_C32(0xbb668808), SPH_C32(0xda878000), SPH_C32(0xabe70000),
+	  SPH_C32(0x9e0d0000), SPH_C32(0xaf270000), SPH_C32(0x3d180005),
+	  SPH_C32(0x2c4f1fd3), SPH_C32(0x74f61695), SPH_C32(0xb5c347eb),
+	  SPH_C32(0x3c5dfffe) },
+	{ SPH_C32(0x01930000), SPH_C32(0xe7820000), SPH_C32(0xedfb0000),
+	  SPH_C32(0xcf0c000b), SPH_C32(0x8dd08d58), SPH_C32(0xbca3b42e),
+	  SPH_C32(0x063661e1), SPH_C32(0x536f9e7b), SPH_C32(0x92280000),
+	  SPH_C32(0xdc850000), SPH_C32(0x57fa0000), SPH_C32(0x56dc0003),
+	  SPH_C32(0xbae92316), SPH_C32(0x5aefa30c), SPH_C32(0x90cef752),
+	  SPH_C32(0x7b1675d7) },
+	{ SPH_C32(0x92280000), SPH_C32(0xdc850000), SPH_C32(0x57fa0000),
+	  SPH_C32(0x56dc0003), SPH_C32(0xbae92316), SPH_C32(0x5aefa30c),
+	  SPH_C32(0x90cef752), SPH_C32(0x7b1675d7), SPH_C32(0x93bb0000),
+	  SPH_C32(0x3b070000), SPH_C32(0xba010000), SPH_C32(0x99d00008),
+	  SPH_C32(0x3739ae4e), SPH_C32(0xe64c1722), SPH_C32(0x96f896b3),
+	  SPH_C32(0x2879ebac) },
+	{ SPH_C32(0x5fa80000), SPH_C32(0x56030000), SPH_C32(0x43ae0000),
+	  SPH_C32(0x64f30013), SPH_C32(0x257e86bf), SPH_C32(0x1311944e),
+	  SPH_C32(0x541e95bf), SPH_C32(0x8ea4db69), SPH_C32(0x00440000),
+	  SPH_C32(0x7f480000), SPH_C32(0xda7c0000), SPH_C32(0x2a230001),
+	  SPH_C32(0x3badc9cc), SPH_C32(0xa9b69c87), SPH_C32(0x030a9e60),
+	  SPH_C32(0xbe0a679e) },
+	{ SPH_C32(0x00440000), SPH_C32(0x7f480000), SPH_C32(0xda7c0000),
+	  SPH_C32(0x2a230001), SPH_C32(0x3badc9cc), SPH_C32(0xa9b69c87),
+	  SPH_C32(0x030a9e60), SPH_C32(0xbe0a679e), SPH_C32(0x5fec0000),
+	  SPH_C32(0x294b0000), SPH_C32(0x99d20000), SPH_C32(0x4ed00012),
+	  SPH_C32(0x1ed34f73), SPH_C32(0xbaa708c9), SPH_C32(0x57140bdf),
+	  SPH_C32(0x30aebcf7) },
+	{ SPH_C32(0xee930000), SPH_C32(0xd6070000), SPH_C32(0x92c10000),
+	  SPH_C32(0x2b9801e0), SPH_C32(0x9451287c), SPH_C32(0x3b6cfb57),
+	  SPH_C32(0x45312374), SPH_C32(0x201f6a64), SPH_C32(0x7b280000),
+	  SPH_C32(0x57420000), SPH_C32(0xa9e50000), SPH_C32(0x634300a0),
+	  SPH_C32(0x9edb442f), SPH_C32(0x6d9995bb), SPH_C32(0x27f83b03),
+	  SPH_C32(0xc7ff60f0) },
+	{ SPH_C32(0x7b280000), SPH_C32(0x57420000), SPH_C32(0xa9e50000),
+	  SPH_C32(0x634300a0), SPH_C32(0x9edb442f), SPH_C32(0x6d9995bb),
+	  SPH_C32(0x27f83b03), SPH_C32(0xc7ff60f0), SPH_C32(0x95bb0000),
+	  SPH_C32(0x81450000), SPH_C32(0x3b240000), SPH_C32(0x48db0140),
+	  SPH_C32(0x0a8a6c53), SPH_C32(0x56f56eec), SPH_C32(0x62c91877),
+	  SPH_C32(0xe7e00a94) }
+};
+
+#define INPUT_BIG \
+do { \
+  __m256i db = *buf; \
+  const sph_u32 *tp = &T512[0][0]; \
+  m0 = m256_zero; \
+  m1 = m256_zero; \
+  m2 = m256_zero; \
+  m3 = m256_zero; \
+  m4 = m256_zero; \
+  m5 = m256_zero; \
+  m6 = m256_zero; \
+  m7 = m256_zero; \
+  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] ) ) ); \
+     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] ) ) ); \
+     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] ) ) ); \
+     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] ) ) ); \
+     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] ) ) ); \
+     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] ) ) ); \
+     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] ) ) ); \
+     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; \
+     db = _mm256_srli_epi64( db, 1 ); \
+  } \
+} while (0)
+
+#define SBOX( a, b, c, d ) \
+do { \
+  __m256i t; \
+  t = a; \
+  a = _mm256_and_si256( a, c ); \
+  a = _mm256_xor_si256( a, d ); \
+  c = _mm256_xor_si256( c, b ); \
+  c = _mm256_xor_si256( c, a ); \
+  d = _mm256_or_si256( d, t ); \
+  d = _mm256_xor_si256( d, b ); \
+  t = _mm256_xor_si256( t, c ); \
+  b = d; \
+  d = _mm256_or_si256( d, t ); \
+  d = _mm256_xor_si256( d, a ); \
+  a = _mm256_and_si256( a, b ); \
+  t = _mm256_xor_si256( t, a ); \
+  b = _mm256_xor_si256( b, d ); \
+  b = _mm256_xor_si256( b, t ); \
+  a = c; \
+  c = b; \
+  b = d; \
+  d = mm256_not( t ); \
+} while (0)
+
+#define L( a, b, c, d ) \
+do { \
+   a = mm256_rotl_32( a, 13 ); \
+   c = mm256_rotl_32( c,  3 ); \
+   b = _mm256_xor_si256( b, _mm256_xor_si256( a, c ) ); \
+   d = _mm256_xor_si256( d, _mm256_xor_si256( c, \
+                                              _mm256_slli_epi32( a, 3 ) ) ); \
+   b = mm256_rotl_32( b, 1 ); \
+   d = mm256_rotl_32( d, 7 ); \
+   a = _mm256_xor_si256( a, _mm256_xor_si256( b, d ) ); \
+   c = _mm256_xor_si256( c, _mm256_xor_si256( d, \
+                                              _mm256_slli_epi32( b, 7 ) ) ); \
+   a = mm256_rotl_32( a,  5 ); \
+   c = mm256_rotl_32( c, 22 ); \
+} while (0)
+
+#define DECL_STATE_BIG \
+   __m256i c0, c1, c2, c3, c4, c5, c6, c7; \
+
+#define READ_STATE_BIG(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_BIG(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 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
+
+#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] ) ); \
+\
+  SBOX( s0, s4, s8, sC ); \
+  SBOX( s1, s5, s9, sD ); \
+  SBOX( s2, s6, sA, sE ); \
+  SBOX( s3, s7, sB, sF ); \
+\
+  t1 = _mm256_blend_epi32( _mm256_bsrli_epi128( s4, 4 ), \
+                           _mm256_bslli_epi128( s5, 4 ), 0xAA ); \
+  t3 = _mm256_blend_epi32( _mm256_bsrli_epi128( sD, 4 ), \
+                           _mm256_bslli_epi128( sE, 4 ), 0xAA ); \
+  L( s0, t1, s9, t3 ); \
+  s4 = _mm256_blend_epi32( s4, _mm256_bslli_epi128( t1, 4 ), 0xAA );\
+  s5 = _mm256_blend_epi32( s5, _mm256_bsrli_epi128( t1, 4 ), 0x55 );\
+  sD = _mm256_blend_epi32( sD, _mm256_bslli_epi128( t3, 4 ), 0xAA );\
+  sE = _mm256_blend_epi32( sE, _mm256_bsrli_epi128( t3, 4 ), 0x55 );\
+\
+  t1 = _mm256_blend_epi32( _mm256_bsrli_epi128( s5, 4 ), \
+                           _mm256_bslli_epi128( s6, 4 ), 0xAA ); \
+  t3 = _mm256_blend_epi32( _mm256_bsrli_epi128( sE, 4 ), \
+                           _mm256_bslli_epi128( sF, 4 ), 0xAA ); \
+  L( s1, t1, sA, t3 ); \
+  s5 = _mm256_blend_epi32( s5, _mm256_bslli_epi128( t1, 4 ), 0xAA );\
+  s6 = _mm256_blend_epi32( s6, _mm256_bsrli_epi128( t1, 4 ), 0x55 );\
+  sE = _mm256_blend_epi32( sE, _mm256_bslli_epi128( t3, 4 ), 0xAA );\
+  sF = _mm256_blend_epi32( sF, _mm256_bsrli_epi128( t3, 4 ), 0x55 );\
+\
+  t1 = _mm256_blend_epi32( _mm256_bsrli_epi128( s6, 4 ), \
+                           _mm256_bslli_epi128( s7, 4 ), 0xAA ); \
+  t3 = _mm256_blend_epi32( _mm256_bsrli_epi128( sF, 4 ), \
+                           _mm256_bslli_epi128( sC, 4 ), 0xAA ); \
+  L( s2, t1, sB, t3 ); \
+  s6 = _mm256_blend_epi32( s6, _mm256_bslli_epi128( t1, 4 ), 0xAA );\
+  s7 = _mm256_blend_epi32( s7, _mm256_bsrli_epi128( t1, 4 ), 0x55 );\
+  sF = _mm256_blend_epi32( sF, _mm256_bslli_epi128( t3, 4 ), 0xAA );\
+  sC = _mm256_blend_epi32( sC, _mm256_bsrli_epi128( t3, 4 ), 0x55 );\
+\
+  t1 = _mm256_blend_epi32( _mm256_bsrli_epi128( s7, 4 ), \
+                           _mm256_bslli_epi128( s4, 4 ), 0xAA ); \
+  t3 = _mm256_blend_epi32( _mm256_bsrli_epi128( sC, 4 ), \
+                           _mm256_bslli_epi128( sD, 4 ), 0xAA ); \
+  L( s3, t1, s8, t3 ); \
+  s7 = _mm256_blend_epi32( s7, _mm256_bslli_epi128( t1, 4 ), 0xAA );\
+  s4 = _mm256_blend_epi32( s4, _mm256_bsrli_epi128( t1, 4 ), 0x55 );\
+  sC = _mm256_blend_epi32( sC, _mm256_bslli_epi128( t3, 4 ), 0xAA );\
+  sD = _mm256_blend_epi32( sD, _mm256_bsrli_epi128( t3, 4 ), 0x55 );\
+\
+  t0 = _mm256_blend_epi32( s0, _mm256_bslli_epi128( s8, 4 ), 0xAA ); \
+  t1 = _mm256_blend_epi32( s1, s9, 0xAA ); \
+  t2 = _mm256_blend_epi32( _mm256_bsrli_epi128( s2, 4 ), sA, 0xAA ); \
+  t3 = _mm256_blend_epi32( _mm256_bsrli_epi128( s3, 4 ), \
+                           _mm256_bslli_epi128( sB, 4 ), 0xAA ); \
+  L( t0, t1, t2, t3 ); \
+  s0 = _mm256_blend_epi32( s0, t0, 0x55 ); \
+  s8 = _mm256_blend_epi32( s8, _mm256_bsrli_epi128( t0, 4 ), 0x55 ); \
+  s1 = _mm256_blend_epi32( s1, t1, 0x55 ); \
+  s9 = _mm256_blend_epi32( s9, t1, 0xAA ); \
+  s2 = _mm256_blend_epi32( s2, _mm256_bslli_epi128( t2, 4 ), 0xAA ); \
+  sA = _mm256_blend_epi32( sA, t2, 0xAA ); \
+  s3 = _mm256_blend_epi32( s3, _mm256_bslli_epi128( t3, 4 ), 0xAA ); \
+  sB = _mm256_blend_epi32( sB, _mm256_bsrli_epi128( t3, 4 ), 0x55 ); \
+\
+  t0 = _mm256_blend_epi32( _mm256_bsrli_epi128( s4, 4 ), sC, 0xAA ); \
+  t1 = _mm256_blend_epi32( _mm256_bsrli_epi128( s5, 4 ), \
+                           _mm256_bslli_epi128( sD, 4 ), 0xAA ); \
+  t2 = _mm256_blend_epi32( s6, _mm256_bslli_epi128( sE, 4 ), 0xAA ); \
+  t3 = _mm256_blend_epi32( s7, sF, 0xAA ); \
+  L( t0, t1, t2, t3 ); \
+  s4 = _mm256_blend_epi32( s4, _mm256_bslli_epi128( t0, 4 ), 0xAA ); \
+  sC = _mm256_blend_epi32( sC, t0, 0xAA ); \
+  s5 = _mm256_blend_epi32( s5, _mm256_bslli_epi128( t1, 4 ), 0xAA ); \
+  sD = _mm256_blend_epi32( sD, _mm256_bsrli_epi128( t1, 4 ), 0x55 ); \
+  s6 = _mm256_blend_epi32( s6, t2, 0x55 ); \
+  sE = _mm256_blend_epi32( sE, _mm256_bsrli_epi128( t2, 4 ), 0x55 ); \
+  s7 = _mm256_blend_epi32( s7, t3, 0x55 ); \
+  sF = _mm256_blend_epi32( sF, t3, 0xAA ); \
+} while (0)
+
+#define P_BIG \
+do { \
+   ROUND_BIG(0, alpha_n); \
+   ROUND_BIG(1, alpha_n); \
+   ROUND_BIG(2, alpha_n); \
+   ROUND_BIG(3, alpha_n); \
+   ROUND_BIG(4, alpha_n); \
+   ROUND_BIG(5, alpha_n); \
+} while (0)
+
+#define PF_BIG \
+do { \
+   ROUND_BIG( 0, alpha_f); \
+   ROUND_BIG( 1, alpha_f); \
+   ROUND_BIG( 2, alpha_f); \
+   ROUND_BIG( 3, alpha_f); \
+   ROUND_BIG( 4, alpha_f); \
+   ROUND_BIG( 5, alpha_f); \
+   ROUND_BIG( 6, alpha_f); \
+   ROUND_BIG( 7, alpha_f); \
+   ROUND_BIG( 8, alpha_f); \
+   ROUND_BIG( 9, alpha_f); \
+   ROUND_BIG(10, alpha_f); \
+   ROUND_BIG(11, alpha_f); \
+} while (0)
+
+#define T_BIG \
+do { /* order is important */ \
+   c7 = sc->h[ 0x7 ] = _mm256_xor_si256( sc->h[ 0x7 ], sB ); \
+   c6 = sc->h[ 0x6 ] = _mm256_xor_si256( sc->h[ 0x6 ], sA ); \
+   c5 = sc->h[ 0x5 ] = _mm256_xor_si256( sc->h[ 0x5 ], s9 ); \
+   c4 = sc->h[ 0x4 ] = _mm256_xor_si256( sc->h[ 0x4 ], s8 ); \
+   c3 = sc->h[ 0x3 ] = _mm256_xor_si256( sc->h[ 0x3 ], s3 ); \
+   c2 = sc->h[ 0x2 ] = _mm256_xor_si256( sc->h[ 0x2 ], s2 ); \
+   c1 = sc->h[ 0x1 ] = _mm256_xor_si256( sc->h[ 0x1 ], s1 ); \
+   c0 = sc->h[ 0x0 ] = _mm256_xor_si256( sc->h[ 0x0 ], s0 ); \
+} while (0)
+
+void hamsi_big( hamsi_4way_big_context *sc, __m256i *buf, size_t num )
+{
+   DECL_STATE_BIG
+   sph_u32 tmp;
+
+   tmp = SPH_T32( (sph_u32)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_BIG( sc );
+   while ( num-- > 0 )
+   {
+      __m256i m0, m1, m2, m3, m4, m5, m6, m7;
+
+      INPUT_BIG;
+      P_BIG;
+      T_BIG;
+      buf++;
+   }
+   WRITE_STATE_BIG( sc );
+}
+
+void hamsi_big_final( hamsi_4way_big_context *sc, __m256i *buf )
+{
+   __m256i m0, m1, m2, m3, m4, m5, m6, m7;
+   DECL_STATE_BIG
+   READ_STATE_BIG( sc );
+   INPUT_BIG;
+   PF_BIG;
+   T_BIG;
+   WRITE_STATE_BIG( sc );
+}
+
+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] );
+   }
+}
+
+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;
+   memcpy_256( sc->buf, vdata, len>>3 );
+   sc->partial_len = len;
+}
+
+void hamsi512_4way_close( hamsi_4way_big_context *sc, void *dst )
+{
+   __m256i *out = (__m256i*)dst;
+   __m256i pad[1];
+   size_t u;
+   int ch, cl;
+
+   sph_enc32be( &ch, sc->count_high );
+   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 );
+   hamsi_big( sc, sc->buf, 1 );
+   hamsi_big_final( sc, pad );
+   for ( u = 0; u < 8; u ++ )
+      out[u] = mm256_bswap_32( sc->h[u] );
+}
+
+#ifdef __cplusplus
+}
+#endif
+#endif
--- a/algo/hamsi/hamsi-hash-4way.h
+++ b/algo/hamsi/hamsi-hash-4way.h
@@ -0,0 +1,72 @@
+/* $Id: sph_hamsi.h 216 2010-06-08 09:46:57Z tp $ */
+/**
+ * Hamsi interface. This code implements Hamsi with the recommended
+ * parameters for SHA-3, with outputs of 224, 256, 384 and 512 bits.
+ *
+ * ==========================(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)=============================
+ *
+ * @file     sph_hamsi.h
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#ifndef HAMSI_4WAY_H__
+#define HAMSI_4WAY_H__
+
+#include <stddef.h>
+#include "algo/sha/sph_types.h"
+
+#if defined (__AVX__)
+
+#include "avxdefs.h"
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#define SPH_SIZE_hamsi512   512
+
+// Partial is only scalar but needs pointer ref for hamsi-helper
+// deprecate partial_len
+typedef struct {
+   __m256i h[8];
+   __m256i buf[1];
+   size_t partial_len;
+   sph_u32 count_high, count_low;
+} hamsi_4way_big_context;
+
+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_close( hamsi512_4way_context *sc, void *dst );
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
+
+#endif
--- a/algo/haval/haval-4way-helper.c
+++ b/algo/haval/haval-4way-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), _4way)
+( haval_4way_context *sc, const void *data, size_t len )
+{
+   __m128i *vdata = (__m128i*)data;
+   unsigned current;
+
+   current = (unsigned)sc->count_low & 127U;
+   while ( len > 0 )
+   {
+      unsigned clen;
+      sph_u32 clow, clow2;
+
+      clen = 128U - current;
+      if ( clen > len )
+         clen = len;
+      memcpy_128( sc->buf + (current>>2), vdata, clen>>2 );
+      vdata += clen>>2;
+      current += clen;
+      len -= clen;
+      if ( current == 128U )
+      {
+         DSTATE;
+         IN_PREPARE(sc->buf);
+         RSTATE;
+         SPH_XCAT(CORE, PASSES)(INW);
+         WSTATE;
+         current = 0;
+      }
+      clow = sc->count_low;
+      clow2 = SPH_T32(clow + clen);
+      sc->count_low = clow2;
+      if ( clow2 < clow )
+         sc->count_high ++;
+   }
+}
+
+static void
+SPH_XCAT(SPH_XCAT(haval, PASSES), _4way_close)( haval_4way_context *sc,
+                                                void *dst)
+{
+   unsigned current;
+   DSTATE;
+
+   current = (unsigned)sc->count_low & 127UL;
+
+   sc->buf[ current>>2 ] = m128_one_32;
+   current += 4;   
+   RSTATE;
+   if ( current > 116UL )
+   {
+      memset_zero_128( sc->buf + ( current>>2 ), (128UL-current) >> 2 );
+      do
+      {
+         IN_PREPARE(sc->buf);
+         SPH_XCAT(CORE, PASSES)(INW);
+      } while (0);
+      current = 0;
+   }
+
+   uint32_t t1, t2;
+   memset_zero_128( sc->buf + ( current>>2 ), (116UL-current) >> 2 );
+   t1 = 0x01 | (PASSES << 3);
+   t2 = sc->olen << 3;
+   sc->buf[ 116>>2 ] = _mm_set1_epi32( ( t1 << 16 ) | ( t2 << 24 ) );
+   sc->buf[ 120>>2 ] = _mm_set1_epi32( sc->count_low << 3 );
+   sc->buf[ 124>>2 ] = _mm_set1_epi32( (sc->count_high << 3)
+                                     | (sc->count_low >> 29) );
+   do
+   {
+      IN_PREPARE(sc->buf);
+      SPH_XCAT(CORE, PASSES)(INW);
+   } while (0);
+   WSTATE;
+   haval_4way_out( sc, dst );
+}
--- a/algo/haval/haval-hash-4way.c
+++ b/algo/haval/haval-hash-4way.c
@@ -0,0 +1,522 @@
+/* $Id: haval.c 227 2010-06-16 17:28:38Z tp $ */
+/*
+ * HAVAL implementation.
+ *
+ * The HAVAL reference paper is of questionable clarity with regards to
+ * some details such as endianness of bits within a byte, bytes within
+ * a 32-bit word, or the actual ordering of words within a stream of
+ * words. This implementation has been made compatible with the reference
+ * implementation available on: http://labs.calyptix.com/haval.php
+ *
+ * ==========================(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>
+ */
+
+#include <stddef.h>
+#include <string.h>
+#include "haval-hash-4way.h"
+
+#if defined (__AVX__)
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+//#if SPH_SMALL_FOOTPRINT && !defined SPH_SMALL_FOOTPRINT_HAVAL
+#define SPH_SMALL_FOOTPRINT_HAVAL   1
+//#endif
+
+#define F1(x6, x5, x4, x3, x2, x1, x0) \
+   _mm_xor_si128( x0, \
+       _mm_xor_si128( _mm_and_si128(_mm_xor_si128( x0, x4 ), x1 ), \
+                      _mm_xor_si128( _mm_and_si128( x2, x5 ), \
+                                     _mm_and_si128( x3, x6 ) ) ) ) \
+
+#define F2(x6, x5, x4, x3, x2, x1, x0) \
+   _mm_xor_si128( \
+      _mm_and_si128( x2, \
+         _mm_xor_si128( _mm_andnot_si128( x3, x1 ), \
+                        _mm_xor_si128( _mm_and_si128( x4, x5 ), \
+                                       _mm_xor_si128( x6, x0 ) ) ) ), \
+         _mm_xor_si128( \
+             _mm_and_si128( x4, _mm_xor_si128( x1, x5 ) ), \
+             _mm_xor_si128( _mm_and_si128( x3, x5 ), x0 ) ) ) \
+
+#define F3(x6, x5, x4, x3, x2, x1, x0) \
+  _mm_xor_si128( \
+    _mm_and_si128( x3, \
+      _mm_xor_si128( _mm_and_si128( x1, x2 ), \
+                     _mm_xor_si128( x6, x0 ) ) ), \
+      _mm_xor_si128( _mm_xor_si128(_mm_and_si128( x1, x4 ), \
+                                   _mm_and_si128( x2, x5 ) ), x0 ) )
+
+#define F4(x6, x5, x4, x3, x2, x1, x0) \
+  _mm_xor_si128( \
+     _mm_xor_si128( \
+        _mm_and_si128( x3, \
+           _mm_xor_si128( _mm_xor_si128( _mm_and_si128( x1, x2 ), \
+                                         _mm_or_si128( x4, x6 ) ), x5 ) ), \
+        _mm_and_si128( x4, \
+           _mm_xor_si128( _mm_xor_si128( _mm_and_si128( mm_not(x2), x5 ), \
+                          _mm_xor_si128( x1, x6 ) ), x0 ) ) ), \
+     _mm_xor_si128( _mm_and_si128( x2, x6 ), x0 ) )
+
+
+#define F5(x6, x5, x4, x3, x2, x1, x0) \
+   _mm_xor_si128( \
+       _mm_and_si128( x0, \
+            mm_not( _mm_xor_si128( \
+                    _mm_and_si128( _mm_and_si128( x1, x2 ), x3 ), x5 ) ) ), \
+      _mm_xor_si128( _mm_xor_si128( _mm_and_si128( x1, x4 ), \
+                                    _mm_and_si128( x2, x5 ) ), \
+                                    _mm_and_si128( x3, x6 ) ) )
+
+/*
+ * The macros below integrate the phi() permutations, depending on the
+ * pass and the total number of passes.
+ */
+
+#define FP3_1(x6, x5, x4, x3, x2, x1, x0) \
+	F1(x1, x0, x3, x5, x6, x2, x4)
+#define FP3_2(x6, x5, x4, x3, x2, x1, x0) \
+	F2(x4, x2, x1, x0, x5, x3, x6)
+#define FP3_3(x6, x5, x4, x3, x2, x1, x0) \
+	F3(x6, x1, x2, x3, x4, x5, x0)
+
+#define FP4_1(x6, x5, x4, x3, x2, x1, x0) \
+	F1(x2, x6, x1, x4, x5, x3, x0)
+#define FP4_2(x6, x5, x4, x3, x2, x1, x0) \
+	F2(x3, x5, x2, x0, x1, x6, x4)
+#define FP4_3(x6, x5, x4, x3, x2, x1, x0) \
+	F3(x1, x4, x3, x6, x0, x2, x5)
+#define FP4_4(x6, x5, x4, x3, x2, x1, x0) \
+	F4(x6, x4, x0, x5, x2, x1, x3)
+
+#define FP5_1(x6, x5, x4, x3, x2, x1, x0) \
+	F1(x3, x4, x1, x0, x5, x2, x6)
+#define FP5_2(x6, x5, x4, x3, x2, x1, x0) \
+	F2(x6, x2, x1, x0, x3, x4, x5)
+#define FP5_3(x6, x5, x4, x3, x2, x1, x0) \
+	F3(x2, x6, x0, x4, x3, x1, x5)
+#define FP5_4(x6, x5, x4, x3, x2, x1, x0) \
+	F4(x1, x5, x3, x2, x0, x4, x6)
+#define FP5_5(x6, x5, x4, x3, x2, x1, x0) \
+	F5(x2, x5, x0, x6, x4, x3, x1)
+
+/*
+ * One step, for "n" passes, pass number "p" (1 <= p <= n), using
+ * input word number "w" and step constant "c".
+ */
+#define STEP(n, p, x7, x6, x5, x4, x3, x2, x1, x0, w, c) \
+do { \
+   __m128i t = FP ## n ## _ ## p(x6, x5, x4, x3, x2, x1, x0); \
+   x7 = _mm_add_epi32( _mm_add_epi32( mm_rotr_32( t, 7 ), \
+                                      mm_rotr_32( x7, 11 ) ), \
+                       _mm_add_epi32( w, _mm_set1_epi32( c ) ) ); \
+} while (0)
+
+/*
+ * PASSy(n, in) computes pass number "y", for a total of "n", using the
+ * one-argument macro "in" to access input words. Current state is assumed
+ * to be held in variables "s0" to "s7".
+ */
+
+//#if SPH_SMALL_FOOTPRINT_HAVAL
+
+#define PASS1(n, in)   do { \
+		unsigned pass_count; \
+		for (pass_count = 0; pass_count < 32; pass_count += 8) { \
+			STEP(n, 1, s7, s6, s5, s4, s3, s2, s1, s0, \
+				in(pass_count + 0), SPH_C32(0x00000000)); \
+			STEP(n, 1, s6, s5, s4, s3, s2, s1, s0, s7, \
+				in(pass_count + 1), SPH_C32(0x00000000)); \
+			STEP(n, 1, s5, s4, s3, s2, s1, s0, s7, s6, \
+				in(pass_count + 2), SPH_C32(0x00000000)); \
+			STEP(n, 1, s4, s3, s2, s1, s0, s7, s6, s5, \
+				in(pass_count + 3), SPH_C32(0x00000000)); \
+			STEP(n, 1, s3, s2, s1, s0, s7, s6, s5, s4, \
+				in(pass_count + 4), SPH_C32(0x00000000)); \
+			STEP(n, 1, s2, s1, s0, s7, s6, s5, s4, s3, \
+				in(pass_count + 5), SPH_C32(0x00000000)); \
+			STEP(n, 1, s1, s0, s7, s6, s5, s4, s3, s2, \
+				in(pass_count + 6), SPH_C32(0x00000000)); \
+			STEP(n, 1, s0, s7, s6, s5, s4, s3, s2, s1, \
+				in(pass_count + 7), SPH_C32(0x00000000)); \
+   		} \
+	} while (0)
+
+#define PASSG(p, n, in)   do { \
+		unsigned pass_count; \
+		for (pass_count = 0; pass_count < 32; pass_count += 8) { \
+			STEP(n, p, s7, s6, s5, s4, s3, s2, s1, s0, \
+				in(MP ## p[pass_count + 0]), \
+				RK ## p[pass_count + 0]); \
+			STEP(n, p, s6, s5, s4, s3, s2, s1, s0, s7, \
+				in(MP ## p[pass_count + 1]), \
+				RK ## p[pass_count + 1]); \
+			STEP(n, p, s5, s4, s3, s2, s1, s0, s7, s6, \
+				in(MP ## p[pass_count + 2]), \
+				RK ## p[pass_count + 2]); \
+			STEP(n, p, s4, s3, s2, s1, s0, s7, s6, s5, \
+				in(MP ## p[pass_count + 3]), \
+				RK ## p[pass_count + 3]); \
+			STEP(n, p, s3, s2, s1, s0, s7, s6, s5, s4, \
+				in(MP ## p[pass_count + 4]), \
+				RK ## p[pass_count + 4]); \
+			STEP(n, p, s2, s1, s0, s7, s6, s5, s4, s3, \
+				in(MP ## p[pass_count + 5]), \
+				RK ## p[pass_count + 5]); \
+			STEP(n, p, s1, s0, s7, s6, s5, s4, s3, s2, \
+				in(MP ## p[pass_count + 6]), \
+				RK ## p[pass_count + 6]); \
+			STEP(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(n, in)    PASSG(2, n, in)
+#define PASS3(n, in)    PASSG(3, n, in)
+#define PASS4(n, in)    PASSG(4, n, in)
+#define PASS5(n, in)    PASSG(5, n, in)
+
+static const unsigned MP2[32] = {
+	 5, 14, 26, 18, 11, 28,  7, 16,
+	 0, 23, 20, 22,  1, 10,  4,  8,
+	30,  3, 21,  9, 17, 24, 29,  6,
+	19, 12, 15, 13,  2, 25, 31, 27
+};
+
+static const unsigned MP3[32] = {
+	19,  9,  4, 20, 28, 17,  8, 22,
+	29, 14, 25, 12, 24, 30, 16, 26,
+	31, 15,  7,  3,  1,  0, 18, 27,
+	13,  6, 21, 10, 23, 11,  5,  2
+};
+
+static const unsigned MP4[32] = {
+	24,  4,  0, 14,  2,  7, 28, 23,
+	26,  6, 30, 20, 18, 25, 19,  3,
+	22, 11, 31, 21,  8, 27, 12,  9,
+	 1, 29,  5, 15, 17, 10, 16, 13
+};
+
+static const unsigned MP5[32] = {
+	27,  3, 21, 26, 17, 11, 20, 29,
+	19,  0, 12,  7, 13,  8, 31, 10,
+	 5,  9, 14, 30, 18,  6, 28, 24,
+	 2, 23, 16, 22,  4,  1, 25, 15
+};
+
+static const sph_u32 RK2[32] = {
+	SPH_C32(0x452821E6), SPH_C32(0x38D01377),
+	SPH_C32(0xBE5466CF), SPH_C32(0x34E90C6C),
+	SPH_C32(0xC0AC29B7), SPH_C32(0xC97C50DD),
+	SPH_C32(0x3F84D5B5), SPH_C32(0xB5470917),
+	SPH_C32(0x9216D5D9), SPH_C32(0x8979FB1B),
+	SPH_C32(0xD1310BA6), SPH_C32(0x98DFB5AC),
+	SPH_C32(0x2FFD72DB), SPH_C32(0xD01ADFB7),
+	SPH_C32(0xB8E1AFED), SPH_C32(0x6A267E96),
+	SPH_C32(0xBA7C9045), SPH_C32(0xF12C7F99),
+	SPH_C32(0x24A19947), SPH_C32(0xB3916CF7),
+	SPH_C32(0x0801F2E2), SPH_C32(0x858EFC16),
+	SPH_C32(0x636920D8), SPH_C32(0x71574E69),
+	SPH_C32(0xA458FEA3), SPH_C32(0xF4933D7E),
+	SPH_C32(0x0D95748F), SPH_C32(0x728EB658),
+	SPH_C32(0x718BCD58), SPH_C32(0x82154AEE),
+	SPH_C32(0x7B54A41D), SPH_C32(0xC25A59B5)
+};
+
+static const sph_u32 RK3[32] = {
+	SPH_C32(0x9C30D539), SPH_C32(0x2AF26013),
+	SPH_C32(0xC5D1B023), SPH_C32(0x286085F0),
+	SPH_C32(0xCA417918), SPH_C32(0xB8DB38EF),
+	SPH_C32(0x8E79DCB0), SPH_C32(0x603A180E),
+	SPH_C32(0x6C9E0E8B), SPH_C32(0xB01E8A3E),
+	SPH_C32(0xD71577C1), SPH_C32(0xBD314B27),
+	SPH_C32(0x78AF2FDA), SPH_C32(0x55605C60),
+	SPH_C32(0xE65525F3), SPH_C32(0xAA55AB94),
+	SPH_C32(0x57489862), SPH_C32(0x63E81440),
+	SPH_C32(0x55CA396A), SPH_C32(0x2AAB10B6),
+	SPH_C32(0xB4CC5C34), SPH_C32(0x1141E8CE),
+	SPH_C32(0xA15486AF), SPH_C32(0x7C72E993),
+	SPH_C32(0xB3EE1411), SPH_C32(0x636FBC2A),
+	SPH_C32(0x2BA9C55D), SPH_C32(0x741831F6),
+	SPH_C32(0xCE5C3E16), SPH_C32(0x9B87931E),
+	SPH_C32(0xAFD6BA33), SPH_C32(0x6C24CF5C)
+};
+
+static const sph_u32 RK4[32] = {
+	SPH_C32(0x7A325381), SPH_C32(0x28958677),
+	SPH_C32(0x3B8F4898), SPH_C32(0x6B4BB9AF),
+	SPH_C32(0xC4BFE81B), SPH_C32(0x66282193),
+	SPH_C32(0x61D809CC), SPH_C32(0xFB21A991),
+	SPH_C32(0x487CAC60), SPH_C32(0x5DEC8032),
+	SPH_C32(0xEF845D5D), SPH_C32(0xE98575B1),
+	SPH_C32(0xDC262302), SPH_C32(0xEB651B88),
+	SPH_C32(0x23893E81), SPH_C32(0xD396ACC5),
+	SPH_C32(0x0F6D6FF3), SPH_C32(0x83F44239),
+	SPH_C32(0x2E0B4482), SPH_C32(0xA4842004),
+	SPH_C32(0x69C8F04A), SPH_C32(0x9E1F9B5E),
+	SPH_C32(0x21C66842), SPH_C32(0xF6E96C9A),
+	SPH_C32(0x670C9C61), SPH_C32(0xABD388F0),
+	SPH_C32(0x6A51A0D2), SPH_C32(0xD8542F68),
+	SPH_C32(0x960FA728), SPH_C32(0xAB5133A3),
+	SPH_C32(0x6EEF0B6C), SPH_C32(0x137A3BE4)
+};
+
+static const sph_u32 RK5[32] = {
+	SPH_C32(0xBA3BF050), SPH_C32(0x7EFB2A98),
+	SPH_C32(0xA1F1651D), SPH_C32(0x39AF0176),
+	SPH_C32(0x66CA593E), SPH_C32(0x82430E88),
+	SPH_C32(0x8CEE8619), SPH_C32(0x456F9FB4),
+	SPH_C32(0x7D84A5C3), SPH_C32(0x3B8B5EBE),
+	SPH_C32(0xE06F75D8), SPH_C32(0x85C12073),
+	SPH_C32(0x401A449F), SPH_C32(0x56C16AA6),
+	SPH_C32(0x4ED3AA62), SPH_C32(0x363F7706),
+	SPH_C32(0x1BFEDF72), SPH_C32(0x429B023D),
+	SPH_C32(0x37D0D724), SPH_C32(0xD00A1248),
+	SPH_C32(0xDB0FEAD3), SPH_C32(0x49F1C09B),
+	SPH_C32(0x075372C9), SPH_C32(0x80991B7B),
+	SPH_C32(0x25D479D8), SPH_C32(0xF6E8DEF7),
+	SPH_C32(0xE3FE501A), SPH_C32(0xB6794C3B),
+	SPH_C32(0x976CE0BD), SPH_C32(0x04C006BA),
+	SPH_C32(0xC1A94FB6), SPH_C32(0x409F60C4)
+};
+
+#define SAVE_STATE \
+   __m128i 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 \
+do { \
+   s0 = _mm_add_epi32( s0, u0 ); \
+   s1 = _mm_add_epi32( s1, u1 ); \
+   s2 = _mm_add_epi32( s2, u2 ); \
+   s3 = _mm_add_epi32( s3, u3 ); \
+   s4 = _mm_add_epi32( s4, u4 ); \
+   s5 = _mm_add_epi32( s5, u5 ); \
+   s6 = _mm_add_epi32( s6, u6 ); \
+   s7 = _mm_add_epi32( s7, u7 ); \
+} while (0)
+
+/*
+ * COREn(in) performs the core HAVAL computation for "n" passes, using
+ * the one-argument macro "in" to access the input words. Running state
+ * is held in variable "s0" to "s7".
+ */
+/*
+#define CORE3(in)  do { \
+		SAVE_STATE; \
+		PASS1(3, in); \
+		PASS2(3, in); \
+		PASS3(3, in); \
+		UPDATE_STATE; \
+	} while (0)
+
+#define CORE4(in)  do { \
+		SAVE_STATE; \
+		PASS1(4, in); \
+		PASS2(4, in); \
+		PASS3(4, in); \
+		PASS4(4, in); \
+		UPDATE_STATE; \
+	} while (0)
+*/
+#define CORE5(in)  do { \
+		SAVE_STATE; \
+		PASS1(5, in); \
+		PASS2(5, in); \
+		PASS3(5, in); \
+		PASS4(5, in); \
+		PASS5(5, in); \
+		UPDATE_STATE; \
+	} while (0)
+
+/*
+ * DSTATE declares the state variables "s0" to "s7".
+ */
+#define DSTATE   __m128i s0, s1, s2, s3, s4, s5, s6, s7
+
+/*
+ * RSTATE fills the state variables from the context "sc".
+ */
+#define RSTATE \
+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)
+
+/*
+ * WSTATE updates the context "sc" from the state variables.
+ */
+#define WSTATE \
+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)
+
+/*
+ * Initialize a context. "olen" is the output length, in 32-bit words
+ * (between 4 and 8, inclusive). "passes" is the number of passes
+ * (3, 4 or 5).
+ */
+static void
+haval_4way_init( haval_4way_context *sc, unsigned olen, unsigned passes )
+{
+   sc->s0 = _mm_set1_epi32( 0x243F6A88UL );
+   sc->s1 = _mm_set1_epi32( 0x85A308D3UL );
+   sc->s2 = _mm_set1_epi32( 0x13198A2EUL );
+   sc->s3 = _mm_set1_epi32( 0x03707344UL );
+   sc->s4 = _mm_set1_epi32( 0xA4093822UL );
+   sc->s5 = _mm_set1_epi32( 0x299F31D0UL );
+   sc->s6 = _mm_set1_epi32( 0x082EFA98UL );
+   sc->s7 = _mm_set1_epi32( 0xEC4E6C89UL );
+   sc->olen = olen;
+   sc->passes = passes;
+   sc->count_high = 0;
+   sc->count_low = 0;
+	
+}
+
+#define IN_PREPARE(indata) const __m128i *const load_ptr = (indata)
+
+#define INW(i)   load_ptr[ i ] 
+
+/*
+ * Write out HAVAL output. The output length is tailored to the requested
+ * length.
+ */
+static void
+haval_4way_out( haval_4way_context *sc, void *dst )
+{
+   __m128i *buf = (__m128i*)dst;
+   DSTATE;
+   RSTATE;
+
+   buf[0] = s0;
+   buf[1] = s1;
+   buf[2] = s2;
+   buf[3] = s3;
+   buf[4] = s4;
+   buf[5] = s5;
+   buf[6] = s6;
+   buf[7] = s7;
+}
+
+/*
+ * The main core functions inline the code with the COREx() macros. We
+ * use a helper file, included three times, which avoids code copying.
+ */
+/*
+#undef PASSES
+#define PASSES   3
+#include "haval-helper.c"
+
+#undef PASSES
+#define PASSES   4
+#include "haval-helper.c"
+*/
+
+#undef PASSES
+#define PASSES   5
+#include "haval-4way-helper.c"
+
+/* ====================================================================== */
+
+#define API(xxx, y) \
+void \
+haval ## xxx ## _ ## y ## _4way_init(void *cc) \
+{ \
+	haval_4way_init(cc, xxx >> 5, y); \
+} \
+ \
+void \
+haval ## xxx ## _ ## y ## _4way (void *cc, const void *data, size_t len) \
+{ \
+	haval ## y ## _4way(cc, data, len); \
+} \
+ \
+void \
+haval ## xxx ## _ ## y ## _4way_close(void *cc, void *dst) \
+{ \
+	haval ## y ## _4way_close(cc, dst); \
+} \
+
+API(256, 5)
+
+#define RVAL \
+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 \
+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(i)   msg[i]
+
+#ifdef __cplusplus
+}
+#endif	
+#endif
--- a/algo/haval/haval-hash-4way.h
+++ b/algo/haval/haval-hash-4way.h
@@ -0,0 +1,95 @@
+/* $Id: sph_haval.h 218 2010-06-08 17:06:34Z tp $ */
+/**
+ * HAVAL interface.
+ *
+ * HAVAL is actually a family of 15 hash functions, depending on whether
+ * the internal computation uses 3, 4 or 5 passes, and on the output
+ * length, which is 128, 160, 192, 224 or 256 bits. This implementation
+ * provides interface functions for all 15, which internally map to
+ * three cores (depending on the number of passes). Note that output
+ * lengths other than 256 bits are not obtained by a simple truncation
+ * of a longer result; the requested length is encoded within the
+ * padding data.
+ *
+ * HAVAL was published in: Yuliang Zheng, Josef Pieprzyk and Jennifer
+ * Seberry: "HAVAL -- a one-way hashing algorithm with variable length
+ * of output", Advances in Cryptology -- AUSCRYPT'92, Lecture Notes in
+ * Computer Science, Vol.718, pp.83-104, Springer-Verlag, 1993.
+ *
+ * This paper, and a reference implementation, are available on the
+ * Calyptix web site: http://labs.calyptix.com/haval.php
+ *
+ * The HAVAL reference paper is quite unclear on the data encoding
+ * details, i.e. endianness (both byte order within a 32-bit word, and
+ * word order within a message block). This implementation has been
+ * made compatible with the reference implementation referenced above.
+ *
+ * @warning   A collision for HAVAL-128/3 (HAVAL with three passes and
+ * 128-bit output) has been published; this function is thus considered
+ * as cryptographically broken. The status for other variants is unclear;
+ * use only with care.
+ *
+ * ==========================(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)=============================
+ *
+ * @file     sph_haval.h
+ * @author   Thomas Pornin <thomas.pornin@cryptolog.com>
+ */
+
+#ifndef HAVAL_HASH_4WAY_H__
+#define HAVAL_HASH_4WAY_H__
+
+#if defined(__AVX__)
+
+#ifdef __cplusplus
+extern "C"{
+#endif
+
+#include <stddef.h>
+#include "algo/sha/sph_types.h"
+#include "avxdefs.h"
+
+#define SPH_SIZE_haval256_5   256
+
+typedef struct {
+   __m128i buf[32];
+   __m128i s0, s1, s2, s3, s4, s5, s6, s7;
+   unsigned olen, passes;
+   sph_u32 count_high, count_low;
+} haval_4way_context;
+
+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_close( void *cc, void *dst );
+
+#ifdef __cplusplus
+}
+#endif
+#endif
+#endif
--- a/algo/heavy/bastion.c
+++ b/algo/heavy/bastion.c
@@ -15,7 +15,7 @@
 #include "algo/shabal/sph_shabal.h"
 #include "algo/echo/sph_echo.h"
 #include "algo/hamsi/sph_hamsi.h"
-#include "algo/luffa/sse2/luffa_for_sse2.h"
+#include "algo/luffa/luffa_for_sse2.h"
 #include "algo/skein/sse2/skein.c"

 #ifndef NO_AES_NI
--- a/algo/hodl/hodl-gate.c
+++ b/algo/hodl/hodl-gate.c
@@ -42,15 +42,83 @@ void hodl_le_build_stratum_request( char* req, struct work* work,
   free( xnonce2str );
 }

-void hodl_build_extraheader( struct work* g_work, struct stratum_ctx *sctx )
+char* hodl_malloc_txs_request( struct work *work )
+{
+  char* req;
+  json_t *val;
+  char data_str[2 * sizeof(work->data) + 1];
+  int i;
+
+  for ( i = 0; i < ARRAY_SIZE(work->data); i++ )
+    be32enc( work->data + i, work->data[i] );
+
+  bin2hex( data_str, (unsigned char *)work->data, 88 );
+  if ( work->workid )
+  {
+    char *params;
+    val = json_object();
+    json_object_set_new( val, "workid", json_string( work->workid ) );
+    params = json_dumps( val, 0 );
+    json_decref( val );
+    req = malloc( 128 + 2*88 + strlen( work->txs ) + strlen( params ) );
+    sprintf( req,
+     "{\"method\": \"submitblock\", \"params\": [\"%s%s\", %s], \"id\":1}\r\n",
+      data_str, work->txs, params);
+    free( params );
+  }
+  else
+  {
+    req = malloc( 128 + 2*88 + strlen(work->txs));
+    sprintf( req,
+       "{\"method\": \"submitblock\", \"params\": [\"%s%s\"], \"id\":1}\r\n",
+        data_str, work->txs);
+  }
+  return req;
+}
+
+
+void hodl_build_block_header( struct work* g_work, uint32_t version,
+                              uint32_t *prevhash, uint32_t *merkle_tree,
+                              uint32_t ntime, uint32_t nbits )
 {
-   uchar merkle_root[64] = { 0 };
-   size_t t;
   int i;

-   algo_gate.gen_merkle_root( merkle_root, sctx );
+   memset( g_work->data, 0, sizeof(g_work->data) );
+   g_work->data[0] = version;
+
+   if ( have_stratum )
+      for ( i = 0; i < 8; i++ )
+         g_work->data[1 + i] = le32dec( prevhash + i );
+   else
+      for (i = 0; i < 8; i++)
+         g_work->data[ 8-i ] = le32dec( prevhash + i );
+
+   for ( i = 0; i < 8; i++ )
+      g_work->data[9 + i] = be32dec(  merkle_tree + i );
+
+   g_work->data[ algo_gate.ntime_index ] =  ntime;
+   g_work->data[ algo_gate.nbits_index ] =  nbits;
+   g_work->data[22] = 0x80000000;
+   g_work->data[31] = 0x00000280;
+}
+
+// hodl build_extra_header is redundant, hodl can use std_build_extra_header
+// and call hodl_build_block_header.
+#if 0
+void hodl_build_extraheader( struct work* g_work, struct stratum_ctx *sctx )
+{
+   uchar merkle_tree[64] = { 0 };
+   size_t t;
+//   int i;
+
+   algo_gate.gen_merkle_root( merkle_tree, sctx );
   // Increment extranonce2
   for ( t = 0; t < sctx->xnonce2_size && !( ++sctx->job.xnonce2[t] ); t++ );
+
+   algo_gate.build_block_header( g_work, le32dec( sctx->job.version ),
+          (uint32_t*) sctx->job.prevhash, (uint32_t*) merkle_tree,
+          le32dec( sctx->job.ntime ), le32dec( sctx->job.nbits ) );
+/*
   // Assemble block header
   memset( g_work->data, 0, sizeof(g_work->data) );
   g_work->data[0] = le32dec( sctx->job.version );
@@ -63,7 +131,9 @@ void hodl_build_extraheader( struct work* g_work, struct stratum_ctx *sctx )
   g_work->data[ algo_gate.nbits_index ] = le32dec( sctx->job.nbits );
   g_work->data[22] = 0x80000000;
   g_work->data[31] = 0x00000280;
+*/
 }
+#endif

 // called only by thread 0, saves a backup of g_work
 void hodl_get_new_work( struct work* work, struct work* g_work)
@@ -73,6 +143,22 @@ void hodl_get_new_work( struct work* work, struct work* g_work)
     hodl_work.data[ algo_gate.nonce_index ] = ( clock() + rand() ) % 9999;
 }

+json_t *hodl_longpoll_rpc_call( CURL *curl, int *err, char* lp_url )
+{
+   json_t *val;
+   char *req = NULL;
+
+   if ( have_gbt )
+   {
+      req = malloc( strlen( gbt_lp_req ) + strlen( lp_id ) + 1 );
+      sprintf( req, gbt_lp_req, lp_id );
+   }
+   val = json_rpc_call( curl, lp_url, rpc_userpass,
+                        req ? req : getwork_req, err, JSON_RPC_LONGPOLL );
+   free( req );
+   return val;
+}
+
 // called by every thread, copies the backup to each thread's work.
 void hodl_resync_threads( struct work* work )
 {
@@ -95,10 +181,11 @@ int hodl_scanhash( int thr_id, struct work* work, uint32_t max_nonce,
                   uint64_t *hashes_done )
 {
 #ifndef NO_AES_NI
-  GenRandomGarbage( hodl_scratchbuf, work->data, thr_id );
+  GenRandomGarbage( (CacheEntry*)hodl_scratchbuf, work->data, thr_id );
  pthread_barrier_wait( &hodl_barrier );
  return scanhash_hodl_wolf( thr_id, work, max_nonce, hashes_done );
 #endif
+  return false;
 }

 bool register_hodl_algo( algo_gate_t* gate )
@@ -111,13 +198,17 @@ bool register_hodl_algo( algo_gate_t* gate )
  gate->optimizations         = SSE2_OPT | AES_OPT | AVX_OPT | AVX2_OPT;
  gate->scanhash              = (void*)&hodl_scanhash;
  gate->get_new_work          = (void*)&hodl_get_new_work;
+  gate->longpoll_rpc_call     = (void*)&hodl_longpoll_rpc_call;
  gate->set_target            = (void*)&hodl_set_target;
  gate->build_stratum_request = (void*)&hodl_le_build_stratum_request;
-  gate->build_extraheader     = (void*)&hodl_build_extraheader;
+  gate->malloc_txs_request    = (void*)&hodl_malloc_txs_request;
+  gate->build_block_header    = (void*)&hodl_build_block_header;
+//  gate->build_extraheader     = (void*)&hodl_build_extraheader;
  gate->resync_threads        = (void*)&hodl_resync_threads;
  gate->do_this_thread        = (void*)&hodl_do_this_thread;
  gate->work_cmp_size         = 76;
  hodl_scratchbuf = (unsigned char*)malloc( 1 << 30 );
+  allow_getwork = false;
  return ( hodl_scratchbuf != NULL );
 }

--- a/algo/hodl/hodl-wolf.c
+++ b/algo/hodl/hodl-wolf.c
@@ -150,6 +150,9 @@ int scanhash_hodl_wolf( int threadNumber, struct work* work, uint32_t max_nonce,
        int searchNumber = COMPARE_SIZE / opt_n_threads;
        int startLoc = threadNumber * searchNumber;

+        if ( opt_debug )
+           applog( LOG_DEBUG,"Hash target= %08lx", ptarget[7] );
+
        for(int32_t k = startLoc; k < startLoc + searchNumber && !work_restart[threadNumber].restart; k++)
        {
           // copy data to first l2 cache
--- a/algo/jh/jha-4way.c
+++ b/algo/jh/jha-4way.c
@@ -23,12 +23,12 @@ void jha_hash_4way( void *out, const void *input )
    uint64_t hash2[8] __attribute__ ((aligned (64)));
    uint64_t hash3[8] __attribute__ ((aligned (64)));
    uint64_t vhash[8*4] __attribute__ ((aligned (64)));
-    uint64_t vhash0[8*4] __attribute__ ((aligned (64)));
-    uint64_t vhash1[8*4] __attribute__ ((aligned (64)));
-    __m256i mask, mask0, mask1;
-    __m256i* vh = (__m256i*)vhash;
-    __m256i* vh0 = (__m256i*)vhash0;
-    __m256i* vh1 = (__m256i*)vhash1;
+    uint64_t vhashA[8*4] __attribute__ ((aligned (64)));
+    uint64_t vhashB[8*4] __attribute__ ((aligned (64)));
+    __m256i* vh  = (__m256i*)vhash;
+    __m256i* vhA = (__m256i*)vhashA;
+    __m256i* vhB = (__m256i*)vhashB;
+    __m256i vh_mask;

    blake512_4way_context  ctx_blake;
    hashState_groestl      ctx_groestl;
@@ -40,127 +40,65 @@ void jha_hash_4way( void *out, const void *input )
    keccak512_4way( &ctx_keccak, input, 80 );
    keccak512_4way_close( &ctx_keccak, vhash );

-//    memcpy( &ctx_keccak, &jha_kec_mid, sizeof jha_kec_mid );
-//    keccak512_4way( &ctx_keccak, input + (64<<2), 16 );
-//    keccak512_4way_close( &ctx_keccak, vhash );
-
    // Heavy & Light Pair Loop
    for ( int round = 0; round < 3; round++ )
    {
-      // select next function based on bit 0 of previous hash.
-      // Specutively execute both functions and use mask to
-      // select results from correct function for each lane.
-      // hash = mask : vhash0 ? vhash1
-      mask = mm256_negate_64(
-                     _mm256_and_si256( vh[0], _mm256_set1_epi64x( 0x1 ) ) );
-
-// second version
-//      mask0 = mask
-//      mask1 = mm256_not( mask );
-
-// first version
-//       mask = _mm256_sub_epi64( _mm256_and_si256( vh[0],
-//                     _mm256_set1_epi64x( 0x1 ) ), _mm256_set1_epi64x( 0x1 ) );
-
-       // groestl (serial) vs skein
+       vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256(
+               vh[0], _mm256_set1_epi64x( 1 ) ), m256_zero );

       mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
-
       init_groestl( &ctx_groestl, 64 );
       update_and_final_groestl( &ctx_groestl, (char*)hash0,
-                                 (char*)hash0, 512 );
+                                               (char*)hash0, 512 );
       init_groestl( &ctx_groestl, 64 );
       update_and_final_groestl( &ctx_groestl, (char*)hash1,
-                                 (char*)hash1, 512 );
+                                               (char*)hash1, 512 );
       init_groestl( &ctx_groestl, 64 );
       update_and_final_groestl( &ctx_groestl, (char*)hash2,
-                                 (char*)hash2, 512 );
+                                               (char*)hash2, 512 );
       init_groestl( &ctx_groestl, 64 );
       update_and_final_groestl( &ctx_groestl, (char*)hash3,
-                                 (char*)hash3, 512 );
-
-       mm256_interleave_4x64( vhash0, hash0, hash1, hash2, hash3, 512 );
-
-       // skein
+                                               (char*)hash3, 512 );
+       mm256_interleave_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );

       skein512_4way_init( &ctx_skein );
       skein512_4way( &ctx_skein, vhash, 64 );
-       skein512_4way_close( &ctx_skein, vhash1 );
+       skein512_4way_close( &ctx_skein, vhashB );

-       // merge vectored hash
       for ( int i = 0; i < 8; i++ )
-       {
-          // blend should be faster
-          vh[i] = _mm256_blendv_epi8( vh0[i], vh1[i], mask );
-
-// second version
-//          vh[i] = _mm256_or_si256( _mm256_and_si256( vh0[i], mask0 ),
-//                                   _mm256_and_si256( vh1[i], mask1 ) );
-
-// first version
-/*
-          vh0[i] = _mm256_maskload_epi64( 
-                                      vhash0 + i*4, mm256_not( mask ) );
-          vh1[i] = _mm256_maskload_epi64(
-                                      vhash1 + i*4, mask );
-          vh[i]  = _mm256_or_si256( vh0[i], vh1[i] );
-*/
-       }
-
-       // blake v jh
+          vh[i] = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );

       blake512_4way_init( &ctx_blake );
       blake512_4way( &ctx_blake, vhash, 64 );
-       blake512_4way_close( &ctx_blake, vhash0 );
+       blake512_4way_close( &ctx_blake, vhashA );

       jh512_4way_init( &ctx_jh );
       jh512_4way( &ctx_jh, vhash, 64 );
-       jh512_4way_close( &ctx_jh, vhash1 );
+       jh512_4way_close( &ctx_jh, vhashB );

-       // merge hash
       for ( int i = 0; i < 8; i++ )
-       {
-          vh[i] = _mm256_or_si256( _mm256_and_si256( vh0[i], mask0 ),
-                                   _mm256_and_si256( vh1[i], mask1 ) );
-/*
-          vha256[i] = _mm256_maskload_epi64(
-                                      vhasha + i*4, mm256_not( mask ) );
-          vhb256[i] = _mm256_maskload_epi64(
-                                      vhashb + i*4, mask );
-          vh256[i]  = _mm256_or_si256( vha256[i], vhb256[i] );
-*/
-       }
+          vh[i] = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );
    }

    mm256_deinterleave_4x64( out, out+32, out+64, out+96, vhash, 256 );
-
-//    memcpy( output,       hash0, 32 );
-//    memcpy( output+32,    hash1, 32 );
-//    memcpy( output+64,    hash2, 32 );
-//    memcpy( output+96,    hash3, 32 );
-
 }

 int scanhash_jha_4way( int thr_id, struct work *work, uint32_t max_nonce,
                       uint64_t *hashes_done )
 {
-     uint32_t hash[8*4] __attribute__ ((aligned (64)));
-     uint32_t vdata[20*4] __attribute__ ((aligned (64)));
-     uint32_t endiandata[20] __attribute__((aligned(64)));
-	uint32_t *pdata = work->data;
-	uint32_t *ptarget = work->target;
-	const uint32_t first_nonce = pdata[19];
-	const uint32_t Htarg = ptarget[7];
-	uint32_t n = pdata[19];
-     uint32_t *nonces = work->nonces;
-     bool *found = work->nfound;
-     int num_found = 0;
-     uint32_t *noncep0 = vdata + 73;   // 9*8 + 1
-     uint32_t *noncep1 = vdata + 75;
-     uint32_t *noncep2 = vdata + 77;
-     uint32_t *noncep3 = vdata + 79;
+   uint32_t hash[8*4] __attribute__ ((aligned (64)));
+   uint32_t vdata[20*4] __attribute__ ((aligned (64)));
+   uint32_t endiandata[20] __attribute__((aligned(64)));
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   const uint32_t first_nonce = pdata[19];
+   const uint32_t Htarg = ptarget[7];
+   uint32_t n = pdata[19];
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   uint32_t *noncep = vdata + 73;   // 9*8 + 1

-	uint64_t htmax[] = {
+   uint64_t htmax[] = {
 		0,
 		0xF,
 		0xFF,
@@ -168,7 +106,7 @@ int scanhash_jha_4way( int thr_id, struct work *work, uint32_t max_nonce,
 		0xFFFF,
 		0x10000000
 	};
-	uint32_t masks[] = {
+   uint32_t masks[] = {
 		0xFFFFFFFF,
 		0xFFFFFFF0,
 		0xFFFFFF00,
@@ -177,73 +115,40 @@ int scanhash_jha_4way( int thr_id, struct work *work, uint32_t max_nonce,
 		0
 	};

-   // we need bigendian data...
   for ( int i=0; i < 19; i++ )
      be32enc( &endiandata[i], pdata[i] );

   uint64_t *edata = (uint64_t*)endiandata;
   mm256_interleave_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );

-   // precalc midstate for keccak
-//   keccak512_4way_init( &jha_kec_mid );
-//   keccak512_4way( &jha_kec_mid, vdata, 64 );
-
   for ( int m = 0; m < 6; m++ )
   {
      if ( Htarg <= htmax[m] )
      {
         uint32_t mask = masks[m];
         do {
-              found[0] = found[1] = found[2] = found[3] = false;
-              be32enc( noncep0, n   );
-              be32enc( noncep1, n+1 );
-              be32enc( noncep2, n+2 );
-              be32enc( noncep3, n+3 );
+              be32enc( noncep,   n   );
+              be32enc( noncep+2, n+1 );
+              be32enc( noncep+4, n+2 );
+              be32enc( noncep+6, n+3 );

              jha_hash_4way( hash, vdata );
-
              pdata[19] = n;

-              if ( ( !(hash[7] & mask) )
-                   && fulltest( hash, ptarget ) )
+              for ( int i = 0; i < 4; i++ )
+              if ( ( !( (hash+(i<<3))[7] & mask ) == 0 )
+                  && fulltest( hash+(i<<3), ptarget ) )
              {
-                 found[0] = true;
-                 num_found++;
-                 nonces[0] = n;
-                 work_set_target_ratio( work, hash );
-              }
-              if ( ( !((hash+8)[7] & mask) )
-                   && fulltest( hash+8, ptarget ) )
-              {
-                 found[1] = true;
-                 num_found++;
-                 nonces[1] = n+1;
-                 work_set_target_ratio( work, hash+8 );
-              }
-              if ( ( !((hash+16)[7] & mask) )
-                 && fulltest( hash+16, ptarget ) )
-              {
-                 found[2] = true;
-                 num_found++;
-                 nonces[2] = n+2;
-                 work_set_target_ratio( work, hash+16 );
-              }
-              if ( ( !((hash+24)[7] & mask) )
-                   && fulltest( hash+24, ptarget ) )
-              {
-                 found[3] = true;
-                 num_found++;
-                 nonces[3] = n+3;
-                 work_set_target_ratio( work, hash+24 );
+                 pdata[19] = n;
+                 nonces[ num_found++ ] = n+i;
+                 work_set_target_ratio( work, hash+(i<<3) );
              }
              n += 4;
         } while ( ( num_found == 0 ) && ( n < max_nonce )
                     && !work_restart[thr_id].restart );
-
         break;
      }
   }
-
   *hashes_done = n - first_nonce + 1;
   return num_found;
 }
--- a/algo/jh/jha-gate.c
+++ b/algo/jh/jha-gate.c
@@ -5,14 +5,13 @@ bool register_jha_algo( algo_gate_t* gate )
 {
 #if defined (JHA_4WAY)
  four_way_not_tested();
-  gate->optimizations = SSE2_OPT | AES_OPT | FOUR_WAY_OPT;
  gate->scanhash         = (void*)&scanhash_jha_4way;
  gate->hash             = (void*)&jha_hash_4way;
 #else
-  gate->optimizations = SSE2_OPT | AES_OPT | FOUR_WAY_OPT;
  gate->scanhash         = (void*)&scanhash_jha;
  gate->hash             = (void*)&jha_hash;
 #endif
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
  gate->set_target       = (void*)&scrypt_set_target;
  return true;
 };
--- a/algo/jh/jha-gate.h
+++ b/algo/jh/jha-gate.h
@@ -5,7 +5,7 @@
 #include <stdint.h>


-#if defined(FOUR_WAY) && defined(__AVX2__) && !defined(NO_AES_NI)
+#if defined(__AVX2__) && defined(__AES__)
  #define JHA_4WAY
 #endif

--- a/algo/jh/sse2/jh_sse2_opt64.h
+++ b/algo/jh/sse2/jh_sse2_opt64.h
@@ -339,13 +339,13 @@ do { \
        jhSbuffer[53] = 0x00, \
        jhSbuffer[54] = 0x00, \
        jhSbuffer[55] = 0x00; \
-        jhSbuffer[56] = ((64*8) >> 56) & 0xff, \
-        jhSbuffer[57] = ((64*8) >> 48) & 0xff, \
-        jhSbuffer[58] = ((64*8) >> 40) & 0xff, \
-        jhSbuffer[59] = ((64*8) >> 32) & 0xff, \
-        jhSbuffer[60] = ((64*8) >> 24) & 0xff, \
-        jhSbuffer[61] = ((64*8) >> 16) & 0xff, \
-        jhSbuffer[62] = ((64*8) >> 8) & 0xff, \
+        jhSbuffer[56] = ((char)((uint64_t)(64*8) >> 56)) & 0xff, \
+        jhSbuffer[57] = ((char)((uint64_t)(64*8) >> 48)) & 0xff, \
+        jhSbuffer[58] = ((char)((uint64_t)(64*8) >> 40)) & 0xff, \
+        jhSbuffer[59] = ((char)((uint64_t)(64*8) >> 32)) & 0xff, \
+        jhSbuffer[60] = ((char)((uint64_t)(64*8) >> 24)) & 0xff, \
+        jhSbuffer[61] = ((char)((uint64_t)(64*8) >> 16)) & 0xff, \
+        jhSbuffer[62] = ((char)((uint64_t)(64*8) >> 8)) & 0xff, \
        jhSbuffer[63] = (64*8) & 0xff; \
        b = true; \
    } \
--- a/algo/keccak/keccak-4way.c
+++ b/algo/keccak/keccak-4way.c
@@ -32,12 +32,8 @@ int scanhash_keccak_4way( int thr_id, struct work *work, uint32_t max_nonce,
 //   const uint32_t Htarg = ptarget[7];
   uint32_t endiandata[20];
   uint32_t *nonces = work->nonces;
-   bool *found = work->nfound;
   int num_found = 0;
-   uint32_t *noncep0 = vdata + 73;   // 9*8 + 1
-   uint32_t *noncep1 = vdata + 75;
-   uint32_t *noncep2 = vdata + 77;
-   uint32_t *noncep3 = vdata + 79;
+   uint32_t *noncep = vdata + 73;   // 9*8 + 1

   for ( int i=0; i < 19; i++ ) 
      be32enc( &endiandata[i], pdata[i] );
@@ -46,42 +42,20 @@ int scanhash_keccak_4way( int thr_id, struct work *work, uint32_t max_nonce,
   mm256_interleave_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );

   do {
-      found[0] = found[1] = found[2] = found[3] = false;
-      be32enc( noncep0, n   );
-      be32enc( noncep1, n+1 );
-      be32enc( noncep2, n+2 );
-      be32enc( noncep3, n+3 );
+      be32enc( noncep,   n   );
+      be32enc( noncep+2, n+1 );
+      be32enc( noncep+4, n+2 );
+      be32enc( noncep+6, n+3 );
 	
      keccakhash_4way( hash, vdata );

-      if ( ( ( hash[7] & 0xFFFFFF00 ) == 0 )
-         && fulltest( hash, ptarget) )
+      for ( int i = 0; i < 4; i++ )
+      if ( ( ( (hash+(i<<3))[7] & 0xFFFFFF00 ) == 0 )
+           && fulltest( hash+(i<<3), ptarget ) )
      {
-          found[0] = true;
-          num_found++;
-          nonces[0] = n;
-          pdata[19] = n;
-      }
-      if ( ( ( (hash+8)[7] & 0xFFFFFF00 ) == 0 )
-         && fulltest( hash+8, ptarget) ) 
-      {
-          found[1] = true;
-          num_found++;
-          nonces[1] = n+1;
-      }
-      if ( ( ( (hash+16) [7] & 0xFFFFFF00 ) == 0 )
-         && fulltest( hash+16, ptarget) )
-      {
-          found[2] = true;
-          num_found++;
-          nonces[2] = n+2;
-      }
-      if ( ( ( (hash+24)[7] & 0xFFFFFF00 ) == 0 )
-         && fulltest( hash+24, ptarget) )
-      {
-          found[3] = true;
-          num_found++;
-          nonces[3] = n+3;
+         pdata[19] = n+i;
+         nonces[ num_found++ ] = n+i;
+         work_set_target_ratio( work, hash+(i<<3) );
      }
      n += 4;

--- a/algo/keccak/keccak-gate.c
+++ b/algo/keccak/keccak-gate.c
@@ -9,7 +9,7 @@ int64_t keccak_get_max64() { return 0x7ffffLL; }

 bool register_keccak_algo( algo_gate_t* gate )
 {
-  gate->optimizations = FOUR_WAY_OPT;
+  gate->optimizations = AVX2_OPT;
  gate->gen_merkle_root = (void*)&SHA256_gen_merkle_root;
  gate->set_target      = (void*)&keccak_set_target;
  gate->get_max64       = (void*)&keccak_get_max64;
@@ -30,7 +30,7 @@ void keccakc_set_target( struct work* work, double job_diff )

 bool register_keccakc_algo( algo_gate_t* gate )
 {
-  gate->optimizations = FOUR_WAY_OPT;
+  gate->optimizations = AVX2_OPT;
  gate->gen_merkle_root = (void*)&sha256d_gen_merkle_root;
  gate->set_target      = (void*)&keccakc_set_target;
  gate->get_max64       = (void*)&keccak_get_max64;
--- a/algo/keccak/keccak-gate.h
+++ b/algo/keccak/keccak-gate.h
@@ -4,7 +4,7 @@
 #include "algo-gate-api.h"
 #include <stdint.h>

-#if defined(FOUR_WAY) && defined(__AVX2__)
+#if defined(__AVX2__)
  #define KECCAK_4WAY
 #endif

--- a/algo/keccak/keccak-hash-4way.c
+++ b/algo/keccak/keccak-hash-4way.c
@@ -59,7 +59,7 @@ static const sph_u64 RC[] = {
 #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,mm256_neg1))
+#define NOT64(d, s)      (d = _mm256_xor_si256(s,m256_neg1))
 #define ROL64(d, v, n)   (d = mm256_rotl_64(v, n))
 #define XOR64_IOTA       XOR64

@@ -375,12 +375,12 @@ static void keccak64_init( keccak64_ctx_m256i *kc, unsigned out_size )
          kc->w[i] = _mm256_setzero_si256();

   // Initialization for the "lane complement".
-   kc->w[ 1] = mm256_neg1;
-   kc->w[ 2] = mm256_neg1;
-   kc->w[ 8] = mm256_neg1;
-   kc->w[12] = mm256_neg1;
-   kc->w[17] = mm256_neg1;
-   kc->w[20] = mm256_neg1;
+   kc->w[ 1] = m256_neg1;
+   kc->w[ 2] = m256_neg1;
+   kc->w[ 8] = m256_neg1;
+   kc->w[12] = m256_neg1;
+   kc->w[17] = m256_neg1;
+   kc->w[20] = m256_neg1;
   kc->ptr = 0;
   kc->lim = 200 - (out_size >> 2);
 }
--- a/algo/keccak/sse2/keccak.c
+++ b/algo/keccak/sse2/keccak.c
@@ -775,10 +775,8 @@ static const sph_u64 RC[] = {
 			KF_ELT( 5,  6, RC[j + 5]); \
 			KF_ELT( 6,  7, RC[j + 6]); \
 			KF_ELT( 7,  8, RC[j + 7]); \
-*/
-
-	//kekDECL_STATE \
-        
+	kekDECL_STATE \
+*/        
 #define DECL_KEC  


--- a/algo/luffa/luffa-hash-2way.c
+++ b/algo/luffa/luffa-hash-2way.c
@@ -0,0 +1,583 @@
+/*
+ * luffa_for_sse2.c
+ * Version 2.0 (Sep 15th 2009)
+ *
+ * Copyright (C) 2008-2009 Hitachi, Ltd. All rights reserved.
+ *
+ * Hitachi, Ltd. is the owner of this software and hereby grant
+ * the U.S. Government and any interested party the right to use
+ * this software for the purposes of the SHA-3 evaluation process,
+ * notwithstanding that this software is copyrighted.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <string.h>
+#include <immintrin.h>
+#include "luffa-hash-2way.h"
+
+#if defined(__AVX2__)
+
+#include "avxdefs.h"
+
+#define MASK _mm256_set_epi32( 0UL, 0UL, 0UL, 0xffffffffUL, \
+                               0UL, 0UL, 0UL, 0xffffffffUL )
+
+#define ADD_CONSTANT(a,b,c0,c1)\
+    a = _mm256_xor_si256(a,c0);\
+    b = _mm256_xor_si256(b,c1);\
+
+#define MULT2(a0,a1) \
+do { \
+  register __m256i b = _mm256_xor_si256( a0, \
+                   _mm256_shuffle_epi32( _mm256_and_si256(a1,MASK), 16 ) ); \
+  a0 = _mm256_or_si256( _mm256_srli_si256(b,4), _mm256_slli_si256(a1,12) ); \
+  a1 = _mm256_or_si256( _mm256_srli_si256(a1,4), _mm256_slli_si256(b,12) );  \
+} while(0)
+
+// confirm pointer arithmetic
+// ok but use array indexes
+#define STEP_PART(x,c,t)\
+    SUBCRUMB(*x,*(x+1),*(x+2),*(x+3),*t);\
+    SUBCRUMB(*(x+5),*(x+6),*(x+7),*(x+4),*t);\
+    MIXWORD(*x,*(x+4),*t,*(t+1));\
+    MIXWORD(*(x+1),*(x+5),*t,*(t+1));\
+    MIXWORD(*(x+2),*(x+6),*t,*(t+1));\
+    MIXWORD(*(x+3),*(x+7),*t,*(t+1));\
+    ADD_CONSTANT(*x, *(x+4), *c, *(c+1));
+
+#define SUBCRUMB(a0,a1,a2,a3,t)\
+    t  = _mm256_load_si256(&a0);\
+    a0 = _mm256_or_si256(a0,a1);\
+    a2 = _mm256_xor_si256(a2,a3);\
+    a1 = _mm256_andnot_si256(a1, m256_neg1 );\
+    a0 = _mm256_xor_si256(a0,a3);\
+    a3 = _mm256_and_si256(a3,t);\
+    a1 = _mm256_xor_si256(a1,a3);\
+    a3 = _mm256_xor_si256(a3,a2);\
+    a2 = _mm256_and_si256(a2,a0);\
+    a0 = _mm256_andnot_si256(a0, m256_neg1 );\
+    a2 = _mm256_xor_si256(a2,a1);\
+    a1 = _mm256_or_si256(a1,a3);\
+    t  = _mm256_xor_si256(t,a1);\
+    a3 = _mm256_xor_si256(a3,a2);\
+    a2 = _mm256_and_si256(a2,a1);\
+    a1 = _mm256_xor_si256(a1,a0);\
+    a0 = _mm256_load_si256(&t);\
+
+#define MIXWORD(a,b,t1,t2)\
+    b  = _mm256_xor_si256(a,b);\
+    t1 = _mm256_slli_epi32(a,2);\
+    t2 = _mm256_srli_epi32(a,30);\
+     a = _mm256_or_si256(t1,t2);\
+    a  = _mm256_xor_si256(a,b);\
+    t1 = _mm256_slli_epi32(b,14);\
+    t2 = _mm256_srli_epi32(b,18);\
+    b  = _mm256_or_si256(t1,t2);\
+    b  = _mm256_xor_si256(a,b);\
+    t1 = _mm256_slli_epi32(a,10);\
+    t2 = _mm256_srli_epi32(a,22);\
+    a  = _mm256_or_si256(t1,t2);\
+    a  = _mm256_xor_si256(a,b);\
+    t1 = _mm256_slli_epi32(b,1);\
+    t2 = _mm256_srli_epi32(b,31);\
+    b  = _mm256_or_si256(t1,t2);
+
+#define STEP_PART2(a0,a1,t0,t1,c0,c1,tmp0,tmp1)\
+    a1 = _mm256_shuffle_epi32(a1,147);\
+    t0 = _mm256_load_si256(&a1);\
+    a1 = _mm256_unpacklo_epi32(a1,a0);\
+    t0 = _mm256_unpackhi_epi32(t0,a0);\
+    t1 = _mm256_shuffle_epi32(t0,78);\
+    a0 = _mm256_shuffle_epi32(a1,78);\
+    SUBCRUMB(t1,t0,a0,a1,tmp0);\
+    t0 = _mm256_unpacklo_epi32(t0,t1);\
+    a1 = _mm256_unpacklo_epi32(a1,a0);\
+    a0 = _mm256_load_si256(&a1);\
+    a0 = _mm256_unpackhi_epi64(a0,t0);\
+    a1 = _mm256_unpacklo_epi64(a1,t0);\
+    a1 = _mm256_shuffle_epi32(a1,57);\
+    MIXWORD(a0,a1,tmp0,tmp1);\
+    ADD_CONSTANT(a0,a1,c0,c1);
+
+#define NMLTOM768(r0,r1,r2,s0,s1,s2,s3,p0,p1,p2,q0,q1,q2,q3)\
+    s2 = _mm256_load_si256(&r1);\
+    q2 = _mm256_load_si256(&p1);\
+    r2 = _mm256_shuffle_epi32(r2,216);\
+    p2 = _mm256_shuffle_epi32(p2,216);\
+    r1 = _mm256_unpacklo_epi32(r1,r0);\
+    p1 = _mm256_unpacklo_epi32(p1,p0);\
+    s2 = _mm256_unpackhi_epi32(s2,r0);\
+    q2 = _mm256_unpackhi_epi32(q2,p0);\
+    s0 = _mm256_load_si256(&r2);\
+    q0 = _mm256_load_si256(&p2);\
+    r2 = _mm256_unpacklo_epi64(r2,r1);\
+    p2 = _mm256_unpacklo_epi64(p2,p1);\
+    s1 = _mm256_load_si256(&s0);\
+    q1 = _mm256_load_si256(&q0);\
+    s0 = _mm256_unpackhi_epi64(s0,r1);\
+    q0 = _mm256_unpackhi_epi64(q0,p1);\
+    r2 = _mm256_shuffle_epi32(r2,225);\
+    p2 = _mm256_shuffle_epi32(p2,225);\
+    r0 = _mm256_load_si256(&s1);\
+    p0 = _mm256_load_si256(&q1);\
+    s0 = _mm256_shuffle_epi32(s0,225);\
+    q0 = _mm256_shuffle_epi32(q0,225);\
+    s1 = _mm256_unpacklo_epi64(s1,s2);\
+    q1 = _mm256_unpacklo_epi64(q1,q2);\
+    r0 = _mm256_unpackhi_epi64(r0,s2);\
+    p0 = _mm256_unpackhi_epi64(p0,q2);\
+    s2 = _mm256_load_si256(&r0);\
+    q2 = _mm256_load_si256(&p0);\
+    s3 = _mm256_load_si256(&r2);\
+    q3 = _mm256_load_si256(&p2);\
+
+#define MIXTON768(r0,r1,r2,r3,s0,s1,s2,p0,p1,p2,p3,q0,q1,q2)\
+    s0 = _mm256_load_si256(&r0);\
+    q0 = _mm256_load_si256(&p0);\
+    s1 = _mm256_load_si256(&r2);\
+    q1 = _mm256_load_si256(&p2);\
+    r0 = _mm256_unpackhi_epi32(r0,r1);\
+    p0 = _mm256_unpackhi_epi32(p0,p1);\
+    r2 = _mm256_unpackhi_epi32(r2,r3);\
+    p2 = _mm256_unpackhi_epi32(p2,p3);\
+    s0 = _mm256_unpacklo_epi32(s0,r1);\
+    q0 = _mm256_unpacklo_epi32(q0,p1);\
+    s1 = _mm256_unpacklo_epi32(s1,r3);\
+    q1 = _mm256_unpacklo_epi32(q1,p3);\
+    r1 = _mm256_load_si256(&r0);\
+    p1 = _mm256_load_si256(&p0);\
+    r0 = _mm256_unpackhi_epi64(r0,r2);\
+    p0 = _mm256_unpackhi_epi64(p0,p2);\
+    s0 = _mm256_unpackhi_epi64(s0,s1);\
+    q0 = _mm256_unpackhi_epi64(q0,q1);\
+    r1 = _mm256_unpacklo_epi64(r1,r2);\
+    p1 = _mm256_unpacklo_epi64(p1,p2);\
+    s2 = _mm256_load_si256(&r0);\
+    q2 = _mm256_load_si256(&p0);\
+    s1 = _mm256_load_si256(&r1);\
+    q1 = _mm256_load_si256(&p1);\
+
+#define NMLTOM1024(r0,r1,r2,r3,s0,s1,s2,s3,p0,p1,p2,p3,q0,q1,q2,q3)\
+    s1 = _mm256_load_si256(&r3);\
+    q1 = _mm256_load_si256(&p3);\
+    s3 = _mm256_load_si256(&r3);\
+    q3 = _mm256_load_si256(&p3);\
+    s1 = _mm256_unpackhi_epi32(s1,r2);\
+    q1 = _mm256_unpackhi_epi32(q1,p2);\
+    s3 = _mm256_unpacklo_epi32(s3,r2);\
+    q3 = _mm256_unpacklo_epi32(q3,p2);\
+    s0 = _mm256_load_si256(&s1);\
+    q0 = _mm256_load_si256(&q1);\
+    s2 = _mm256_load_si256(&s3);\
+    q2 = _mm256_load_si256(&q3);\
+    r3 = _mm256_load_si256(&r1);\
+    p3 = _mm256_load_si256(&p1);\
+    r1 = _mm256_unpacklo_epi32(r1,r0);\
+    p1 = _mm256_unpacklo_epi32(p1,p0);\
+    r3 = _mm256_unpackhi_epi32(r3,r0);\
+    p3 = _mm256_unpackhi_epi32(p3,p0);\
+    s0 = _mm256_unpackhi_epi64(s0,r3);\
+    q0 = _mm256_unpackhi_epi64(q0,p3);\
+    s1 = _mm256_unpacklo_epi64(s1,r3);\
+    q1 = _mm256_unpacklo_epi64(q1,p3);\
+    s2 = _mm256_unpackhi_epi64(s2,r1);\
+    q2 = _mm256_unpackhi_epi64(q2,p1);\
+    s3 = _mm256_unpacklo_epi64(s3,r1);\
+    q3 = _mm256_unpacklo_epi64(q3,p1);
+
+#define MIXTON1024(r0,r1,r2,r3,s0,s1,s2,s3,p0,p1,p2,p3,q0,q1,q2,q3)\
+    NMLTOM1024(r0,r1,r2,r3,s0,s1,s2,s3,p0,p1,p2,p3,q0,q1,q2,q3);
+
+/* initial values of chaining variables */
+static const uint32 IV[40] __attribute((aligned(32))) = {
+    0xdbf78465,0x4eaa6fb4,0x44b051e0,0x6d251e69,
+    0xdef610bb,0xee058139,0x90152df4,0x6e292011,
+    0xde099fa3,0x70eee9a0,0xd9d2f256,0xc3b44b95,
+    0x746cd581,0xcf1ccf0e,0x8fc944b3,0x5d9b0557,
+    0xad659c05,0x04016ce5,0x5dba5781,0xf7efc89d,
+    0x8b264ae7,0x24aa230a,0x666d1836,0x0306194f,
+    0x204b1f67,0xe571f7d7,0x36d79cce,0x858075d5,
+    0x7cde72ce,0x14bcb808,0x57e9e923,0x35870c6a,
+    0xaffb4363,0xc825b7c7,0x5ec41e22,0x6c68e9be,
+    0x03e86cea,0xb07224cc,0x0fc688f1,0xf5df3999
+};
+
+/* Round Constants */
+static const uint32 CNS_INIT[128] __attribute((aligned(32))) = {
+    0xb213afa5,0xfc20d9d2,0xb6de10ed,0x303994a6,
+    0xe028c9bf,0xe25e72c1,0x01685f3d,0xe0337818,
+    0xc84ebe95,0x34552e25,0x70f47aae,0xc0e65299,
+    0x44756f91,0xe623bb72,0x05a17cf4,0x441ba90d,
+    0x4e608a22,0x7ad8818f,0x0707a3d4,0x6cc33a12,
+    0x7e8fce32,0x5c58a4a4,0xbd09caca,0x7f34d442,
+    0x56d858fe,0x8438764a,0x1c1e8f51,0xdc56983e,
+    0x956548be,0x1e38e2e7,0xf4272b28,0x9389217f,
+    0x343b138f,0xbb6de032,0x707a3d45,0x1e00108f,
+    0xfe191be2,0x78e38b9d,0x144ae5cc,0xe5a8bce6,
+    0xd0ec4e3d,0xedb780c8,0xaeb28562,0x7800423d,
+    0x3cb226e5,0x27586719,0xfaa7ae2b,0x5274baf4,
+    0x2ceb4882,0xd9847356,0xbaca1589,0x8f5b7882,
+    0x5944a28e,0x36eda57f,0x2e48f1c1,0x26889ba7,
+    0xb3ad2208,0xa2c78434,0x40a46f3e,0x96e1db12,
+    0xa1c4c355,0x703aace7,0xb923c704,0x9a226e9d,
+    0x00000000,0x00000000,0x00000000,0xf0d2e9e3,
+    0x00000000,0x00000000,0x00000000,0x5090d577,
+    0x00000000,0x00000000,0x00000000,0xac11d7fa,
+    0x00000000,0x00000000,0x00000000,0x2d1925ab,
+    0x00000000,0x00000000,0x00000000,0x1bcb66f2,
+    0x00000000,0x00000000,0x00000000,0xb46496ac,
+    0x00000000,0x00000000,0x00000000,0x6f2d9bc9,
+    0x00000000,0x00000000,0x00000000,0xd1925ab0,
+    0x00000000,0x00000000,0x00000000,0x78602649,
+    0x00000000,0x00000000,0x00000000,0x29131ab6,
+    0x00000000,0x00000000,0x00000000,0x8edae952,
+    0x00000000,0x00000000,0x00000000,0x0fc053c3,
+    0x00000000,0x00000000,0x00000000,0x3b6ba548,
+    0x00000000,0x00000000,0x00000000,0x3f014f0c,
+    0x00000000,0x00000000,0x00000000,0xedae9520,
+    0x00000000,0x00000000,0x00000000,0xfc053c31
+};
+
+__m256i CNS[32];
+
+/***************************************************/
+/* Round function         */
+/* state: hash context    */
+
+void rnd512_2way( luffa_2way_context *state, __m256i *msg )
+{
+    __m256i t0, t1;
+    __m256i *chainv = state->chainv;
+    __m256i msg0, msg1;
+    __m256i tmp[2];
+    __m256i x[8];
+
+    t0 = chainv[0];
+    t1 = chainv[1];
+
+    t0 = _mm256_xor_si256( t0, chainv[2] );
+    t1 = _mm256_xor_si256( t1, chainv[3] );
+    t0 = _mm256_xor_si256( t0, chainv[4] );
+    t1 = _mm256_xor_si256( t1, chainv[5] );
+    t0 = _mm256_xor_si256( t0, chainv[6] );
+    t1 = _mm256_xor_si256( t1, chainv[7] );
+    t0 = _mm256_xor_si256( t0, chainv[8] );
+    t1 = _mm256_xor_si256( t1, chainv[9] );
+
+    MULT2( t0, t1 );
+
+    msg0 = _mm256_shuffle_epi32( msg[0], 27 );
+    msg1 = _mm256_shuffle_epi32( msg[1], 27 );
+
+    chainv[0] = _mm256_xor_si256( chainv[0], t0 );
+    chainv[1] = _mm256_xor_si256( chainv[1], t1 );
+    chainv[2] = _mm256_xor_si256( chainv[2], t0 );
+    chainv[3] = _mm256_xor_si256( chainv[3], t1 );
+    chainv[4] = _mm256_xor_si256( chainv[4], t0 );
+    chainv[5] = _mm256_xor_si256( chainv[5], t1 );
+    chainv[6] = _mm256_xor_si256( chainv[6], t0 );
+    chainv[7] = _mm256_xor_si256( chainv[7], t1 );
+    chainv[8] = _mm256_xor_si256( chainv[8], t0 );
+    chainv[9] = _mm256_xor_si256( chainv[9], t1 );
+
+    t0 = chainv[0];
+    t1 = chainv[1];
+
+    MULT2( chainv[0], chainv[1]);
+    chainv[0] = _mm256_xor_si256( chainv[0], chainv[2] );
+    chainv[1] = _mm256_xor_si256( chainv[1], chainv[3] );
+
+    MULT2( chainv[2], chainv[3]);
+    chainv[2] = _mm256_xor_si256(chainv[2], chainv[4]);
+    chainv[3] = _mm256_xor_si256(chainv[3], chainv[5]);
+
+    MULT2( chainv[4], chainv[5]);
+    chainv[4] = _mm256_xor_si256(chainv[4], chainv[6]);
+    chainv[5] = _mm256_xor_si256(chainv[5], chainv[7]);
+
+    MULT2( chainv[6], chainv[7]);
+    chainv[6] = _mm256_xor_si256(chainv[6], chainv[8]);
+    chainv[7] = _mm256_xor_si256(chainv[7], chainv[9]);
+
+    MULT2( chainv[8], chainv[9]);
+    chainv[8] = _mm256_xor_si256( chainv[8], t0 );
+    chainv[9] = _mm256_xor_si256( chainv[9], t1 );
+
+    t0 = chainv[8];
+    t1 = chainv[9];
+
+    MULT2( chainv[8], chainv[9]);
+    chainv[8] = _mm256_xor_si256( chainv[8], chainv[6] );
+    chainv[9] = _mm256_xor_si256( chainv[9], chainv[7] );
+
+    MULT2( chainv[6], chainv[7]);
+    chainv[6] = _mm256_xor_si256( chainv[6], chainv[4] );
+    chainv[7] = _mm256_xor_si256( chainv[7], chainv[5] );
+
+    MULT2( chainv[4], chainv[5]);
+    chainv[4] = _mm256_xor_si256( chainv[4], chainv[2] );
+    chainv[5] = _mm256_xor_si256( chainv[5], chainv[3] );
+
+    MULT2( chainv[2], chainv[3] );
+    chainv[2] = _mm256_xor_si256( chainv[2], chainv[0] );
+    chainv[3] = _mm256_xor_si256( chainv[3], chainv[1] );
+
+    MULT2( chainv[0], chainv[1] );
+    chainv[0] = _mm256_xor_si256( _mm256_xor_si256( chainv[0], t0 ), msg0 );
+    chainv[1] = _mm256_xor_si256( _mm256_xor_si256( chainv[1], t1 ), msg1 );
+
+    MULT2( msg0, msg1);
+    chainv[2] = _mm256_xor_si256( chainv[2], msg0 );
+    chainv[3] = _mm256_xor_si256( chainv[3], msg1 );
+
+    MULT2( msg0, msg1);
+    chainv[4] = _mm256_xor_si256( chainv[4], msg0 );
+    chainv[5] = _mm256_xor_si256( chainv[5], msg1 );
+
+    MULT2( msg0, msg1);
+    chainv[6] = _mm256_xor_si256( chainv[6], msg0 );
+    chainv[7] = _mm256_xor_si256( chainv[7], msg1 );
+
+    MULT2( msg0, msg1);
+    chainv[8] = _mm256_xor_si256( chainv[8], msg0 );
+    chainv[9] = _mm256_xor_si256( chainv[9], msg1 );
+
+    MULT2( msg0, msg1);
+
+    chainv[3] = _mm256_or_si256( _mm256_slli_epi32( chainv[3],  1 ),
+                                 _mm256_srli_epi32( chainv[3], 31 ) );
+    chainv[5] = _mm256_or_si256( _mm256_slli_epi32( chainv[5],  2 ),
+                                 _mm256_srli_epi32( chainv[5], 30 ) );
+    chainv[7] = _mm256_or_si256( _mm256_slli_epi32( chainv[7],  3 ),
+                                 _mm256_srli_epi32( chainv[7], 29 ) );
+    chainv[9] = _mm256_or_si256( _mm256_slli_epi32( chainv[9],  4 ),
+                                 _mm256_srli_epi32( chainv[9], 28 ) );
+
+    NMLTOM1024( chainv[0], chainv[2], chainv[4], chainv[6],
+                x[0], x[1], x[2], x[3],
+                chainv[1],chainv[3],chainv[5],chainv[7],
+                x[4], x[5], x[6], x[7] );
+
+    STEP_PART( &x[0], &CNS[ 0], &tmp[0] );
+    STEP_PART( &x[0], &CNS[ 2], &tmp[0] );
+    STEP_PART( &x[0], &CNS[ 4], &tmp[0] );
+    STEP_PART( &x[0], &CNS[ 6], &tmp[0] );
+    STEP_PART( &x[0], &CNS[ 8], &tmp[0] );
+    STEP_PART( &x[0], &CNS[10], &tmp[0] );
+    STEP_PART( &x[0], &CNS[12], &tmp[0] );
+    STEP_PART( &x[0], &CNS[14], &tmp[0] );
+
+    MIXTON1024( x[0], x[1], x[2], x[3],
+                chainv[0], chainv[2], chainv[4],chainv[6],
+                x[4], x[5], x[6], x[7],
+                chainv[1],chainv[3],chainv[5],chainv[7]);
+
+    /* Process last 256-bit block */
+    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[16], CNS[17],
+                tmp[0], tmp[1] );
+    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[18], CNS[19],
+                tmp[0], tmp[1] );
+    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[20], CNS[21],
+                tmp[0], tmp[1] );
+    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[22], CNS[23],
+                tmp[0], tmp[1] );
+    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[24], CNS[25],
+                tmp[0], tmp[1] );
+    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[26], CNS[27],
+                tmp[0], tmp[1] );
+    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[28], CNS[29],
+                tmp[0], tmp[1] );
+    STEP_PART2( chainv[8], chainv[9], t0, t1, CNS[30], CNS[31],
+                tmp[0], tmp[1] );
+}
+
+
+/***************************************************/
+/* Finalization function  */
+/* state: hash context    */
+/* b[8]: hash values      */
+
+void finalization512_2way( luffa_2way_context *state, uint32 *b )
+{
+    uint32 hash[8] __attribute((aligned(64)));
+    __m256i* chainv = state->chainv;
+    __m256i t[2];
+    __m256i zero[2];
+    zero[0] = zero[1] = _mm256_setzero_si256();
+
+    /*---- blank round with m=0 ----*/
+    rnd512_2way( state, zero );
+
+    t[0] = chainv[0];
+    t[1] = chainv[1];
+
+    t[0] = _mm256_xor_si256( t[0], chainv[2] );
+    t[1] = _mm256_xor_si256( t[1], chainv[3] );
+    t[0] = _mm256_xor_si256( t[0], chainv[4] );
+    t[1] = _mm256_xor_si256( t[1], chainv[5] );
+    t[0] = _mm256_xor_si256( t[0], chainv[6] );
+    t[1] = _mm256_xor_si256( t[1], chainv[7] );
+    t[0] = _mm256_xor_si256( t[0], chainv[8] );
+    t[1] = _mm256_xor_si256( t[1], chainv[9] );
+
+    t[0] = _mm256_shuffle_epi32( t[0], 27 );
+    t[1] = _mm256_shuffle_epi32( t[1], 27 );
+
+    _mm256_store_si256( (__m256i*)&hash[0], t[0] );
+    _mm256_store_si256( (__m256i*)&hash[8], t[1] );
+
+    casti_m256i( b, 0 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );
+    casti_m256i( b, 1 ) = mm256_bswap_32( casti_m256i( hash, 1 ) );
+
+    rnd512_2way( state, zero );
+
+    t[0] = chainv[0];
+    t[1] = chainv[1];
+    t[0] = _mm256_xor_si256( t[0], chainv[2] );
+    t[1] = _mm256_xor_si256( t[1], chainv[3] );
+    t[0] = _mm256_xor_si256( t[0], chainv[4] );
+    t[1] = _mm256_xor_si256( t[1], chainv[5] );
+    t[0] = _mm256_xor_si256( t[0], chainv[6] );
+    t[1] = _mm256_xor_si256( t[1], chainv[7] );
+    t[0] = _mm256_xor_si256( t[0], chainv[8] );
+    t[1] = _mm256_xor_si256( t[1], chainv[9] );
+
+    t[0] = _mm256_shuffle_epi32( t[0], 27 );
+    t[1] = _mm256_shuffle_epi32( t[1], 27 );
+
+    _mm256_store_si256( (__m256i*)&hash[0], t[0] );
+    _mm256_store_si256( (__m256i*)&hash[8], t[1] );
+
+    casti_m256i( b, 2 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );
+    casti_m256i( b, 3 ) = mm256_bswap_32( casti_m256i( hash, 1 ) );
+}
+
+int luffa_2way_init( luffa_2way_context *state, int hashbitlen )
+{
+    int i;
+    state->hashbitlen = hashbitlen;
+
+    for ( i=0; i<32; i++ ) CNS[i] =
+          _mm256_set_epi32( CNS_INIT[ (i<<2) + 3 ], CNS_INIT[ (i<<2) +2 ],
+                            CNS_INIT[ (i<<2) + 1 ], CNS_INIT[ (i<<2)    ],
+                            CNS_INIT[ (i<<2) + 3 ], CNS_INIT[ (i<<2) +2 ],
+                            CNS_INIT[ (i<<2) + 1 ], CNS_INIT[ (i<<2)    ] );
+
+    for ( i=0; i<10; i++ ) state->chainv[i] =
+          _mm256_set_epi32( IV[ (i<<2) +3 ], IV[ (i<<2) +2 ],
+                            IV[ (i<<2) +1 ], IV[ (i<<2)    ],
+                            IV[ (i<<2) +3 ], IV[ (i<<2) +2 ],
+                            IV[ (i<<2) +1 ], IV[ (i<<2)    ] );
+
+    ((__m256i*)state->buffer)[0] = m256_zero;
+    ((__m256i*)state->buffer)[1] = m256_zero;
+
+    return 0;
+}
+
+// Do not call luffa_update_close after having called luffa_update.
+// Once luffa_update has been called only call luffa_update or luffa_close.
+int luffa_2way_update( luffa_2way_context *state, const void *data,
+                       size_t len )
+{
+    __m256i *vdata  = (__m256i*)data;
+    __m256i *buffer = (__m256i*)state->buffer;
+    __m256i msg[2];
+    int i;
+    int blocks = (int)len >> 5;
+    state-> rembytes = (int)len & 0x1F;
+
+    // full blocks
+    for ( i = 0; i < blocks; i++, vdata+=2 )
+    {
+       msg[0] = mm256_bswap_32( vdata[ 0] );
+       msg[1] = mm256_bswap_32( vdata[ 1 ] );
+       rnd512_2way( state, msg );
+    }
+
+    // 16 byte partial block exists for 80 byte len
+    // store in buffer for transform in final for midstate to work
+    if ( state->rembytes  )
+    {
+      // remaining data bytes
+      buffer[0] = mm256_bswap_32( vdata[0] );
+      buffer[1] = _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0,
+                                   0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
+    }
+    return 0;
+}
+
+int luffa_2way_close( luffa_2way_context *state, void *hashval )
+{
+    __m256i *buffer = (__m256i*)state->buffer;
+    __m256i msg[2];
+
+    // transform pad block
+    if ( state->rembytes )
+      // not empty, data is in buffer
+      rnd512_2way( state, buffer );
+    else
+    {     // empty pad block, constant data
+      msg[0] = _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0,
+                                0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
+      msg[1] = m256_zero;
+      rnd512_2way( state, msg );
+    }
+    finalization512_2way( state, (uint32*)hashval );
+
+    if ( state->hashbitlen > 512 )
+        finalization512_2way( state, (uint32*)( hashval+32 ) );
+    return 0;
+}
+
+int luffa_2way_update_close( luffa_2way_context *state,
+                 void *output, const void *data, size_t inlen )
+{
+// Optimized for integrals of 16 bytes, good for 64 and 80 byte len
+    const __m256i *vdata  = (__m256i*)data;
+    __m256i msg[2];
+    int i;
+    const int blocks = (int)( inlen >> 5 );
+    state->rembytes = inlen & 0x1F;
+
+    // full blocks
+    for ( i = 0; i < blocks; i++, vdata+=2 )
+    {
+       msg[0] = mm256_bswap_32( vdata[ 0 ] );
+       msg[1] = mm256_bswap_32( vdata[ 1 ] );
+       rnd512_2way( state, msg );
+    }
+
+    // 16 byte partial block exists for 80 byte len
+    if ( state->rembytes  )
+    {
+       // padding of partial block
+       msg[0] = mm256_bswap_32( vdata[0] );
+       msg[1] = _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0,
+                                 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
+       rnd512_2way( state, msg );
+    }
+    else
+    {
+       // empty pad block
+       msg[0] = _mm256_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0,
+                                 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
+       msg[1] = m256_zero;
+       rnd512_2way( state, msg );
+    }
+
+    finalization512_2way( state, (uint32*)output );
+    if ( state->hashbitlen > 512 )
+        finalization512_2way( state, (uint32*)( output+32 ) );
+
+    return 0;
+}
+
+#endif
--- a/algo/luffa/luffa-hash-2way.h
+++ b/algo/luffa/luffa-hash-2way.h
@@ -0,0 +1,69 @@
+#if !defined(LUFFA_HASH_2WAY_H__)
+#define LUFFA_HASH_2WAY_H__ 1
+/*
+ * luffa_for_sse2.h
+ * Version 2.0 (Sep 15th 2009)
+ *
+ * Copyright (C) 2008-2009 Hitachi, Ltd. All rights reserved.
+ *
+ * Hitachi, Ltd. is the owner of this software and hereby grant
+ * the U.S. Government and any interested party the right to use
+ * this software for the purposes of the SHA-3 evaluation process,
+ * notwithstanding that this software is copyrighted.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#include "algo/sha/sha3-defs.h"
+#include "avxdefs.h"
+
+/* The length of digests*/
+#define DIGEST_BIT_LEN_224 224
+#define DIGEST_BIT_LEN_256 256
+#define DIGEST_BIT_LEN_384 384
+#define DIGEST_BIT_LEN_512 512
+
+/*********************************/
+/* The parameters of Luffa       */
+#define MSG_BLOCK_BIT_LEN 256  /*The bit length of a message block*/
+#define MSG_BLOCK_BYTE_LEN (MSG_BLOCK_BIT_LEN >> 3) /* The byte length
+                                                     * of a message block*/
+
+/* The number of blocks in Luffa */
+#define WIDTH_224 3
+#define WIDTH_256 3
+#define WIDTH_384 4
+#define WIDTH_512 5
+
+/* The limit of the length of message */
+#define LIMIT_224 64
+#define LIMIT_256 64
+#define LIMIT_384 128
+#define LIMIT_512 128
+/*********************************/
+
+typedef struct {
+    uint32 buffer[8*2] __attribute((aligned(64)));
+    __m256i chainv[10] __attribute((aligned(32)));   /* Chaining values */
+    int hashbitlen;
+    int rembytes;
+} luffa_2way_context;
+
+int luffa_2way_init( luffa_2way_context *state, int hashbitlen );
+int luffa_2way_update( luffa_2way_context *state, const void *data,
+                       size_t len );
+int luffa_2way_close( luffa_2way_context *state, void *hashval );
+int luffa_2way_update_close( luffa_2way_context *state, void *output,
+                                   const void *data, size_t inlen );
+
+#endif
+#endif
--- a/algo/luffa/sse2/luffa_for_sse2.c
+++ b/algo/luffa/sse2/luffa_for_sse2.c
@@ -272,8 +272,8 @@ HashReturn update_luffa( hashState_luffa *state, const BitSequence *data,
    // full blocks
    for ( i = 0; i < blocks; i++ )
    {
-       rnd512( state, mm_byteswap_32( casti_m128i( data, 1 ) ),
-                      mm_byteswap_32( casti_m128i( data, 0 ) ) );
+       rnd512( state, mm_bswap_32( casti_m128i( data, 1 ) ),
+                      mm_bswap_32( casti_m128i( data, 0 ) ) );
       data += MSG_BLOCK_BYTE_LEN;
    }

@@ -282,7 +282,7 @@ HashReturn update_luffa( hashState_luffa *state, const BitSequence *data,
    if ( state->rembytes  )
    {
      // remaining data bytes
-      casti_m128i( state->buffer, 0 ) = mm_byteswap_32( cast_m128i( data ) );
+      casti_m128i( state->buffer, 0 ) = mm_bswap_32( cast_m128i( data ) );
      // padding of partial block
      casti_m128i( state->buffer, 1 ) =
            _mm_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 );
@@ -324,8 +324,8 @@ HashReturn update_and_final_luffa( hashState_luffa *state, BitSequence* output,
    // full blocks
    for ( i = 0; i < blocks; i++ )
    {
-       rnd512( state, mm_byteswap_32( casti_m128i( data, 1 ) ),
-                      mm_byteswap_32( casti_m128i( data, 0 ) ) );
+       rnd512( state, mm_bswap_32( casti_m128i( data, 1 ) ),
+                      mm_bswap_32( casti_m128i( data, 0 ) ) );
       data += MSG_BLOCK_BYTE_LEN;
    }

@@ -334,7 +334,7 @@ HashReturn update_and_final_luffa( hashState_luffa *state, BitSequence* output,
    {
      // padding of partial block
      rnd512( state, _mm_set_epi8( 0,0,0,0, 0,0,0,0, 0,0,0,0, 0x80,0,0,0 ),
-                      mm_byteswap_32( cast_m128i( data ) ) );
+                      mm_bswap_32( cast_m128i( data ) ) );
    }
    else
    {
@@ -542,7 +542,7 @@ static void finalization512( hashState_luffa *state, uint32 *b )

    _mm256_store_si256( (__m256i*)hash, t );

-    casti_m256i( b, 0 ) = mm256_byteswap_32( casti_m256i( hash, 0 ) );
+    casti_m256i( b, 0 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );

    rnd512( state, zero, zero );

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

    _mm256_store_si256( (__m256i*)hash, t );

-    casti_m256i( b, 1 ) = mm256_byteswap_32( casti_m256i( hash, 0 ) );
+    casti_m256i( b, 1 ) = mm256_bswap_32( casti_m256i( hash, 0 ) );
 }

 #else
@@ -587,8 +587,8 @@ static void finalization512( hashState_luffa *state, uint32 *b )
    _mm_store_si128((__m128i*)&hash[0], t[0]);
    _mm_store_si128((__m128i*)&hash[4], t[1]);

-    casti_m128i( b, 0 ) = mm_byteswap_32( casti_m128i( hash, 0 ) );
-    casti_m128i( b, 1 ) = mm_byteswap_32( casti_m128i( hash, 1 ) );
+    casti_m128i( b, 0 ) = mm_bswap_32( casti_m128i( hash, 0 ) );
+    casti_m128i( b, 1 ) = mm_bswap_32( casti_m128i( hash, 1 ) );

    rnd512( state, zero, zero );

@@ -609,8 +609,8 @@ static void finalization512( hashState_luffa *state, uint32 *b )
    _mm_store_si128((__m128i*)&hash[0], t[0]);
    _mm_store_si128((__m128i*)&hash[4], t[1]);

-    casti_m128i( b, 2 ) = mm_byteswap_32( casti_m128i( hash, 0 ) );
-    casti_m128i( b, 3 ) = mm_byteswap_32( casti_m128i( hash, 1 ) );
+    casti_m128i( b, 2 ) = mm_bswap_32( casti_m128i( hash, 0 ) );
+    casti_m128i( b, 3 ) = mm_bswap_32( casti_m128i( hash, 1 ) );
 }
 #endif

--- a/algo/luffa/sse2/luffa_for_sse2.h
+++ b/algo/luffa/sse2/luffa_for_sse2.h
--- a/algo/lyra2/allium-4way.c
+++ b/algo/lyra2/allium-4way.c
@@ -0,0 +1,139 @@
+#include "allium-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/sse2/cubehash_sse2.h"
+#include "algo/groestl/aes_ni/hash-groestl256.h"
+
+typedef struct {
+   blake256_4way_context     blake;
+   keccak256_4way_context    keccak;
+   cubehashParam             cube;
+   skein256_4way_context     skein;
+   hashState_groestl256      groestl;
+
+} allium_4way_ctx_holder;
+
+static __thread allium_4way_ctx_holder allium_4way_ctx;
+
+bool init_allium_4way_ctx()
+{
+   keccak256_4way_init( &allium_4way_ctx.keccak );
+   cubehashInit( &allium_4way_ctx.cube, 256, 16, 32 );
+   skein256_4way_init( &allium_4way_ctx.skein );
+   init_groestl256( &allium_4way_ctx.groestl, 32 );
+   return true;
+}
+
+void allium_4way_hash( void *state, const void *input )
+{
+   uint32_t hash0[8] __attribute__ ((aligned (64)));
+   uint32_t hash1[8] __attribute__ ((aligned (32)));
+   uint32_t hash2[8] __attribute__ ((aligned (32)));
+   uint32_t hash3[8] __attribute__ ((aligned (32)));
+   uint32_t vhash32[8*4] __attribute__ ((aligned (64)));
+   uint32_t vhash64[8*4] __attribute__ ((aligned (64)));
+   allium_4way_ctx_holder ctx __attribute__ ((aligned (64))); 
+
+   memcpy( &ctx, &allium_4way_ctx, sizeof(allium_4way_ctx) );
+   blake256_4way( &ctx.blake, input + (64<<2), 16 );
+   blake256_4way_close( &ctx.blake, vhash32 );
+
+   mm256_reinterleave_4x64( vhash64, vhash32, 256 );
+   keccak256_4way( &ctx.keccak, vhash64, 32 );
+   keccak256_4way_close( &ctx.keccak, vhash64 );
+   mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+
+   LYRA2RE( hash0, 32, hash0, 32, hash0, 32, 1, 8, 8 );
+   LYRA2RE( hash1, 32, hash1, 32, hash1, 32, 1, 8, 8 );
+   LYRA2RE( hash2, 32, hash2, 32, hash2, 32, 1, 8, 8 );
+   LYRA2RE( hash3, 32, hash3, 32, hash3, 32, 1, 8, 8 );
+
+   cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*)hash0, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*)hash1, (const byte*)hash1, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*)hash2, (const byte*)hash2, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*)hash3, (const byte*)hash3, 32 );
+
+   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 );
+
+   mm256_interleave_4x64( vhash64, hash0, hash1, hash2, hash3, 256 );
+   skein256_4way( &ctx.skein, vhash64, 32 );
+   skein256_4way_close( &ctx.skein, vhash64 );
+   mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+
+   update_and_final_groestl256( &ctx.groestl, hash0, hash0, 256 );
+   memcpy( &ctx.groestl, &allium_4way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, hash1, hash1, 256 );
+   memcpy( &ctx.groestl, &allium_4way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, hash2, hash2, 256 );
+   memcpy( &ctx.groestl, &allium_4way_ctx.groestl,
+           sizeof(hashState_groestl256) );
+   update_and_final_groestl256( &ctx.groestl, hash3, hash3, 256 );
+
+   memcpy( state,    hash0, 32 );
+   memcpy( state+32, hash1, 32 );
+   memcpy( state+64, hash2, 32 );
+   memcpy( state+96, hash3, 32 );
+}
+
+int scanhash_allium_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                             uint64_t *hashes_done )
+{
+   uint32_t hash[8*4] __attribute__ ((aligned (64)));
+   uint32_t vdata[20*4] __attribute__ ((aligned (64)));
+   uint32_t _ALIGN(64) edata[20];
+   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 Htarg = ptarget[7];
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   uint32_t *noncep = vdata + 76; // 19*4
+
+   if ( opt_benchmark )
+      ( (uint32_t*)ptarget )[7] = 0x0000ff;
+
+   swab32_array( edata, pdata, 20 );
+   mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
+   blake256_4way_init( &allium_4way_ctx.blake );
+   blake256_4way( &allium_4way_ctx.blake, vdata, 64 );
+
+   do {
+     be32enc( noncep,   n   );
+     be32enc( noncep+1, n+1 );
+     be32enc( noncep+2, n+2 );
+     be32enc( noncep+3, n+3 );
+
+     allium_4way_hash( hash, vdata );
+     pdata[19] = n;
+
+     for ( int i = 0; i < 4; i++ )
+     if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
+     {
+         pdata[19] = n+i;
+         nonces[ num_found++ ] = n+i;
+         work_set_target_ratio( work, hash+(i<<3) );
+     }
+     n += 4;
+   } while ( (num_found == 0) && (n < max_nonce-4)
+                   && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#endif
--- a/algo/lyra2/allium-gate.c
+++ b/algo/lyra2/allium-gate.c
@@ -0,0 +1,22 @@
+#include "allium-gate.h"
+
+int64_t get_max64_0xFFFFLL() { return 0xFFFFLL; }
+
+bool register_allium_algo( algo_gate_t* gate )
+{
+#if defined (ALLIUM_4WAY)
+  gate->miner_thread_init = (void*)&init_allium_4way_ctx;
+  gate->scanhash  = (void*)&scanhash_allium_4way;
+  gate->hash      = (void*)&allium_4way_hash;
+#else
+  gate->miner_thread_init = (void*)&init_allium_ctx;
+  gate->scanhash  = (void*)&scanhash_allium;
+  gate->hash      = (void*)&allium_hash;
+#endif
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX_OPT | AVX2_OPT;
+  gate->set_target        = (void*)&alt_set_target;
+  gate->get_max64         = (void*)&get_max64_0xFFFFLL;
+  return true;
+};
+
+
--- a/algo/lyra2/allium-gate.h
+++ b/algo/lyra2/allium-gate.h
@@ -0,0 +1,29 @@
+#ifndef ALLIUM_GATE_H__
+#define ALLIUM_GATE_H__ 1
+
+#include "algo-gate-api.h"
+#include <stdint.h>
+#include "lyra2.h"
+
+#if defined(__AVX2__) && defined(__AES__)
+  #define ALLIUM_4WAY
+#endif
+
+bool register_allium_algo( algo_gate_t* gate );
+
+#if defined(ALLIUM_4WAY)
+
+void allium_4way_hash( void *state, const void *input );
+int scanhash_allium_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done );
+bool init_allium_4way_ctx();
+
+#endif
+
+void allium_hash( void *state, const void *input );
+int scanhash_allium( int thr_id, struct work *work, uint32_t max_nonce,
+                     uint64_t *hashes_done );
+bool init_allium_ctx();
+
+#endif
+
--- a/algo/lyra2/allium.c
+++ b/algo/lyra2/allium.c
@@ -0,0 +1,112 @@
+#include "allium-gate.h"
+#include <memory.h>
+#include "algo/blake/sph_blake.h"
+#include "algo/keccak/sph_keccak.h"
+#include "algo/skein/sph_skein.h"
+#include "algo/cubehash/sse2/cubehash_sse2.h" 
+#if defined(__AES__)
+#include "algo/groestl/aes_ni/hash-groestl256.h"
+#else
+#include "algo/groestl/sph_groestl.h"
+#endif
+#include "lyra2.h"
+
+typedef struct {
+        sph_blake256_context     blake;
+        sph_keccak256_context    keccak;
+        cubehashParam            cube;
+        sph_skein256_context     skein;
+#if defined (__AES__)
+        hashState_groestl256     groestl;
+#else
+        sph_groestl256_context   groestl;
+#endif
+} allium_ctx_holder;
+
+static __thread allium_ctx_holder allium_ctx;
+
+bool init_allium_ctx()
+{
+        sph_keccak256_init( &allium_ctx.keccak );
+        cubehashInit( &allium_ctx.cube, 256, 16, 32 );
+        sph_skein256_init( &allium_ctx.skein );
+#if defined (__AES__)
+        init_groestl256( &allium_ctx.groestl, 32 );
+#else
+        sph_groestl256_init( &allium_ctx.groestl );
+#endif
+        return true;
+}
+
+void allium_hash(void *state, const void *input)
+{
+    uint32_t hash[8] __attribute__ ((aligned (64)));
+    allium_ctx_holder ctx __attribute__ ((aligned (32)));
+
+    memcpy( &ctx, &allium_ctx, sizeof(allium_ctx) );
+    sph_blake256( &ctx.blake, input + 64, 16 );
+    sph_blake256_close( &ctx.blake, hash );
+
+    sph_keccak256( &ctx.keccak, hash, 32 );
+    sph_keccak256_close( &ctx.keccak, hash );
+
+    LYRA2RE( hash, 32, hash, 32, hash, 32, 1, 8, 8 );
+
+    cubehashUpdateDigest( &ctx.cube, (byte*)hash, (const byte*)hash, 32 );
+
+    LYRA2RE( hash, 32, hash, 32, hash, 32, 1, 8, 8 );
+
+    sph_skein256( &ctx.skein, hash, 32 );
+    sph_skein256_close( &ctx.skein, hash );
+
+#if defined (__AES__)
+   update_and_final_groestl256( &ctx.groestl, hash, hash, 256 );
+#else
+   sph_groestl256( &ctx.groestl, hash, 32 );
+   sph_groestl256_close( &ctx.groestl, hash );
+#endif
+
+    memcpy(state, hash, 32);
+}
+
+int scanhash_allium( int thr_id, struct work *work, uint32_t max_nonce,
+                     uint64_t *hashes_done )
+{
+    uint32_t _ALIGN(128) hash[8];
+    uint32_t _ALIGN(128) endiandata[20];
+    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 nonce = first_nonce;
+
+    if ( opt_benchmark )
+        ptarget[7] = 0x3ffff;
+
+    for ( int i = 0; i < 19; i++ )
+        be32enc( &endiandata[i], pdata[i] );
+
+    sph_blake256_init( &allium_ctx.blake );
+    sph_blake256( &allium_ctx.blake, endiandata, 64 );
+
+    do {
+        be32enc( &endiandata[19], nonce );
+        allium_hash( hash, endiandata );
+
+        if ( hash[7] <= Htarg && fulltest( hash, ptarget ) )
+        {
+            work_set_target_ratio( work, hash );
+            pdata[19] = nonce;
+            *hashes_done = pdata[19] - first_nonce;
+            return 1;
+        }
+        nonce++;
+
+    } while (nonce < max_nonce && !work_restart[thr_id].restart);
+
+    pdata[19] = nonce;
+    *hashes_done = pdata[19] - first_nonce + 1;
+    return 0;
+}
+
--- a/algo/lyra2/lyra2.c
+++ b/algo/lyra2/lyra2.c
@@ -47,8 +47,9 @@
 */

 int LYRA2REV2( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
-               uint64_t pwdlen, const void *salt, uint64_t saltlen,
-               uint64_t timeCost, const uint64_t nRows, const uint64_t nCols )
+               const uint64_t pwdlen, const void *salt, const uint64_t saltlen,
+               const uint64_t timeCost, const uint64_t nRows,
+               const uint64_t nCols )
 {
   //====================== Basic variables ============================//
   uint64_t _ALIGN(256) state[16];
@@ -73,6 +74,8 @@ int LYRA2REV2( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
                                          : BLOCK_LEN_BLAKE2_SAFE_BYTES;
   uint64_t *ptrWord = wholeMatrix;

+//   memset( wholeMatrix, 0, ROW_LEN_BYTES * nRows );
+
   //=== Getting the password + salt + basil padded with 10*1 ==========//
   //OBS.:The memory matrix will temporarily hold the password: not for saving memory,
   //but this ensures that the password copied locally will be overwritten as soon as possible
@@ -209,8 +212,9 @@ int LYRA2REV2( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
 }

 int LYRA2Z( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
-            uint64_t pwdlen, const void *salt, uint64_t saltlen,
-            uint64_t timeCost, uint64_t nRows, uint64_t nCols )
+            const uint64_t pwdlen, const void *salt, const uint64_t saltlen,
+            const uint64_t timeCost, const uint64_t nRows,
+            const uint64_t nCols )
 {
    //========================== Basic variables ============================//
    uint64_t _ALIGN(256) state[16];
@@ -230,6 +234,8 @@ int LYRA2Z( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
    const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * nCols;
 //    const int64_t ROW_LEN_BYTES = ROW_LEN_INT64 * 8;

+//    memset( wholeMatrix, 0, ROW_LEN_BYTES * nRows );
+
    //==== Getting the password + salt + basil padded with 10*1 ============//
    //OBS.:The memory matrix will temporarily hold the password: not for saving memory,
    //but this ensures that the password copied locally will be overwritten as soon as possible
@@ -347,9 +353,9 @@ int LYRA2Z( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
 }

 // Lyra2RE doesn't like the new wholeMatrix implementation
-int LYRA2RE( void *K, uint64_t kLen, const void *pwd,
-             uint64_t pwdlen, const void *salt, uint64_t saltlen,
-             uint64_t timeCost, const uint64_t nRows, const uint64_t nCols )
+int LYRA2RE( void *K, uint64_t kLen, const void *pwd, const uint64_t pwdlen,
+             const void *salt, const uint64_t saltlen, const uint64_t timeCost,
+             const uint64_t nRows, const uint64_t nCols )
 {
   //====================== Basic variables ============================//
   uint64_t _ALIGN(256) state[16];
@@ -377,15 +383,15 @@ int LYRA2RE( void *K, uint64_t kLen, const void *pwd,
   uint64_t *wholeMatrix = _mm_malloc( i, 64 );
   if (wholeMatrix == NULL)
      return -1;
-/*
-#if defined (__AVX2__)
-   memset_zero_m256i( (__m256i*)wholeMatrix, i/32 );
+
+#if defined(__AVX2__)
+   memset_zero_256( (__m256i*)wholeMatrix, i>>5 );
 #elif defined(__AVX__)
-   memset_zero_m128i( (__m128i*)wholeMatrix, i/16 );
+   memset_zero_128( (__m128i*)wholeMatrix, i>>4 );   
 #else
-   memset(wholeMatrix, 0, i);
+   memset( wholeMatrix, 0, i );
 #endif
-*/
+
   uint64_t *ptrWord = wholeMatrix;

   //=== Getting the password + salt + basil padded with 10*1 ==========//
@@ -406,8 +412,8 @@ int LYRA2RE( void *K, uint64_t kLen, const void *pwd,
   memcpy(ptrByte, salt, saltlen);
   ptrByte += saltlen;

-   memset( ptrByte, 0, nBlocksInput * BLOCK_LEN_BLAKE2_SAFE_BYTES
-                       - (saltlen + pwdlen) );
+//   memset( ptrByte, 0, nBlocksInput * BLOCK_LEN_BLAKE2_SAFE_BYTES
+//                       - (saltlen + pwdlen) );

   //Concatenates the basil: every integer passed as parameter, in the order they are provided by the interface
   memcpy(ptrByte, &kLen, sizeof(int64_t));
--- a/algo/lyra2/lyra2.h
+++ b/algo/lyra2/lyra2.h
@@ -54,4 +54,6 @@ int LYRA2Z( uint64_t*, void *K, uint64_t kLen, const void *pwd,
            uint64_t pwdlen, const void *salt, uint64_t saltlen,
            uint64_t timeCost, uint64_t nRows, uint64_t nCols );

+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);
+
 #endif /* LYRA2_H_ */
--- a/algo/lyra2/lyra2h-4way.c
+++ b/algo/lyra2/lyra2h-4way.c
@@ -0,0 +1,102 @@
+#include "lyra2h-gate.h"
+
+#ifdef LYRA2H_4WAY
+
+#include <memory.h>
+#include <mm_malloc.h>
+#include "lyra2.h"
+#include "algo/blake/sph_blake.h"
+#include "algo/blake/blake-hash-4way.h"
+
+__thread uint64_t* lyra2h_4way_matrix;
+
+bool lyra2h_4way_thread_init()
+{
+ return ( lyra2h_4way_matrix = _mm_malloc( LYRA2H_MATRIX_SIZE, 64 ) );
+}
+
+static __thread blake256_4way_context l2h_4way_blake_mid;
+
+void lyra2h_4way_midstate( const void* input )
+{
+       blake256_4way_init( &l2h_4way_blake_mid );
+       blake256_4way( &l2h_4way_blake_mid, input, 64 );
+}
+
+void lyra2h_4way_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 vhash[8*4] __attribute__ ((aligned (64)));
+     blake256_4way_context ctx_blake __attribute__ ((aligned (64)));
+
+     memcpy( &ctx_blake, &l2h_4way_blake_mid, sizeof l2h_4way_blake_mid );
+     blake256_4way( &ctx_blake, input + (64*4), 16 );
+     blake256_4way_close( &ctx_blake, vhash );
+
+     mm_deinterleave_4x32( hash0, hash1, hash2, hash3, vhash, 256 );
+
+     LYRA2Z( lyra2h_4way_matrix, hash0, 32, hash0, 32, hash0, 32, 16, 16, 16 );
+     LYRA2Z( lyra2h_4way_matrix, hash1, 32, hash1, 32, hash1, 32, 16, 16, 16 );
+     LYRA2Z( lyra2h_4way_matrix, hash2, 32, hash2, 32, hash2, 32, 16, 16, 16 );
+     LYRA2Z( lyra2h_4way_matrix, hash3, 32, hash3, 32, hash3, 32, 16, 16, 16 );
+
+     memcpy( state,    hash0, 32 );
+     memcpy( state+32, hash1, 32 );
+     memcpy( state+64, hash2, 32 );
+     memcpy( state+96, hash3, 32 );
+}
+
+int scanhash_lyra2h_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done )
+{
+   uint32_t hash[8*4] __attribute__ ((aligned (64)));
+   uint32_t vdata[20*4] __attribute__ ((aligned (64)));
+   uint32_t _ALIGN(64) edata[20];
+   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;
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   uint32_t *noncep= vdata + 76; // 19*4
+
+   if ( opt_benchmark )
+      ptarget[7] = 0x0000ff;
+
+   for ( int i=0; i < 19; i++ )
+      be32enc( &edata[i], pdata[i] );
+
+   mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
+
+   lyra2h_4way_midstate( vdata );
+
+   do {
+      be32enc( noncep,   n   );
+      be32enc( noncep+1, n+1 );
+      be32enc( noncep+2, n+2 );
+      be32enc( noncep+3, n+3 );
+
+      be32enc( &edata[19], n );
+      lyra2h_4way_hash( hash, vdata );
+
+      for ( int i = 0; i < 4; i++ )
+      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
+      {
+          pdata[19] = n+i;         
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+      }
+      n += 4;
+   } while ( (num_found == 0) && (n < max_nonce-4)
+                   && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#endif
+
--- a/algo/lyra2/lyra2h-gate.c
+++ b/algo/lyra2/lyra2h-gate.c
@@ -0,0 +1,25 @@
+#include "lyra2h-gate.h"
+#include "lyra2.h"
+
+void lyra2h_set_target( struct work* work, double job_diff )
+{
+ work_set_target( work, job_diff / (256.0 * opt_diff_factor) );
+}
+
+bool register_lyra2h_algo( algo_gate_t* gate )
+{
+#ifdef LYRA2H_4WAY
+  gate->miner_thread_init = (void*)&lyra2h_4way_thread_init;
+  gate->scanhash   = (void*)&scanhash_lyra2h_4way;
+  gate->hash       = (void*)&lyra2h_4way_hash;
+#else
+  gate->miner_thread_init = (void*)&lyra2h_thread_init;
+  gate->scanhash   = (void*)&scanhash_lyra2h;
+  gate->hash       = (void*)&lyra2h_hash;
+#endif
+  gate->optimizations = AVX_OPT | AVX2_OPT;
+  gate->get_max64  = (void*)&get_max64_0xffffLL;
+  gate->set_target = (void*)&lyra2h_set_target;
+  return true;
+};
+
--- a/algo/lyra2/lyra2h-gate.h
+++ b/algo/lyra2/lyra2h-gate.h
@@ -0,0 +1,32 @@
+#ifndef LYRA2H_GATE_H__
+#define LYRA2H_GATE_H__
+
+#include "algo-gate-api.h"
+#include <stdint.h>
+
+#if defined(__AVX2__)
+  #define LYRA2H_4WAY
+#endif
+
+#define LYRA2H_MATRIX_SIZE  BLOCK_LEN_INT64 * 16 * 16 * 8
+
+#if defined(LYRA2H_4WAY)
+
+void lyra2h_4way_hash( void *state, const void *input );
+
+int scanhash_lyra2h_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+
+bool lyra2h_4way_thread_init();
+
+#endif
+
+void lyra2h_hash( void *state, const void *input );
+
+int scanhash_lyra2h( int thr_id, struct work *work, uint32_t max_nonce,
+                    uint64_t *hashes_done );
+
+bool lyra2h_thread_init();
+
+#endif
+
--- a/algo/lyra2/lyra2h.c
+++ b/algo/lyra2/lyra2h.c
@@ -1,6 +1,6 @@
+#include "lyra2h-gate.h"
 #include <memory.h>
 #include <mm_malloc.h>
-#include "algo-gate-api.h"
 #include "lyra2.h"
 #include "algo/blake/sph_blake.h"

@@ -8,8 +8,7 @@ __thread uint64_t* lyra2h_matrix;

 bool lyra2h_thread_init()
 {
-   const int i = 16 * 16 * 96;
-   lyra2h_matrix = _mm_malloc( i, 64 );
+   lyra2h_matrix = _mm_malloc( LYRA2H_MATRIX_SIZE, 64 );
   return lyra2h_matrix;
 }

@@ -74,20 +73,3 @@ int scanhash_lyra2h( int thr_id, struct work *work, uint32_t max_nonce,
 	*hashes_done = pdata[19] - first_nonce + 1;
 	return 0;
 }
-
-void lyra2h_set_target( struct work* work, double job_diff )
-{
- work_set_target( work, job_diff / (256.0 * opt_diff_factor) );
-}
-
-bool register_lyra2h_algo( algo_gate_t* gate )
-{
-  gate->optimizations = AVX_OPT | AVX2_OPT;
-  gate->miner_thread_init = (void*)&lyra2h_thread_init;
-  gate->scanhash   = (void*)&scanhash_lyra2h;
-  gate->hash       = (void*)&lyra2h_hash;
-  gate->get_max64  = (void*)&get_max64_0xffffLL;
-  gate->set_target = (void*)&lyra2h_set_target;
-  return true;
-};
-
--- a/algo/lyra2/lyra2re.c
+++ b/algo/lyra2/lyra2re.c
@@ -106,6 +106,7 @@ int scanhash_lyra2re(int thr_id, struct work *work,
                   {
 			pdata[19] = nonce;
 			*hashes_done = pdata[19] - first_nonce;
+                        work_set_target_ratio( work, hash );
 			return 1;
                   }
 		}
--- a/algo/lyra2/lyra2rev2-4way.c
+++ b/algo/lyra2/lyra2rev2-4way.c
@@ -0,0 +1,133 @@
+#include "lyra2rev2-gate.h"
+#include <memory.h>
+
+#if defined (__AVX2__)	
+
+#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/sse2/cubehash_sse2.h" 
+
+typedef struct {
+   blake256_4way_context     blake;
+   keccak256_4way_context    keccak;
+   cubehashParam             cube;
+   skein256_4way_context     skein;
+   bmw256_4way_context          bmw;
+} lyra2v2_4way_ctx_holder;
+
+static lyra2v2_4way_ctx_holder l2v2_4way_ctx;
+
+bool init_lyra2rev2_4way_ctx()
+{
+   keccak256_4way_init( &l2v2_4way_ctx.keccak );
+   cubehashInit( &l2v2_4way_ctx.cube, 256, 16, 32 );
+   skein256_4way_init( &l2v2_4way_ctx.skein );
+   bmw256_4way_init( &l2v2_4way_ctx.bmw );
+   return true;
+}
+
+void lyra2rev2_4way_hash( void *state, const void *input )
+{
+   uint32_t hash0[8] __attribute__ ((aligned (64)));
+   uint32_t hash1[8] __attribute__ ((aligned (32)));
+   uint32_t hash2[8] __attribute__ ((aligned (32)));
+   uint32_t hash3[8] __attribute__ ((aligned (32)));
+   uint32_t vhash[8*4] __attribute__ ((aligned (64)));
+   uint64_t vhash64[4*4] __attribute__ ((aligned (64)));
+   lyra2v2_4way_ctx_holder ctx __attribute__ ((aligned (64))); 
+   memcpy( &ctx, &l2v2_4way_ctx, sizeof(l2v2_4way_ctx) );
+
+   blake256_4way( &ctx.blake, input + (64<<2), 16 );
+   blake256_4way_close( &ctx.blake, vhash );
+
+   mm256_reinterleave_4x64( vhash64, vhash, 256 );
+   keccak256_4way( &ctx.keccak, vhash64, 32 );
+   keccak256_4way_close( &ctx.keccak, vhash64 );
+   mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash0, (const byte*) hash0, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash1, (const byte*) hash1, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash2, (const byte*) hash2, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash3, (const byte*) hash3, 32 );
+
+   LYRA2REV2( l2v2_wholeMatrix, hash0, 32, hash0, 32, hash0, 32, 1, 4, 4 );
+   LYRA2REV2( l2v2_wholeMatrix, hash1, 32, hash1, 32, hash1, 32, 1, 4, 4 );
+   LYRA2REV2( l2v2_wholeMatrix, hash2, 32, hash2, 32, hash2, 32, 1, 4, 4 );
+   LYRA2REV2( l2v2_wholeMatrix, hash3, 32, hash3, 32, hash3, 32, 1, 4, 4 );
+
+   mm256_interleave_4x64( vhash64, hash0, hash1, hash2, hash3, 256 );
+   skein256_4way( &ctx.skein, vhash64, 32 );
+   skein256_4way_close( &ctx.skein, vhash64 );
+   mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash0, (const byte*) hash0, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash1, (const byte*) hash1, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash2, (const byte*) hash2, 32 );
+   cubehashReinit( &ctx.cube );
+   cubehashUpdateDigest( &ctx.cube, (byte*) hash3, (const byte*) hash3, 32 );
+
+   mm_interleave_4x32( vhash, hash0, hash1, hash2, hash3, 256 );
+   bmw256_4way( &ctx.bmw, vhash, 32 );
+   bmw256_4way_close( &ctx.bmw, vhash );
+
+   mm_deinterleave_4x32( state, state+32, state+64, state+96, vhash, 256 );
+}
+
+int scanhash_lyra2rev2_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                             uint64_t *hashes_done )
+{
+   uint32_t hash[8*4] __attribute__ ((aligned (64)));
+   uint32_t vdata[20*4] __attribute__ ((aligned (64)));
+   uint32_t _ALIGN(64) edata[20];
+   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 Htarg = ptarget[7];
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   uint32_t *noncep = vdata + 76; // 19*4
+
+   if ( opt_benchmark )
+      ( (uint32_t*)ptarget )[7] = 0x0000ff;
+
+   swab32_array( edata, pdata, 20 );
+
+   mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
+
+   blake256_4way_init( &l2v2_4way_ctx.blake );
+   blake256_4way( &l2v2_4way_ctx.blake, vdata, 64 );
+
+   do {
+      be32enc( noncep,   n   );
+      be32enc( noncep+1, n+1 );
+      be32enc( noncep+2, n+2 );
+      be32enc( noncep+3, n+3 );
+
+      lyra2rev2_4way_hash( hash, vdata );
+      pdata[19] = n;
+
+      for ( int i = 0; i < 4; i++ )
+      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
+      {
+          pdata[19] = n+i;         
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+      }
+      n += 4;
+   } while ( (num_found == 0) && (n < max_nonce-4)
+                   && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+#endif
--- a/algo/lyra2/lyra2rev2-gate.c
+++ b/algo/lyra2/lyra2rev2-gate.c
@@ -0,0 +1,40 @@
+#include "lyra2rev2-gate.h"
+
+__thread uint64_t* l2v2_wholeMatrix;
+
+void lyra2rev2_set_target( struct work* work, double job_diff )
+{
+ work_set_target( work, job_diff / (256.0 * opt_diff_factor) );
+}
+
+bool lyra2rev2_thread_init()
+{
+   const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * 4; // nCols
+   const int64_t ROW_LEN_BYTES = ROW_LEN_INT64 * 8;
+
+   int i = (int64_t)ROW_LEN_BYTES * 4; // nRows;
+   l2v2_wholeMatrix = _mm_malloc( i, 64 );
+#if defined (LYRA2REV2_4WAY)
+   init_lyra2rev2_4way_ctx();;
+#else
+   init_lyra2rev2_ctx();
+#endif
+   return l2v2_wholeMatrix;
+}
+
+bool register_lyra2rev2_algo( algo_gate_t* gate )
+{
+#if 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 | AVX_OPT | AVX2_OPT;
+  gate->miner_thread_init = (void*)&lyra2rev2_thread_init;
+  gate->set_target        = (void*)&lyra2rev2_set_target;
+  return true;
+};
+
+
--- a/algo/lyra2/lyra2rev2-gate.h
+++ b/algo/lyra2/lyra2rev2-gate.h
@@ -0,0 +1,35 @@
+#ifndef LYRA2REV2_GATE_H__
+#define LYRA2REV2_GATE_H__ 1
+
+#include "algo-gate-api.h"
+#include <stdint.h>
+#include "lyra2.h"
+
+#if defined(__AVX2__)
+  #define LYRA2REV2_4WAY
+#endif
+
+extern __thread uint64_t* l2v2_wholeMatrix;
+
+bool register_lyra2rev2_algo( algo_gate_t* gate );
+
+#if defined(LYRA2REV2_4WAY)
+
+void lyra2rev2_4way_hash( void *state, const void *input );
+
+int scanhash_lyra2rev2_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+
+bool init_lyra2rev2_4way_ctx();
+
+#endif
+
+void lyra2rev2_hash( void *state, const void *input );
+
+int scanhash_lyra2rev2( int thr_id, struct work *work, uint32_t max_nonce,
+                    uint64_t *hashes_done );
+
+bool init_lyra2rev2_ctx();
+
+#endif
+
--- a/algo/lyra2/lyra2rev2.c
+++ b/algo/lyra2/lyra2rev2.c
@@ -1,20 +1,12 @@
+#include "lyra2rev2-gate.h"
 #include <memory.h>
-
-#include "algo-gate-api.h"
-
 #include "algo/blake/sph_blake.h"
 #include "algo/cubehash/sph_cubehash.h"
 #include "algo/keccak/sph_keccak.h"
 #include "algo/skein/sph_skein.h"
 #include "algo/bmw/sph_bmw.h"
 #include "algo/cubehash/sse2/cubehash_sse2.h" 
-#include "lyra2.h"
-#include "avxdefs.h"
-
-// This gets allocated when miner_thread starts up and is never freed.
-// It's not a leak because the only way to allocate it again is to exit
-// the thread and that only occurs when the entire program exits.
-__thread uint64_t* l2v2_wholeMatrix;
+//#include "lyra2.h"

 typedef struct {
        cubehashParam           cube1;
@@ -29,7 +21,7 @@ typedef struct {
 static lyra2v2_ctx_holder lyra2v2_ctx;
 static __thread sph_blake256_context l2v2_blake_mid;

-void init_lyra2rev2_ctx()
+bool init_lyra2rev2_ctx()
 {
        cubehashInit( &lyra2v2_ctx.cube1, 256, 16, 32 );
        cubehashInit( &lyra2v2_ctx.cube2, 256, 16, 32 );
@@ -37,6 +29,7 @@ void init_lyra2rev2_ctx()
        sph_keccak256_init( &lyra2v2_ctx.keccak );
        sph_skein256_init( &lyra2v2_ctx.skein );
        sph_bmw256_init( &lyra2v2_ctx.bmw );
+        return true;
 }

 void l2v2_blake256_midstate( const void* input )
@@ -106,6 +99,7 @@ int scanhash_lyra2rev2(int thr_id, struct work *work,
                   if( fulltest(hash, ptarget) )
                   {
 			pdata[19] = nonce;
+                        work_set_target_ratio( work, hash );
 			*hashes_done = pdata[19] - first_nonce;
 		   	return 1;
 		   }
@@ -119,30 +113,3 @@ int scanhash_lyra2rev2(int thr_id, struct work *work,
 	return 0;
 }

-void lyra2rev2_set_target( struct work* work, double job_diff )
-{
- work_set_target( work, job_diff / (256.0 * opt_diff_factor) );
-}
-
-bool lyra2rev2_thread_init()
-{
-   const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * 4; // nCols
-   const int64_t ROW_LEN_BYTES = ROW_LEN_INT64 * 8;
-
-   int i = (int64_t)ROW_LEN_BYTES * 4; // nRows;
-   l2v2_wholeMatrix = _mm_malloc( i, 64 );
-
-   return l2v2_wholeMatrix;
-}
-
-bool register_lyra2rev2_algo( algo_gate_t* gate )
-{
-  init_lyra2rev2_ctx();
-  gate->optimizations = AVX_OPT | AVX2_OPT;
-  gate->miner_thread_init = (void*)&lyra2rev2_thread_init;
-  gate->scanhash          = (void*)&scanhash_lyra2rev2;
-  gate->hash              = (void*)&lyra2rev2_hash;
-  gate->set_target        = (void*)&lyra2rev2_set_target;
-  return true;
-};
-
--- a/algo/lyra2/lyra2z-4way.c
+++ b/algo/lyra2/lyra2z-4way.c
@@ -61,12 +61,8 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
   const uint32_t first_nonce = pdata[19];
   uint32_t n = first_nonce;
   uint32_t *nonces = work->nonces;
-   bool *found = work->nfound;
   int num_found = 0;
-   uint32_t *noncep0 = vdata + 76; // 19*4
-   uint32_t *noncep1 = vdata + 77;
-   uint32_t *noncep2 = vdata + 78;
-   uint32_t *noncep3 = vdata + 79;
+   uint32_t *noncep = vdata + 76; // 19*4

   if ( opt_benchmark )
      ptarget[7] = 0x0000ff;
@@ -79,42 +75,20 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
   lyra2z_4way_midstate( vdata );

   do {
-      found[0] = found[1] = found[2] = found[3] = false;
-      be32enc( noncep0, n   );
-      be32enc( noncep1, n+1 );
-      be32enc( noncep2, n+2 );
-      be32enc( noncep3, n+3 );
+      be32enc( noncep,   n   );
+      be32enc( noncep+1, n+1 );
+      be32enc( noncep+2, n+2 );
+      be32enc( noncep+3, n+3 );

-      be32enc( &edata[19], n );
      lyra2z_4way_hash( hash, vdata );
+      pdata[19] = n;

-      if ( hash[7] <= Htarg && fulltest( hash, ptarget ) )
+      for ( int i = 0; i < 4; i++ )
+      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
      {
-          found[0] = true;
-          num_found++;
-          nonces[0] = pdata[19] = n;
-          work_set_target_ratio( work, hash );
-      }
-      if ( (hash+8)[7] <= Htarg && fulltest( hash+8, ptarget ) )
-      {
-          found[1] = true;
-          num_found++;
-          nonces[1] = n+1;
-          work_set_target_ratio( work, hash+8 );
-      }
-      if ( (hash+16)[7] <= Htarg && fulltest( hash+16, ptarget ) )
-      {
-          found[2] = true;
-          num_found++;
-          nonces[2] = n+2;
-          work_set_target_ratio( work, hash+16 );
-      }
-      if ( (hash+24)[7] <= Htarg && fulltest( hash+24, ptarget ) )
-      {
-          found[3] = true;
-          num_found++;
-          nonces[3] = n+3;
-          work_set_target_ratio( work, hash+24 );
+          pdata[19] = n+i;         
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
      }
      n += 4;
   } while ( (num_found == 0) && (n < max_nonce-4)
@@ -126,3 +100,115 @@ int scanhash_lyra2z_4way( int thr_id, 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 );
+
+     mm256_deinterleave_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, hash1, 32 );
+     memcpy( state+160, hash2, 32 );
+     memcpy( state+192, hash3, 32 );
+     memcpy( state+224, hash1, 32 );
+}
+
+int scanhash_lyra2z_8way( int thr_id, struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done )
+{
+   uint32_t hash[8*8] __attribute__ ((aligned (64)));
+   uint32_t vdata[20*8] __attribute__ ((aligned (64)));
+   uint32_t _ALIGN(64) edata[20];
+   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;
+   uint32_t *nonces = work->nonces;
+   int num_found = 0;
+   uint32_t *noncep = vdata + 152; // 19*8
+
+   if ( opt_benchmark )
+      ptarget[7] = 0x0000ff;
+
+   for ( int i=0; i < 19; i++ )
+      be32enc( &edata[i], pdata[i] );
+
+   mm256_interleave_8x32( vdata, edata, edata, edata, edata,
+                                 edata, edata, edata, edata, 640 );
+
+   lyra2z_8way_midstate( vdata );
+
+   do {
+      be32enc( noncep,   n   );
+      be32enc( noncep+1, n+1 );
+      be32enc( noncep+2, n+2 );
+      be32enc( noncep+3, n+3 );
+      be32enc( noncep+4, n+4 );
+      be32enc( noncep+5, n+5 );
+      be32enc( noncep+6, n+6 );
+      be32enc( noncep+7, n+7 );
+
+      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 ) )
+      {
+          pdata[19] = n+i;         
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+      }
+      n += 8;
+   } while ( (num_found == 0) && (n < max_nonce-4)
+                   && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return num_found;
+}
+
+
+#endif
--- a/algo/lyra2/lyra2z-gate.c
+++ b/algo/lyra2/lyra2z-gate.c
@@ -8,7 +8,11 @@ void lyra2z_set_target( struct work* work, double job_diff )

 bool register_lyra2z_algo( algo_gate_t* gate )
 {
-#ifdef LYRA2Z_4WAY
+#if defined(LYRA2Z_8WAY)
+  gate->miner_thread_init = (void*)&lyra2z_8way_thread_init;
+  gate->scanhash   = (void*)&scanhash_lyra2z_8way;
+  gate->hash       = (void*)&lyra2z_8way_hash;
+#elif defined(LYRA2Z_4WAY)
  gate->miner_thread_init = (void*)&lyra2z_4way_thread_init;
  gate->scanhash   = (void*)&scanhash_lyra2z_4way;
  gate->hash       = (void*)&lyra2z_4way_hash;
@@ -17,7 +21,7 @@ bool register_lyra2z_algo( algo_gate_t* gate )
  gate->scanhash   = (void*)&scanhash_lyra2z;
  gate->hash       = (void*)&lyra2z_hash;
 #endif
-  gate->optimizations = AVX_OPT | AVX2_OPT | FOUR_WAY_OPT;
+  gate->optimizations = AVX_OPT | AVX2_OPT;
  gate->get_max64  = (void*)&get_max64_0xffffLL;
  gate->set_target = (void*)&lyra2z_set_target;
  return true;
--- a/algo/lyra2/lyra2z-gate.h
+++ b/algo/lyra2/lyra2z-gate.h
@@ -1,17 +1,29 @@
 #ifndef LYRA2Z_GATE_H__
-#define LYRA2Z_GATE_H__
+#define LYRA2Z_GATE_H__ 1

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

-#if defined(HASH_4WAY)
+#if defined(__AVX__)
  #define LYRA2Z_4WAY
 #endif
+#if defined(__AVX2__)
+//  #define LYRA2Z_8WAY
+#endif


 #define LYRA2Z_MATRIX_SIZE  BLOCK_LEN_INT64 * 8 * 8 * 8

-#if defined(LYRA2Z_4WAY)
+#if defined(LYRA2Z_8WAY)
+
+void lyra2z_8way_hash( void *state, const void *input );
+
+int scanhash_lyra2z_8way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+
+bool lyra2z_8way_thread_init();
+
+#elif defined(LYRA2Z_4WAY)

 void lyra2z_4way_hash( void *state, const void *input );

@@ -20,7 +32,7 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,

 bool lyra2z_4way_thread_init();

-#endif
+#else

 void lyra2z_hash( void *state, const void *input );

@@ -31,3 +43,4 @@ bool lyra2z_thread_init();

 #endif

+#endif
--- a/algo/lyra2/lyra2z.c
+++ b/algo/lyra2/lyra2z.c
@@ -82,41 +82,3 @@ int scanhash_lyra2z( int thr_id, struct work *work, uint32_t max_nonce,
 	return 0;
 }

-/*
-//int64_t get_max64_0xffffLL() { return 0xffffLL; };
-
-void lyra2z_set_target( struct work* work, double job_diff )
-{
- work_set_target( work, job_diff / (256.0 * opt_diff_factor) );
-}
-
-bool zcoin_get_work_height( struct work* work, struct stratum_ctx* sctx )
-{
-   work->height = sctx->bloc_height;
-   return false;
-}
-
-
-bool lyra2z_thread_init()
-{
-   const int64_t ROW_LEN_INT64 = BLOCK_LEN_INT64 * 8; // nCols
-   const int64_t ROW_LEN_BYTES = ROW_LEN_INT64 * 8;
-
-   int i = (int64_t)ROW_LEN_BYTES * 8; // nRows;
-   lyra2z_wholeMatrix = _mm_malloc( i, 64 );
-
-   return lyra2z_wholeMatrix;
-}
-
-bool register_lyra2z_algo( algo_gate_t* gate )
-{
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX_OPT | AVX2_OPT;
-  gate->miner_thread_init = (void*)&lyra2z_thread_init;
-  gate->scanhash   = (void*)&scanhash_lyra2z;
-  gate->hash       = (void*)&lyra2z_hash;
-  gate->get_max64  = (void*)&get_max64_0xffffLL;
-  gate->set_target = (void*)&lyra2z_set_target;
-//  gate->prevent_dupes = (void*)&zcoin_get_work_height;
-  return true;
-};
-*/
--- a/algo/m7m.c
+++ b/algo/m7m.c
@@ -346,6 +346,7 @@ int scanhash_m7m_hash( int thr_id, struct work* work,
                    hash_str,
                    target_str);
            }
+            work_set_target_ratio( work, hash );
            pdata[19] = data[19];
            goto out;
 	  }
--- a/algo/neoscrypt/neoscrypt.c
+++ b/algo/neoscrypt/neoscrypt.c
@@ -85,12 +85,12 @@ typedef unsigned int  uint;
    U32TO8_BE((p) + 4, (uint32_t)((v)      ));


-typedef uint8_t hash_digest[SCRYPT_HASH_DIGEST_SIZE];
+typedef uint8_t hash_digest[SCRYPT_HASH_DIGEST_SIZE] __attribute__ ((aligned (16)));


 /* SHA-256 */

-static const uint32_t sha256_constants[64] = {
+static const uint32_t sha256_constants[64] __attribute__ ((aligned (16))) = {
    0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
    0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
    0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
@@ -123,10 +123,10 @@ static const uint32_t sha256_constants[64] = {


 typedef struct sha256_hash_state_t {
-    uint32_t H[8];
+    uint32_t H[8] __attribute__ ((aligned (16)));
    uint64_t T;
    uint32_t leftover;
-    uint8_t buffer[SCRYPT_HASH_BLOCK_SIZE];
+    uint8_t buffer[SCRYPT_HASH_BLOCK_SIZE] __attribute__ ((aligned (16)));
 } sha256_hash_state;


@@ -242,7 +242,7 @@ typedef struct sha256_hmac_state_t {
 } sha256_hmac_state;

 static void neoscrypt_hmac_init_sha256(sha256_hmac_state *st, const uint8_t *key, size_t keylen) {
-    uint8_t pad[SCRYPT_HASH_BLOCK_SIZE] = {0};
+    uint8_t pad[SCRYPT_HASH_BLOCK_SIZE] __attribute__ ((aligned (16))) = {0};
    size_t i;

    neoscrypt_hash_init_sha256(&st->inner);
@@ -570,17 +570,17 @@ typedef struct blake2s_param_t {

 /* State block of 180 bytes */
 typedef struct blake2s_state_t {
-    uint  h[8];
+    uint  h[8] __attribute__ ((aligned (16)));
    uint  t[2];
    uint  f[2];
-    uchar buf[2 * BLAKE2S_BLOCK_SIZE];
+    uchar buf[2 * BLAKE2S_BLOCK_SIZE] __attribute__ ((aligned (16)));
    uint  buflen;
 } blake2s_state;

 static void blake2s_compress(blake2s_state *S, const void *buf) {
    uint i;
-    uint m[16];
-    uint v[16];
+    uint m[16] __attribute__ ((aligned (16)));
+    uint v[16] __attribute__ ((aligned (16)));

    neoscrypt_copy(m, buf, 64);
    neoscrypt_copy(v, S, 32);
@@ -1082,6 +1082,7 @@ void neoscrypt_wait_for_diff( struct stratum_ctx *stratum )

 bool register_neoscrypt_algo( algo_gate_t* gate )
 {
+  gate->optimizations         = SSE2_OPT;
  gate->scanhash              = (void*)&scanhash_neoscrypt;
  gate->hash                  = (void*)&neoscrypt;
  gate->get_max64             = (void*)&get_neoscrypt_max64;
--- a/algo/nist5/nist5-4way.c
+++ b/algo/nist5/nist5-4way.c
@@ -79,12 +79,8 @@ int scanhash_nist5_4way( int thr_id, struct work *work, uint32_t max_nonce,
     const uint32_t first_nonce = pdata[19];
     const uint32_t Htarg = ptarget[7];
     uint32_t *nonces = work->nonces;
-     bool *found = work->nfound;
     int num_found = 0;
-     uint32_t *noncep0 = vdata + 73;   // 9*8 + 1
-     uint32_t *noncep1 = vdata + 75;
-     uint32_t *noncep2 = vdata + 77;
-     uint32_t *noncep3 = vdata + 79;
+     uint32_t *noncep = vdata + 73;   // 9*8 + 1

     uint64_t htmax[] = {          0,
                                 0xF,
@@ -117,47 +113,22 @@ int scanhash_nist5_4way( int thr_id, struct work *work, uint32_t max_nonce,
           uint32_t mask = masks[m];

           do {
-              found[0] = found[1] = found[2] = found[3] = false;
-              be32enc( noncep0, n   );
-              be32enc( noncep1, n+1 );
-              be32enc( noncep2, n+2 );
-              be32enc( noncep3, n+3 );
+              be32enc( noncep,   n   );
+              be32enc( noncep+2, n+1 );
+              be32enc( noncep+4, n+2 );
+              be32enc( noncep+6, n+3 );

              nist5hash_4way( hash, vdata );

              pdata[19] = n;

-              if ( ( !(hash[7] & mask) )
-                   && fulltest( hash, ptarget ) ) 
+              for ( int i = 0; i < 4; i++ )
+              if ( ( !( (hash+(i<<3))[7] & mask ) == 0 )
+                 && fulltest( hash+(i<<3), ptarget ) )
              {
-                 found[0] = true;
-                 num_found++;
-                 nonces[0] = n; 
-                 work_set_target_ratio( work, hash );
-              }
-              if ( ( !((hash+8)[7] & mask) )
-                   && fulltest( hash+8, ptarget ) )
-              {
-                 found[1] = true;
-                 num_found++;
-                 nonces[1] = n+1;
-                 work_set_target_ratio( work, hash+8 );
-              }
-              if ( ( !((hash+16)[7] & mask) )
-                 && fulltest( hash+16, ptarget ) )
-              {
-                 found[2] = true;
-                 num_found++;
-                 nonces[2] = n+2;
-                 work_set_target_ratio( work, hash+16 );
-              }
-              if ( ( !((hash+24)[7] & mask) )
-                   && fulltest( hash+24, ptarget ) )
-              {
-                 found[3] = true;
-                 num_found++;
-                 nonces[3] = n+3;
-                 work_set_target_ratio( work, hash+24 );
+                 pdata[19] = n+i;         
+                 nonces[ num_found++ ] = n+i;
+                 work_set_target_ratio( work, hash+(i<<3) );
              }
              n += 4;
           } while ( ( num_found == 0 ) && ( n < max_nonce )
--- a/algo/nist5/nist5-gate.c
+++ b/algo/nist5/nist5-gate.c
@@ -2,7 +2,7 @@

 bool register_nist5_algo( algo_gate_t* gate )
 {
-    gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | FOUR_WAY_OPT;
+    gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
 #if defined (NIST5_4WAY)
    gate->scanhash = (void*)&scanhash_nist5_4way;
    gate->hash     = (void*)&nist5hash_4way;
--- a/algo/nist5/nist5-gate.h
+++ b/algo/nist5/nist5-gate.h
@@ -4,7 +4,7 @@
 #include "algo-gate-api.h"
 #include <stdint.h>

-#if defined(HASH_4WAY) && defined(__AES__)
+#if defined(__AVX2__) && defined(__AES__)
  #define NIST5_4WAY
 #endif

@@ -21,6 +21,7 @@ void nist5hash( void *state, const void *input );

 int scanhash_nist5( int thr_id, struct work *work, uint32_t max_nonce,
                    uint64_t *hashes_done );
+void init_nist5_ctx();
 #endif

 #endif
--- a/algo/nist5/nist5.c
+++ b/algo/nist5/nist5.c
@@ -132,6 +132,7 @@ int scanhash_nist5(int thr_id, struct work *work,
 				if (!(hash64[7] & mask)) {
 					printf("[%d]",thr_id);
 					if (fulltest(hash64, ptarget)) {
+                                                work_set_target_ratio( work, hash64 );
 						*hashes_done = n - first_nonce + 1;
 						return true;
 					}
--- a/algo/nist5/zr5.c
+++ b/algo/nist5/zr5.c
@@ -172,6 +172,7 @@ int scanhash_zr5( int thr_id, struct work *work,
         pdata[0] = tmpdata[0];
         pdata[19] = nonce;
         *hashes_done = pdata[19] - first_nonce + 1;
+         work_set_target_ratio( work, hash );
         if (opt_debug)
           applog(LOG_INFO, "found nonce %x", nonce);
         return 1;
--- a/algo/quark/anime-4way.c
+++ b/algo/quark/anime-4way.c
@@ -0,0 +1,206 @@
+#include "cpuminer-config.h"
+#include "anime-gate.h"
+
+#if defined (ANIME_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"
+#include "algo/jh/jh-hash-4way.h"
+#include "algo/keccak/keccak-hash-4way.h"
+#include "algo/groestl/aes_ni/hash-groestl.h"
+
+typedef struct {
+    blake512_4way_context  blake;
+    bmw512_4way_context    bmw;
+    hashState_groestl      groestl;
+    jh512_4way_context     jh;
+    skein512_4way_context  skein;
+    keccak512_4way_context keccak;
+} anime_4way_ctx_holder;
+
+anime_4way_ctx_holder anime_4way_ctx __attribute__ ((aligned (64)));
+
+void init_anime_4way_ctx()
+{
+     blake512_4way_init( &anime_4way_ctx.blake );
+     bmw512_4way_init( &anime_4way_ctx.bmw );
+     init_groestl( &anime_4way_ctx.groestl, 64 );
+     skein512_4way_init( &anime_4way_ctx.skein );
+     jh512_4way_init( &anime_4way_ctx.jh );
+     keccak512_4way_init( &anime_4way_ctx.keccak );
+}
+
+void anime_4way_hash( void *state, const void *input )
+{
+    uint64_t hash0[8] __attribute__ ((aligned (64)));
+    uint64_t hash1[8] __attribute__ ((aligned (64)));
+    uint64_t hash2[8] __attribute__ ((aligned (64)));
+    uint64_t hash3[8] __attribute__ ((aligned (64)));
+    uint64_t vhash[8*4] __attribute__ ((aligned (64)));
+    uint64_t vhashA[8*4] __attribute__ ((aligned (64)));
+    uint64_t vhashB[8*4] __attribute__ ((aligned (64)));
+    __m256i* vh  = (__m256i*)vhash;
+    __m256i* vhA = (__m256i*)vhashA;
+    __m256i* vhB = (__m256i*)vhashB;
+    __m256i vh_mask;
+    __m256i bit3_mask; bit3_mask = _mm256_set1_epi64x( 8 );
+    int i;
+    anime_4way_ctx_holder ctx;
+    memcpy( &ctx, &anime_4way_ctx, sizeof(anime_4way_ctx) );
+
+    bmw512_4way( &ctx.bmw, vhash, 80 );
+    bmw512_4way_close( &ctx.bmw, vhash );
+
+    blake512_4way( &ctx.blake, input, 64 );
+    blake512_4way_close( &ctx.blake, vhash );
+
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
+                                  m256_zero );
+
+       mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+       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 );
+       mm256_interleave_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
+
+       skein512_4way( &ctx.skein, vhash, 64 );
+       skein512_4way_close( &ctx.skein, vhashB );
+
+    for ( i = 0; i < 8; i++ )
+       vh[i] = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );
+
+    mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, 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 );
+    mm256_interleave_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+
+    jh512_4way( &ctx.jh, vhash, 64 );
+    jh512_4way_close( &ctx.jh, vhash );
+
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
+                                  m256_zero );
+
+       blake512_4way_init( &ctx.blake );
+       blake512_4way( &ctx.blake, vhash, 64 );
+       blake512_4way_close( &ctx.blake, vhashA );
+
+       bmw512_4way_init( &ctx.bmw );
+       bmw512_4way( &ctx.bmw, vhash, 64 );
+       bmw512_4way_close( &ctx.bmw, vhashB );
+
+    for ( i = 0; i < 8; i++ )
+       vh[i] = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );
+
+    keccak512_4way( &ctx.keccak, vhash, 64 );
+    keccak512_4way_close( &ctx.keccak, vhash );
+
+    skein512_4way_init( &ctx.skein );
+    skein512_4way( &ctx.skein, vhash, 64 );
+    skein512_4way_close( &ctx.skein, vhash );
+
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
+                                  m256_zero );
+
+       keccak512_4way_init( &ctx.keccak );
+       keccak512_4way( &ctx.keccak, vhash, 64 );
+       keccak512_4way_close( &ctx.keccak, vhashA );
+
+       jh512_4way_init( &ctx.jh );
+       jh512_4way( &ctx.jh, vhash, 64 );
+       jh512_4way_close( &ctx.jh, vhashB );
+
+    for ( i = 0; i < 8; i++ )
+       vh[i] = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );
+
+    mm256_deinterleave_4x64( state, state+32, state+64, state+96, vhash, 256 );
+}
+
+int scanhash_anime_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done)
+{
+    uint32_t hash[4*8] __attribute__ ((aligned (64)));
+    uint32_t vdata[24*4] __attribute__ ((aligned (64)));
+    uint32_t endiandata[20] __attribute__((aligned(64)));
+    uint32_t *pdata = work->data;
+    uint32_t *ptarget = work->target;
+    uint32_t n = pdata[19];
+    const uint32_t first_nonce = pdata[19];
+    uint32_t *nonces = work->nonces;
+    int num_found = 0;
+    uint32_t *noncep = vdata + 73;   // 9*8 + 1
+    const uint32_t Htarg = ptarget[7];
+    uint64_t htmax[] = {
+                0,
+                0xF,
+                0xFF,
+                0xFFF,
+                0xFFFF,
+                0x10000000
+        };
+    uint32_t masks[] = {
+                0xFFFFFFFF,
+                0xFFFFFFF0,
+                0xFFFFFF00,
+                0xFFFFF000,
+                0xFFFF0000,
+                0
+        };
+
+    swab32_array( endiandata, pdata, 20 );
+
+    uint64_t *edata = (uint64_t*)endiandata;
+    mm256_interleave_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+
+    for (int m=0; m < 6; m++)
+       if (Htarg <= htmax[m])
+       {
+          uint32_t mask = masks[m];
+
+          do
+          {
+             be32enc( noncep,   n   );
+             be32enc( noncep+2, n+1 );
+             be32enc( noncep+4, n+2 );
+             be32enc( noncep+6, n+3 );
+
+             anime_4way_hash( hash, vdata );
+             pdata[19] = n;
+
+             for ( int i = 0; i < 4; i++ )
+             if ( ( ( (hash+(i<<3))[7] & mask ) == 0 )
+                && fulltest( hash+(i<<3), ptarget ) )
+             {
+                pdata[19] = n+i;
+                nonces[ num_found++ ] = n+i;
+                work_set_target_ratio( work, hash+(i<<3) );
+             }
+             n += 4;
+          } while ( ( num_found == 0 ) && ( n < max_nonce )
+              && !work_restart[thr_id].restart );
+          break;
+       }
+
+    *hashes_done = n - first_nonce + 1;
+    return num_found;
+}
+
+#endif
--- a/algo/quark/anime-gate.c
+++ b/algo/quark/anime-gate.c
@@ -0,0 +1,17 @@
+#include "anime-gate.h"
+
+bool register_anime_algo( algo_gate_t* gate )
+{
+#if defined (ANIME_4WAY)
+  init_anime_4way_ctx();
+  gate->scanhash  = (void*)&scanhash_anime_4way;
+  gate->hash      = (void*)&anime_4way_hash;
+#else
+  init_anime_ctx();
+  gate->scanhash  = (void*)&scanhash_anime;
+  gate->hash      = (void*)&anime_hash;
+#endif
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
+  return true;
+};
+
--- a/algo/quark/anime-gate.h
+++ b/algo/quark/anime-gate.h
@@ -0,0 +1,32 @@
+#ifndef ANIME_GATE_H__
+#define ANIME_GATE_H__ 1
+
+#include "algo-gate-api.h"
+#include <stdint.h>
+
+#if defined(__AVX2__) && defined(__AES__)
+  #define ANIME_4WAY
+#endif
+
+bool register_anime_algo( algo_gate_t* gate );
+
+#if defined(ANIME_4WAY)
+
+void anime_4way_hash( void *state, const void *input );
+
+int scanhash_anime_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+
+void init_anime_4way_ctx();
+
+#endif
+
+void anime_hash( void *state, const void *input );
+
+int scanhash_anime( int thr_id, struct work *work, uint32_t max_nonce,
+                    uint64_t *hashes_done );
+
+void init_anime_ctx();
+
+#endif
+
--- a/algo/quark/anime.c
+++ b/algo/quark/anime.c
@@ -0,0 +1,189 @@
+#include "cpuminer-config.h"
+#include "anime-gate.h"
+#include <stdio.h>
+#include <string.h>
+#include <stdint.h>
+#include "algo/blake/sph_blake.h"
+#include "algo/bmw/sph_bmw.h"
+#include "algo/skein/sph_skein.h"
+#include "algo/jh/sph_jh.h"
+#include "algo/keccak/sph_keccak.h"
+#ifdef __AES__
+ #include "algo/groestl/aes_ni/hash-groestl.h"
+#else
+ #include "algo/groestl/sph_groestl.h"
+#endif
+
+typedef struct {
+    sph_blake512_context  blake;
+    sph_bmw512_context    bmw;
+#ifdef __AES__
+    hashState_groestl groestl;
+#else
+    sph_groestl512_context groestl;
+#endif
+    sph_jh512_context      jh;
+    sph_skein512_context   skein;
+    sph_keccak512_context  keccak;
+} anime_ctx_holder;
+
+anime_ctx_holder anime_ctx __attribute__ ((aligned (64)));
+
+void init_anime_ctx()
+{
+     sph_blake512_init( &anime_ctx.blake );
+     sph_bmw512_init( &anime_ctx.bmw );
+#ifdef __AES__
+    init_groestl( &anime_ctx.groestl, 64 );
+#else
+     sph_groestl512_init( &anime_ctx.groestl );
+#endif
+     sph_skein512_init( &anime_ctx.skein );
+     sph_jh512_init( &anime_ctx.jh );
+     sph_keccak512_init( &anime_ctx.keccak );
+}
+
+void anime_hash( void *state, const void *input )
+{
+    unsigned char hash[128] __attribute__ ((aligned (32)));
+/*
+    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)));
+    uint64_t vhashA[8*4] __attribute__ ((aligned (64)));
+    uint64_t vhashB[8*4] __attribute__ ((aligned (64)));
+    __m256i* vh  = (__m256i*)vhash;
+    __m256i* vhA = (__m256i*)vhashA;
+    __m256i* vhB = (__m256i*)vhashB;
+    __m256i vh_mask;
+    __m256i bit3_mask; bit3_mask = _mm256_set1_epi64x( 8 );
+*/
+    uint32_t mask = 8;
+    anime_ctx_holder ctx;
+    memcpy( &ctx, &anime_ctx, sizeof(anime_ctx) );
+
+    sph_bmw512( &ctx.bmw, input, 80 );
+    sph_bmw512_close( &ctx.bmw, hash );
+
+    sph_blake512( &ctx.blake, hash, 64 );
+    sph_blake512_close( &ctx.blake, hash );
+
+    if ( ( hash[0] & mask ) != 0 ) 
+    {
+#ifdef __AES__
+       update_and_final_groestl( &ctx.groestl, (char*)hash, (char*)hash, 512 );
+       reinit_groestl( &ctx.groestl );
+#else
+       sph_groestl512 ( &ctx.groestl, hash, 64 );
+       sph_groestl512_close( &ctx.groestl, hash );
+       sph_groestl512_init( &ctx.groestl );
+#endif
+    }
+    else
+    {
+       sph_skein512( &ctx.skein, hash, 64 );
+       sph_skein512_close( &ctx.skein, hash );
+       sph_skein512_init( &ctx.skein );
+    }
+
+#ifdef __AES__
+    update_and_final_groestl( &ctx.groestl, (char*)hash, (char*)hash, 512 );
+#else
+    sph_groestl512 ( &ctx.groestl, hash, 64 );
+    sph_groestl512_close( &ctx.groestl, hash );
+#endif
+
+    sph_jh512( &ctx.jh, hash, 64 );
+    sph_jh512_close( &ctx.jh, hash );
+
+    if ( ( hash[0] & mask ) != 0 )
+    {
+       sph_blake512_init( &ctx.blake );
+       sph_blake512( &ctx.blake, hash, 64 );
+       sph_blake512_close( &ctx.blake, hash );
+    }
+    else
+    {
+       sph_bmw512_init( &ctx.bmw );
+       sph_bmw512( &ctx.bmw, hash, 64 );
+       sph_bmw512_close( &ctx.bmw, hash );
+    }
+
+    sph_keccak512( &ctx.keccak, hash, 64 );
+    sph_keccak512_close( &ctx.keccak, hash );
+
+    sph_skein512( &ctx.skein, hash, 64 );
+    sph_skein512_close( &ctx.skein, hash );
+
+    if ( ( hash[0] & mask ) != 0 )
+    {
+       sph_keccak512_init( &ctx.keccak );
+       sph_keccak512( &ctx.keccak, hash, 64 );
+       sph_keccak512_close( &ctx.keccak, hash );
+    }
+    else
+    {
+       sph_jh512_init( &ctx.jh );
+       sph_jh512( &ctx.jh, hash, 64 );
+       sph_jh512_close( &ctx.jh, hash );
+    }
+
+   memcpy( state, hash, 32 );
+}
+
+int scanhash_anime( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done)
+{
+    uint32_t hash[8] __attribute__ ((aligned (64)));
+    uint32_t endiandata[20] __attribute__((aligned(64)));
+    uint32_t *pdata = work->data;
+    uint32_t *ptarget = work->target;
+    uint32_t n = pdata[19];
+    const uint32_t first_nonce = pdata[19];
+    const uint32_t Htarg = ptarget[7];
+    uint64_t htmax[] = {
+                0,
+                0xF,
+                0xFF,
+                0xFFF,
+                0xFFFF,
+                0x10000000
+        };
+    uint32_t masks[] = {
+                0xFFFFFFFF,
+                0xFFFFFFF0,
+                0xFFFFFF00,
+                0xFFFFF000,
+                0xFFFF0000,
+                0
+        };
+
+    swab32_array( endiandata, pdata, 20 );
+
+    for (int m=0; m < 6; m++)
+       if (Htarg <= htmax[m])
+       {
+          uint32_t mask = masks[m];
+          do
+          {
+              be32enc( &endiandata[19], n );
+              anime_hash( hash, endiandata );
+              pdata[19] = n;
+
+             if ( ( hash[7] & mask ) == 0 && fulltest( hash, ptarget ) ) 
+             {
+                work_set_target_ratio( work, hash );
+                *hashes_done = n - first_nonce + 1;
+                return true;
+             }
+             n++;
+          } while ( ( n < max_nonce ) && !work_restart[thr_id].restart );
+          break;
+       }
+
+    pdata[19] = n;
+    return 0;
+}
+
--- a/algo/quark/quark-4way.c
+++ b/algo/quark/quark-4way.c
@@ -0,0 +1,182 @@
+#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"
+#include "algo/jh/jh-hash-4way.h"
+#include "algo/keccak/keccak-hash-4way.h"
+#include "algo/groestl/aes_ni/hash-groestl.h"
+
+typedef struct {
+    blake512_4way_context  blake;
+    bmw512_4way_context    bmw;
+    hashState_groestl      groestl;
+    jh512_4way_context     jh;
+    skein512_4way_context  skein;
+    keccak512_4way_context keccak;
+} quark_4way_ctx_holder;
+
+quark_4way_ctx_holder quark_4way_ctx __attribute__ ((aligned (64)));
+
+void init_quark_4way_ctx()
+{
+     blake512_4way_init( &quark_4way_ctx.blake );
+     bmw512_4way_init( &quark_4way_ctx.bmw );
+     init_groestl( &quark_4way_ctx.groestl, 64 );
+     skein512_4way_init( &quark_4way_ctx.skein );
+     jh512_4way_init( &quark_4way_ctx.jh );
+     keccak512_4way_init( &quark_4way_ctx.keccak );
+}
+
+void quark_4way_hash( void *state, const void *input )
+{
+    uint64_t hash0[8] __attribute__ ((aligned (64)));
+    uint64_t hash1[8] __attribute__ ((aligned (64)));
+    uint64_t hash2[8] __attribute__ ((aligned (64)));
+    uint64_t hash3[8] __attribute__ ((aligned (64)));
+    uint64_t vhash[8*4] __attribute__ ((aligned (64)));
+    uint64_t vhashA[8*4] __attribute__ ((aligned (64)));
+    uint64_t vhashB[8*4] __attribute__ ((aligned (64)));
+    __m256i* vh  = (__m256i*)vhash;
+    __m256i* vhA = (__m256i*)vhashA;
+    __m256i* vhB = (__m256i*)vhashB;
+    __m256i vh_mask;
+    __m256i bit3_mask; bit3_mask = _mm256_set1_epi64x( 8 );
+    int i;
+    quark_4way_ctx_holder ctx;
+    memcpy( &ctx, &quark_4way_ctx, sizeof(quark_4way_ctx) );
+
+    blake512_4way( &ctx.blake, input, 80 );
+    blake512_4way_close( &ctx.blake, vhash );
+
+    bmw512_4way( &ctx.bmw, vhash, 64 );
+    bmw512_4way_close( &ctx.bmw, vhash );
+
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
+                                  m256_zero );
+
+       mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
+       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 );
+       mm256_interleave_4x64( vhashA, hash0, hash1, hash2, hash3, 512 );
+
+       skein512_4way( &ctx.skein, vhash, 64 );
+       skein512_4way_close( &ctx.skein, vhashB );
+
+    for ( i = 0; i < 8; i++ )
+       vh[i] = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );
+
+    mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, 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 );
+    mm256_interleave_4x64( vhash, hash0, hash1, hash2, hash3, 512 );
+
+    jh512_4way( &ctx.jh, vhash, 64 );
+    jh512_4way_close( &ctx.jh, vhash );
+
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
+                                  m256_zero );
+
+       blake512_4way_init( &ctx.blake );
+       blake512_4way( &ctx.blake, vhash, 64 );
+       blake512_4way_close( &ctx.blake, vhashA );
+
+       bmw512_4way_init( &ctx.bmw );
+       bmw512_4way( &ctx.bmw, vhash, 64 );
+       bmw512_4way_close( &ctx.bmw, vhashB );
+
+    for ( i = 0; i < 8; i++ )
+       vh[i] = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );
+
+    keccak512_4way( &ctx.keccak, vhash, 64 );
+    keccak512_4way_close( &ctx.keccak, vhash );
+
+    skein512_4way_init( &ctx.skein );
+    skein512_4way( &ctx.skein, vhash, 64 );
+    skein512_4way_close( &ctx.skein, vhash );
+
+    vh_mask = _mm256_cmpeq_epi64( _mm256_and_si256( vh[0], bit3_mask ),
+                                  m256_zero );
+
+       keccak512_4way_init( &ctx.keccak );
+       keccak512_4way( &ctx.keccak, vhash, 64 );
+       keccak512_4way_close( &ctx.keccak, vhashA );
+
+       jh512_4way_init( &ctx.jh );
+       jh512_4way( &ctx.jh, vhash, 64 );
+       jh512_4way_close( &ctx.jh, vhashB );
+
+    for ( i = 0; i < 8; i++ )
+       vh[i] = _mm256_blendv_epi8( vhA[i], vhB[i], vh_mask );
+
+    mm256_deinterleave_4x64( state, state+32, state+64, state+96, vhash, 256 );
+}
+
+int scanhash_quark_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done)
+{
+    uint32_t hash[4*8] __attribute__ ((aligned (64)));
+    uint32_t vdata[24*4] __attribute__ ((aligned (64)));
+    uint32_t endiandata[20] __attribute__((aligned(64)));
+    uint32_t *pdata = work->data;
+    uint32_t *ptarget = work->target;
+    uint32_t n = pdata[19];
+    const uint32_t first_nonce = pdata[19];
+    uint32_t *nonces = work->nonces;
+    int num_found = 0;
+    uint32_t *noncep = vdata + 73;   // 9*8 + 1
+
+    swab32_array( endiandata, pdata, 20 );
+
+    uint64_t *edata = (uint64_t*)endiandata;
+    mm256_interleave_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
+
+    do
+    {
+       be32enc( noncep,   n   );
+       be32enc( noncep+2, n+1 );
+       be32enc( noncep+4, n+2 );
+       be32enc( noncep+6, n+3 );
+
+       quark_4way_hash( hash, vdata );
+       pdata[19] = n;
+
+       for ( int i = 0; i < 4; i++ )
+       if ( ( ( (hash+(i<<3))[7] & 0xFFFFFF00 ) == 0 )
+            && fulltest( hash+(i<<3), ptarget ) )
+       {
+          pdata[19] = n+i;
+          nonces[ num_found++ ] = n+i;
+          work_set_target_ratio( work, hash+(i<<3) );
+       }
+       n += 4;
+    } while ( ( num_found == 0 ) && ( n < max_nonce )
+              && !work_restart[thr_id].restart );
+
+    *hashes_done = n - first_nonce + 1;
+    return num_found;
+}
+
+#endif
--- a/algo/quark/quark-gate.c
+++ b/algo/quark/quark-gate.c
@@ -0,0 +1,17 @@
+#include "quark-gate.h"
+
+bool register_quark_algo( algo_gate_t* gate )
+{
+#if defined (QUARK_4WAY)
+  init_quark_4way_ctx();
+  gate->scanhash  = (void*)&scanhash_quark_4way;
+  gate->hash      = (void*)&quark_4way_hash;
+#else
+  init_quark_ctx();
+  gate->scanhash  = (void*)&scanhash_quark;
+  gate->hash      = (void*)&quark_hash;
+#endif
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
+  return true;
+};
+
--- a/algo/quark/quark-gate.h
+++ b/algo/quark/quark-gate.h
@@ -0,0 +1,32 @@
+#ifndef QUARK_GATE_H__
+#define QUARK_GATE_H__ 1
+
+#include "algo-gate-api.h"
+#include <stdint.h>
+
+#if defined(__AVX2__) && defined(__AES__)
+  #define QUARK_4WAY
+#endif
+
+bool register_quark_algo( algo_gate_t* gate );
+
+#if defined(QUARK_4WAY)
+
+void quark_4way_hash( void *state, const void *input );
+
+int scanhash_quark_4way( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done );
+
+void init_quark_4way_ctx();
+
+#endif
+
+void quark_hash( void *state, const void *input );
+
+int scanhash_quark( int thr_id, struct work *work, uint32_t max_nonce,
+                    uint64_t *hashes_done );
+
+void init_quark_ctx();
+
+#endif
+
--- a/algo/quark/quark.c
+++ b/algo/quark/quark.c
@@ -1,5 +1,5 @@
 #include "cpuminer-config.h"
-#include "algo-gate-api.h"
+#include "quark-gate.h"

 #include <stdio.h>
 #include <string.h>
@@ -47,7 +47,7 @@ void init_quark_ctx()
 #endif
 }

-inline static void quarkhash(void *state, const void *input)
+void quark_hash(void *state, const void *input)
 {
    unsigned char hashbuf[128];
    size_t hashptr;
@@ -187,11 +187,12 @@ int scanhash_quark( int thr_id, struct work *work, uint32_t max_nonce,
 	do {
 		pdata[19] = ++n;
 		be32enc(&endiandata[19], n); 
-		quarkhash(hash64, &endiandata);
+		quark_hash(hash64, &endiandata);
                if ((hash64[7]&0xFFFFFF00)==0)
                {
                  if (fulltest(hash64, ptarget)) 
                  {
+                    work_set_target_ratio( work, hash64 );
                    *hashes_done = n - first_nonce + 1;
 		    return true;
                  }
@@ -203,12 +204,3 @@ int scanhash_quark( int thr_id, struct work *work, uint32_t max_nonce,
 	return 0;
 }

-bool register_quark_algo( algo_gate_t* gate )
-{
-  init_quark_ctx();
-  gate->optimizations = SSE2_OPT | AES_OPT;
-  gate->scanhash         = (void*)&scanhash_quark;
-  gate->hash             = (void*)&quarkhash;
-  return true;
-};
-
--- a/algo/qubit/deep-2way.c
+++ b/algo/qubit/deep-2way.c
@@ -0,0 +1,124 @@
+#include "deep-gate.h"
+
+#if defined(DEEP_2WAY)
+
+#include <stdlib.h>
+#include <stdint.h>
+#include <string.h>
+#include <stdio.h>
+#include "algo/luffa/luffa-hash-2way.h"
+#include "algo/cubehash/sse2/cubehash_sse2.h" 
+#include "algo/shavite/sph_shavite.h"
+#include "algo/echo/aes_ni/hash_api.h"
+
+typedef struct
+{
+        luffa_2way_context      luffa;
+        cubehashParam           cube;
+        sph_shavite512_context  shavite;
+        hashState_echo          echo;
+} deep_2way_ctx_holder;
+
+deep_2way_ctx_holder deep_2way_ctx;
+
+void init_deep_2way_ctx()
+{
+        luffa_2way_init( &deep_2way_ctx.luffa, 512 );
+        cubehashInit(&deep_2way_ctx.cube,512,16,32);
+        sph_shavite512_init(&deep_2way_ctx.shavite);
+        init_echo(&deep_2way_ctx.echo, 512);
+};
+
+void deep_2way_hash( void *output, const void *input )
+{
+     uint64_t hash0[8] __attribute__ ((aligned (64)));
+     uint64_t hash1[8] __attribute__ ((aligned (64)));
+     uint64_t vhash[8*2] __attribute__ ((aligned (64)));
+     deep_2way_ctx_holder ctx;
+
+     memcpy( &ctx, &deep_2way_ctx, sizeof(deep_2way_ctx) );
+     luffa_2way_update( &ctx.luffa, input + (64<<1), 16 );
+     luffa_2way_close( &ctx.luffa, vhash );
+     mm256_deinterleave_2x128( hash0, hash1, vhash, 512 );
+
+     cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
+                           (const byte*) hash0, 64 );
+     memcpy( &ctx.cube, &deep_2way_ctx.cube, sizeof(cubehashParam) );
+     cubehashUpdateDigest( &ctx.cube, (byte*)hash1, (const byte*) hash1, 64 );
+
+     sph_shavite512( &ctx.shavite, hash0, 64 );
+     sph_shavite512_close( &ctx.shavite, hash0 );
+     memcpy( &ctx.shavite, &deep_2way_ctx.shavite,
+             sizeof(sph_shavite512_context) );
+     sph_shavite512( &ctx.shavite, hash1, 64 );
+     sph_shavite512_close( &ctx.shavite, hash1 );
+
+     update_final_echo( &ctx.echo, (BitSequence *)hash0,
+                       (const BitSequence *) hash0, 512 );
+     memcpy( &ctx.echo, &deep_2way_ctx.echo, sizeof(hashState_echo) );
+     update_final_echo( &ctx.echo, (BitSequence *)hash1,
+                       (const BitSequence *) hash1, 512 );
+
+     memcpy( output,    hash0, 32 );
+     memcpy( output+32, hash1, 32 );
+}
+
+int scanhash_deep_2way( int thr_id, struct work *work,uint32_t max_nonce,
+                         uint64_t *hashes_done )
+{
+     uint32_t hash[4*8] __attribute__ ((aligned (64)));
+     uint32_t vdata[24*4] __attribute__ ((aligned (64)));
+     uint32_t endiandata[20] __attribute__((aligned(64)));
+     uint32_t *pdata = work->data;
+     uint32_t *ptarget = work->target;
+     uint32_t n = pdata[19];
+     const uint32_t first_nonce = pdata[19];
+     uint32_t *nonces = work->nonces;
+     int num_found = 0;
+     uint32_t *noncep = vdata + 32+3;   // 4*8 + 3
+     const uint32_t Htarg = ptarget[7];
+     uint64_t htmax[] = {          0,        0xF,       0xFF,
+                               0xFFF,     0xFFFF, 0x10000000  };
+     uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
+                          0xFFFFF000, 0xFFFF0000,          0  };
+
+     // big endian encode 0..18 uint32_t, 64 bits at a time
+     swab32_array( endiandata, pdata, 20 );
+
+     uint64_t *edata = (uint64_t*)endiandata;
+     mm256_interleave_2x128( (uint64_t*)vdata, edata, edata, 640 );
+
+     luffa_2way_init( &deep_2way_ctx.luffa, 512 );
+     luffa_2way_update( &deep_2way_ctx.luffa, vdata, 64 );
+
+     for ( int m=0; m < 6; m++ ) if ( Htarg <= htmax[m] )
+     {
+        uint32_t mask = masks[m];
+        do
+        {
+            be32enc( noncep,   n   );
+            be32enc( noncep+4, n+1 );
+
+            deep_2way_hash( hash, vdata );
+            pdata[19] = n;
+
+            if ( !( hash[7] & mask ) && fulltest( hash, ptarget) )
+            {
+               nonces[ num_found++ ] = n;
+               work_set_target_ratio( work, hash );
+            }
+            if ( !( (hash+8)[7] & mask ) && fulltest( hash+8, ptarget) )
+            {
+               nonces[ num_found++ ] = n+1;
+               work_set_target_ratio( work, hash+8 );
+            }
+            n += 2;
+         } while ( ( num_found == 0 ) && ( n < max_nonce )
+                   && !work_restart[thr_id].restart );
+         break;
+     }
+     *hashes_done = n - first_nonce + 1;
+     return num_found;
+}
+
+#endif
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Jay D Dee	157508bd07	v3.8.3.3	2018-02-25 14:15:07 -05:00
Jay D Dee	c24a4bdbc2	v3.8.3.2	2018-02-24 14:36:19 -05:00
Jay D Dee	59c7848d91	v3.8.3.1	2018-02-23 15:45:32 -05:00
Jay D Dee	3c02653dbe	v3.8.3	2018-02-23 12:39:15 -05:00
Jay D Dee	502ed0b1fe	v3.8.2.1	2018-02-17 13:52:24 -05:00
Jay D Dee	d60a268972	v3.8.2	2018-02-15 14:48:50 -05:00
Jay D Dee	e4265a6f11	v3.8.1.1	2018-02-09 23:30:14 -05:00
Jay D Dee	a28daca3ce	v3.8.1	2018-02-07 16:38:45 -05:00
Jay D Dee	54b8fd7362	v3.8.0.1	2018-02-05 22:10:18 -05:00
Jay D Dee	ad2275f74a	v3.8.0	2018-01-23 21:02:16 -05:00
Jay D Dee	a90d75b8f5	v3.7.10	2018-01-16 15:11:44 -05:00