v3.18.1

v3.7.4
v3.17.1
2025-09-17 23:44:27 +00:00 · 2021-10-10 22:50:19 -04:00 · 2021-09-29 17:31:16 -04:00 · 2021-07-26 15:01:37 -04:00 · 2021-07-15 20:30:44 -04:00 · 2021-06-26 12:27:44 -04:00
169 changed files with 16844 additions and 6738 deletions
--- a/22
+++ b/22
@@ -32,14 +32,26 @@ but different package names.
 $ sudo apt-get install build-essential automake libssl-dev libcurl4-openssl-dev libjansson-dev libgmp-dev zlib1g-dev git

 SHA support on AMD Ryzen CPUs requires gcc version 5 or higher and
-openssl 1.1.0e or higher. Add one of the following to CFLAGS for SHA
-support depending on your CPU and compiler version:
+openssl 1.1.0e or higher.

-"-march=native" is always the best choice
+znver1 and znver2 should be recognized on most recent version of GCC and
+znver3 is expected with GCC 11. GCC 11 also includes rocketlake support.
+In the meantime here are some suggestions to compile with new CPUs:

-"-march=znver1" for Ryzen 1000 & 2000 series, znver2 for 3000.
+"-march=native" is usually the best choice, used by build.sh.

-"-msha"  Add SHA to other tuning options
+"-march=znver2 -mvaes" can be used for Ryzen 5000 if znver3 is not recongized.
+
+"-mcascadelake -msha" or
+"-mcometlake -mavx512 -msha" can be used for Rocket Lake.
+
+Features can also be added individually:
+
+"-msha" adds support for HW accelerated sha256.
+
+"-mavx512" adds support for 512 bit vectors
+
+"-mvaes" add support for parallel AES

 Additional instructions for static compilalation can be found here:
 https://lxadm.com/Static_compilation_of_cpuminer
--- a/145
+++ b/145
@@ -1,5 +1,9 @@
 Instructions for compiling cpuminer-opt for Windows.

+Thwaw intructions nay be out of date. Please consult the wiki for
+the latest:
+
+https://github.com/JayDDee/cpuminer-opt/wiki/Compiling-from-source

 Windows compilation using Visual Studio is not supported. Mingw64 is
 used on a Linux system (bare metal or virtual machine) to cross-compile
@@ -24,79 +28,76 @@ Refer to Linux compile instructions and install required packages.

 Additionally, install mingw-w64.

-sudo apt-get install mingw-w64
+sudo apt-get install mingw-w64 libz-mingw-w64-dev


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

+$ mkdir $HOME/usr/lib
+
 3. Download and build other packages for mingw that don't have a mingw64
   version available in the repositories.

 Download the following source code packages from their respective and
 respected download locations, copy them to ~/usr/lib/ and uncompress them. 

-openssl
-curl
-gmp
+openssl: https://github.com/openssl/openssl/releases

-In most cases the latest vesrion is ok but it's safest to download
-the same major and minor version as included in your distribution.
+curl: https://github.com/curl/curl/releases

-Run the following commands or follow the supplied instructions.
-Do not run "make install" unless you are using ~/usr/lib, which isn't
-recommended.
+gmp: https://gmplib.org/download/gmp/

-Some instructions insist on running "make check". If make check fails
-it may still work, YMMV.
+In most cases the latest version is ok but it's safest to download the same major and minor version as included in your distribution. The following uses versions from Ubuntu 20.04. Change version numbers as required.

-You can speed up "make" by using all CPU cores available with "-j n" where
-n is the number of CPU threads you want to use.
+Run the following commands or follow the supplied instructions. Do not run "make install" unless you are using /usr/lib, which isn't recommended.
+
+Some instructions insist on running "make check". If make check fails it may still work, YMMV.
+
+You can speed up "make" by using all CPU cores available with "-j n" where n is the number of CPU threads you want to use.

 openssl:

-./Configure mingw64 shared --cross-compile-prefix=x86_64-w64-mingw32
-make
+$ ./Configure mingw64 shared --cross-compile-prefix=x86_64-w64-mingw32-
+$ make
+
+Make may fail with an ld error, just ensure libcrypto-1_1-x64.dll is created.

 curl:

-./configure --with-winssl --with-winidn --host=x86_64-w64-mingw32
-make
+$ ./configure --with-winssl --with-winidn --host=x86_64-w64-mingw32
+$ make

 gmp:

-./configure --host=x86_64-w64-mingw32 
-make
-
-
+$ ./configure --host=x86_64-w64-mingw32
+$ make

 4. Tweak the environment.

-This step is required everytime you login or the commands can be added to
-.bashrc.
+This step is required everytime you login or the commands can be added to .bashrc.

-Define some local variables to point to local library. 
+Define some local variables to point to local library.

-export LOCAL_LIB="$HOME/usr/lib"
+$ export LOCAL_LIB="$HOME/usr/lib"

-export LDFLAGS="-L$LOCAL_LIB/curl/lib/.libs -L$LOCAL_LIB/gmp/.libs -L$LOCAL_LIB/openssl"
+$ export LDFLAGS="-L$LOCAL_LIB/curl/lib/.libs -L$LOCAL_LIB/gmp/.libs -L$LOCAL_LIB/openssl"

-export CONFIGURE_ARGS="--with-curl=$LOCAL_LIB/curl --with-crypto=$LOCAL_LIB/openssl --host=x86_64-w64-mingw32"
+$ export CONFIGURE_ARGS="--with-curl=$LOCAL_LIB/curl --with-crypto=$LOCAL_LIB/openssl --host=x86_64-w64-mingw32"

-Create a release directory and copy some dll files previously built.
-This can be done outside of cpuminer-opt and only needs to be done once.
-If the release directory is in cpuminer-opt directory it needs to be
-recreated every a source package is decompressed.
+Adjust for gcc version:

-mkdir release
-cp /usr/x86_64-w64-mingw32/lib/zlib1.dll release/
-cp /usr/x86_64-w64-mingw32/lib/libwinpthread-1.dll release/
-cp /usr/lib/gcc/x86_64-w64-mingw32/7.3-win32/libstdc++-6.dll release/
-cp /usr/lib/gcc/x86_64-w64-mingw32/7.3-win32/libgcc_s_seh-1.dll release/
-cp $LOCAL_LIB/openssl/libcrypto-1_1-x64.dll release/
-cp $LOCAL_LIB/curl/lib/.libs/libcurl-4.dll release/
+$ export GCC_MINGW_LIB="/usr/lib/gcc/x86_64-w64-mingw32/9.3-win32"

+Create a release directory and copy some dll files previously built. This can be done outside of cpuminer-opt and only needs to be done once. If the release directory is in cpuminer-opt directory it needs to be recreated every time a source package is decompressed.

+$ mkdir release
+$ cp /usr/x86_64-w64-mingw32/lib/zlib1.dll release/
+$ cp /usr/x86_64-w64-mingw32/lib/libwinpthread-1.dll release/
+$ cp $GCC_MINGW_LIB/libstdc++-6.dll release/
+$ cp $GCC_MINGW_LIB/libgcc_s_seh-1.dll release/
+$ cp $LOCAL_LIB/openssl/libcrypto-1_1-x64.dll release/
+$ cp $LOCAL_LIB/curl/lib/.libs/libcurl-4.dll release/

 The following steps need to be done every time a new source package is
 opened.
@@ -110,13 +111,73 @@ https://github.com/JayDDee/cpuminer-opt/releases

 Decompress and change to the cpuminer-opt directory.

-
-
-6. Prepare to compile
+6. compile

 Create a link to the locally compiled version of gmp.h

-ln -s $LOCAL_LIB/gmp-version/gmp.h ./gmp.h
+$ ln -s $LOCAL_LIB/gmp-version/gmp.h ./gmp.h
+
+$ ./autogen.sh
+
+Configure the compiler for the CPU architecture of the host machine:
+
+CFLAGS="-O3 -march=native -Wall" ./configure $CONFIGURE_ARGS
+
+or cross compile for a specific CPU architecture:
+
+CFLAGS="-O3 -march=znver1 -Wall" ./configure $CONFIGURE_ARGS
+
+This will compile for AMD Ryzen.
+
+You can compile more generically for a set of specific CPU features if you know what features you want:
+
+CFLAGS="-O3 -maes -msse4.2 -Wall" ./configure $CONFIGURE_ARGS
+
+This will compile for an older CPU that does not have AVX.
+
+You can find several examples in README.txt
+
+If you have a CPU with more than 64 threads and Windows 7 or higher you can enable the CPU Groups feature by adding the following to CFLAGS:
+
+"-D_WIN32_WINNT=0x0601"
+
+Once you have run configure successfully run the compiler with n CPU threads:
+
+$ make -j n
+
+Copy cpuminer.exe to the release directory, compress and copy the release directory to a Windows system and run cpuminer.exe from the command line.
+
+Run cpuminer
+
+In a command windows change directories to the unzipped release folder. to get a list of all options:
+
+cpuminer.exe --help
+
+Command options are specific to where you mine. Refer to the pool's instructions on how to set them.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+Create a link to the locally compiled version of gmp.h
+
+$ ln -s $LOCAL_LIB/gmp-version/gmp.h ./gmp.h

 Edit configure.ac to fix lipthread package name.

--- a/Makefile.am
+++ b/Makefile.am
@@ -129,7 +129,7 @@ cpuminer_SOURCES = \
  algo/lyra2/allium.c \
  algo/lyra2/phi2-4way.c \
  algo/lyra2/phi2.c \
-  algo//m7m/m7m.c \
+  algo/m7m/m7m.c \
  algo/m7m/magimath.cpp \
  algo/nist5/nist5-gate.c \
  algo/nist5/nist5-4way.c \
@@ -158,13 +158,18 @@ cpuminer_SOURCES = \
  algo/ripemd/lbry.c \
  algo/ripemd/lbry-4way.c \
  algo/scrypt/scrypt.c \
+  algo/scrypt/scrypt-core-4way.c \
  algo/scrypt/neoscrypt.c \
+  algo/sha/sha256-hash.c \
  algo/sha/sph_sha2.c \
  algo/sha/sph_sha2big.c \
  algo/sha/sha256-hash-4way.c \
  algo/sha/sha512-hash-4way.c \
+  algo/sha/sha256-hash-opt.c \
+  algo/sha/sha256-hash-2way-ni.c \
  algo/sha/hmac-sha256-hash.c \
  algo/sha/hmac-sha256-hash-4way.c \
+  algo/sha/sha256d.c \
  algo/sha/sha2.c \
  algo/sha/sha256t-gate.c \
  algo/sha/sha256t-4way.c \
@@ -192,6 +197,11 @@ cpuminer_SOURCES = \
  algo/sm3/sm3-hash-4way.c \
  algo/swifftx/swifftx.c \
  algo/tiger/sph_tiger.c \
+  algo/verthash/verthash-gate.c \
+  algo/verthash/Verthash.c \
+  algo/verthash/fopen_utf8.c \
+  algo/verthash/tiny_sha3/sha3.c \
+  algo/verthash/tiny_sha3/sha3-4way.c \
  algo/whirlpool/sph_whirlpool.c \
  algo/whirlpool/whirlpool-hash-4way.c \
  algo/whirlpool/whirlpool-gate.c \
--- a/README.md
+++ b/README.md
@@ -89,7 +89,7 @@ Supported Algorithms
                          lyra2h        Hppcoin
                          lyra2re       lyra2
                          lyra2rev2     lyra2v2
-                          lyra2rev3     lyrav2v3, Vertcoin
+                          lyra2rev3     lyrav2v3
                          lyra2z        
                          lyra2z330     Lyra2 330 rows, Zoin (ZOI)
                          m7m           Magi (XMG)
@@ -122,6 +122,7 @@ Supported Algorithms
                          tribus        Denarius (DNR)
                          vanilla       blake256r8vnl (VCash)
                          veltor        (VLT)
+                          verthash      Vertcoin
                          whirlpool
                          whirlpoolx
                          x11           Dash
@@ -134,7 +135,7 @@ Supported Algorithms
                          x14           X14
                          x15           X15
                          x16r          
-                          x16rv2        Ravencoin (RVN)
+                          x16rv2        
                          x16rt         Gincoin (GIN)
                          x16rt-veil    Veil (VEIL)
                          x16s          Pigeoncoin (PGN)
--- a/README.txt
+++ b/README.txt
@@ -1,6 +1,10 @@
 This file is included in the Windows binary package. Compile instructions
 for Linux and Windows can be found in RELEASE_NOTES.

+This package is officially avalable only from:
+ https://github.com/JayDDee/cpuminer-opt
+No other sources should be trusted.
+
 cpuminer is a console program that is executed from a DOS or Powershell
 prompt. There is no GUI and no mouse support.

@@ -10,7 +14,7 @@ miners. The source code is open for anyone to inspect. If you don't trust
 the software, don't use it.

 Choose the exe that best matches you CPU's features or use trial and
-error to find the fastest one that doesn't crash. Pay attention to
+error to find the fastest one that works. Pay attention to
 the features listed at cpuminer startup to ensure you are mining at
 optimum speed using the best available features.

@@ -31,30 +35,40 @@ https://en.wikipedia.org/wiki/List_of_Intel_CPU_microarchitectures
 https://en.wikipedia.org/wiki/List_of_AMD_CPU_microarchitectures


-Exe file name              Compile flags              Arch name
+Exe file name                Compile flags              Arch name

-cpuminer-sse2.exe            "-msse2"                 Core2, Nehalem   
-cpuminer-aes-sse42.exe       "-march=westmere"        Westmere
-cpuminer-avx.exe             "-march=corei7-avx"      Sandybridge, Ivybridge
-cpuminer-avx2.exe            "-march=core-avx2 -maes" Haswell*
-cpuminer-avx512.exe          "-march=skylake-avx512"  Skylake-X, Cascadelake-X
-cpuminer-zen.exe             "-march=znver1"          AMD Ryzen, Threadripper
-cpuminer-avx512-sha-vaes.exe "-march=icelake-client"  Icelake*
+cpuminer-sse2.exe            "-msse2"                   Core2, Nehalem   
+cpuminer-aes-sse42.exe       "-march=westmere"          Westmere
+cpuminer-avx.exe             "-march=corei7-avx"        Sandybridge, Ivybridge
+cpuminer-avx2.exe            "-march=core-avx2 -maes"   Haswell(1)
+cpuminer-avx512.exe          "-march=skylake-avx512"    Skylake-X, Cascadelake
+cpuminer-avx512-sha.exe      "-march=cascadelake -msha" Rocketlake(2)
+cpuminer-avx512-sha-vaes.exe "-march=icelake-client"    Icelake, Tigerlake(3)
+cpuminer-zen.exe             "-march=znver1"            AMD Zen1, Zen2
+cpuminer-zen3.exe            "-march=znver2 -mvaes"     Zen3(4)

-* Haswell includes Broadwell, Skylake, Kabylake, Coffeelake & Cometlake. 
-Icelake is only available on some laptops. Mining with a laptop is not
-recommended. The icelake build is included in anticipation of Intel eventually
-releasing a desktop CPU with a microarchitecture newer than Skylake.
+(1) Haswell includes Broadwell, Skylake, Kabylake, Coffeelake & Cometlake. 
+(2) Rocketlake build uses cascadelake+sha as a workaround until Rocketlake
+    compiler support is avalable.
+(3) Icelake & Tigerlake are only available on some laptops. Mining with a
+    laptop is not recommended.
+(4) Zen3 build uses zen2+vaes as a workaround until Zen3 compiler support is
+    available. Zen2 CPUs should use Zen1 build.

 Notes about included DLL files:

 Downloading DLL files from alternative sources presents an inherent
 security risk if their source is unknown. All DLL files included have
-been copied from the Ubuntu-20.04 instalation or compiled by me from
+been copied from the Ubuntu-20.04 installation or compiled by me from
 source code obtained from the author's official repository. The exact
 procedure is documented in the build instructions for Windows:
 https://github.com/JayDDee/cpuminer-opt/wiki/Compiling-from-source

+Some DLL filess may already be installed on the system by Windows or third
+party packages. They often will work and may be used instead of the included
+file. Without a compelling reason to do so it's recommended to use the included
+files as they are packaged.
+
 If you like this software feel free to donate:

 BTC: 12tdvfF7KmAsihBXQXynT6E6th2c2pByTT
--- a/145
+++ b/145
@@ -65,6 +65,151 @@ If not what makes it happen or not happen?
 Change Log
 ----------

+v3.18.1
+
+More speed for scrypt:
+ - additional scryptn2 optimizations for all CPU architectures,
+ - AVX2 is now used by default on CPUS with SHA but not AVX512,
+ - scrypt:1024 performance lost in v3.18.0 is restored,
+ - AVX512 & AVX2 improvements to scrypt:1024.
+
+Big speedup for SwiFFTx AVX2 & SSE4.1: x22i +55%, x25x +22%.
+
+Issue #337: fixed a problem that could display negative stats values in the
+first summary report if the report was forced prematurely due to a stratum
+diff change. The stats will still be invalid but should display zeros.
+
+v3.18.0
+
+Complete rewrite of Scrypt code, optimized for large N factor (scryptn2):
+  - AVX512 & SHA support for sha256, AVX512 has priority,
+  - up to 50% increase in hashrate,
+  - memory requirements reduced 30-60% depending on CPU architecture,
+  - memory usage displayed at startup,
+  - scrypt, default N=1024 (LTC), will likely perform slower.
+
+Improved stale share detection and handling for Scrypt with large N factor:
+  - abort and discard partially computed hash when new work is detected,
+  - quicker response to new job, less time wasted mining stale job.
+
+Improved stale share handling for all algorithms:
+  - report possible stale share when new work received with a previously
+    submitted share still pending,
+  - when new work is detected report the submission of an already completed,
+    otherwise valid, but likely stale, share,
+  - fixed incorrect block height in stale share log.
+
+Small performance improvements to sha, bmw, cube & hamsi for AVX512 & AVX2.
+
+When stratum disconnects miner threads go to idle until reconnected.
+
+Colour changes to some logs.
+
+Some low level function name changes for clarity and consistency.
+
+The reference hashrate in the summary log and the benchmark total hashrate
+are now the mean hashrate for the session. 
+
+v3.17.1
+
+Fixed Windows build for AES+SSE4.2 (Westmere), was missing AES.
+More ternary logic optimizations for AVX512, AVX512+VAES, and AVX512+AES.
+Fixed my-gr algo for VAES.
+
+v3.17.0
+
+AVX512 optimized using ternary logic instructions.
+Faster sha256t on all CPU architectures: AVX512 +30%, SHA +30%, AVX2 +9%.
+Use SHA on supported CPUs to produce merkle hash.
+Fixed byte order in Extranonce2 log & replaced Block height with Job ID.
+
+v3.16.5
+
+#329: Fixed GBT incorrect target diff in stats, second attempt.
+Fixed formatting error in share result log when --no-color option is used.
+
+v3.16.4
+
+Faster sha512 and sha256 when not using SHA CPU extension.
+#329: Fixed GBT incorrect target diff in stats.
+
+v3.16.3
+
+#313 Fix compile error with GCC 11.
+Incremental improvements to verthash.
+
+v3.16.2
+
+Verthash: midstate prehash optimization for all architectures.
+Verthash: AVX2 optimization.
+GBT: added support for Bech32 addresses.
+Linux: added CPU frequency to benchmark log.
+Fixed integer overflow in time calculations.
+
+v3.16.1
+
+New options for verthash:
+  --data-file to specify the name, and optionally the path, of the verthash
+              data file, default is "verthash.dat" in the current directory.
+  --verify to perform the data file integrity check at startup, default is
+           not to verify data file integrity.
+Support for creation of default verthash data file if:
+   1) --data-file option is not used,
+   2) no default data file is found in the current directory, and,
+   3) --verify option is used.
+More detailed logs related to verthash data file.
+Small verthash performance improvement.
+Fixed detection of corrupt stats caused by networking issues.
+
+v3.16.0
+
+Added verthash algo.
+
+v3.15.7
+
+Added accepted/stale/rejected percentage to summary log report.
+Added warning if share counters mismatch which could corrupt stats.
+Linux: CPU temperature reporting is more responsive to rising temperature.
+A few AVX2 & AVX512 tweaks.
+Removed some dead code and other cleanup.
+
+v3.15.6
+
+Implement keccak pre-hash optimization for x16* algos.
+Move conditional mining test to before get_new_work in miner thread.
+Add test for share reject reason when solo mining.
+Add support for floating point, as well as integer, "networkhasps" in
+RPC getmininginfo method.
+
+v3.15.5
+
+Fix stratum jobs lost if 2 jobs received in less than one second.
+
+v3.15.4
+
+Fixed yescryptr16 broken in v3.15.3.
+
+v3.15.3
+
+Yescrypt algos now use yespower v0.5, a little faster.
+New implementation of sha256 using SHA CPU extension.
+Replace Openssl with SPH for sha256 & sha512.
+AVX512 optimization for sha256t & sha256q.
+Faster sha256t, sha256q, x21s, x22i & x25x on CPUs with SHA without AVX512.
+AVX512+SHA build for Intel Rocketlake added to Windows binary package. 
+
+v3.15.2
+
+Zen3 AVX2+VAES optimization for x16*, x17, sonoa, xevan, x21s, x22i, x25x,
+allium.
+Zen3 (AVX2+SHA+VAES) build added to Windows binary package.
+
+v3.15.1
+
+Fix compile on AMD Zen3 CPUs with VAES.
+Force new work immediately after solving a block solo.
+
+
 v3.15.0

 Fugue optimized with AES, improves many sha3 algos.
--- a/algo-gate-api.c
+++ b/algo-gate-api.c
@@ -15,8 +15,6 @@
 #include <stdbool.h>
 #include <memory.h>
 #include <unistd.h>
-#include <openssl/sha.h>
-//#include "miner.h"
 #include "algo-gate-api.h"

 // Define null and standard functions.
@@ -279,9 +277,11 @@ void init_algo_gate( algo_gate_t* gate )
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wimplicit-function-declaration"

-// called by each thread that uses the gate
+// Called once by main
 bool register_algo_gate( int algo, algo_gate_t *gate )
 {
+  bool rc = false;
+
  if ( NULL == gate )
  {
    applog(LOG_ERR,"FAIL: algo_gate registration failed, NULL gate\n");
@@ -290,104 +290,108 @@ bool register_algo_gate( int algo, algo_gate_t *gate )

  init_algo_gate( gate );

-  switch (algo)
+  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_ARGON2D250:    register_argon2d_crds_algo  ( gate ); break;
-    case ALGO_ARGON2D500:    register_argon2d_dyn_algo   ( gate ); break;
-    case ALGO_ARGON2D4096:   register_argon2d4096_algo   ( gate ); break;
-    case ALGO_AXIOM:         register_axiom_algo         ( gate ); break;
-    case ALGO_BLAKE:         register_blake_algo         ( gate ); break;
-    case ALGO_BLAKE2B:       register_blake2b_algo       ( gate ); break;
-    case ALGO_BLAKE2S:       register_blake2s_algo       ( gate ); break;
-    case ALGO_BLAKECOIN:     register_blakecoin_algo     ( gate ); break;
-    case ALGO_BMW512:        register_bmw512_algo        ( gate ); break;
-    case ALGO_C11:           register_c11_algo           ( gate ); break;
-    case ALGO_DECRED:        register_decred_algo        ( gate ); break;
-    case ALGO_DEEP:          register_deep_algo          ( gate ); break;
-    case ALGO_DMD_GR:        register_dmd_gr_algo        ( gate ); break;
-    case ALGO_GROESTL:       register_groestl_algo       ( gate ); break;
-    case ALGO_HEX:           register_hex_algo           ( gate ); break;
-    case ALGO_HMQ1725:       register_hmq1725_algo       ( gate ); break;
-    case ALGO_HODL:          register_hodl_algo          ( gate ); break;
-    case ALGO_JHA:           register_jha_algo           ( gate ); break;
-    case ALGO_KECCAK:        register_keccak_algo        ( gate ); break;
-    case ALGO_KECCAKC:       register_keccakc_algo       ( gate ); break;
-    case ALGO_LBRY:          register_lbry_algo          ( gate ); break;
-    case ALGO_LYRA2H:        register_lyra2h_algo        ( gate ); break;
-    case ALGO_LYRA2RE:       register_lyra2re_algo       ( gate ); break;
-    case ALGO_LYRA2REV2:     register_lyra2rev2_algo     ( gate ); break;
-    case ALGO_LYRA2REV3:     register_lyra2rev3_algo     ( gate ); break;
-    case ALGO_LYRA2Z:        register_lyra2z_algo        ( gate ); break;
-    case ALGO_LYRA2Z330:     register_lyra2z330_algo     ( gate ); break;
-    case ALGO_M7M:           register_m7m_algo           ( gate ); break;
-    case ALGO_MINOTAUR:      register_minotaur_algo      ( gate ); break;
-    case ALGO_MYR_GR:        register_myriad_algo        ( gate ); break;
-    case ALGO_NEOSCRYPT:     register_neoscrypt_algo     ( gate ); break;
-    case ALGO_NIST5:         register_nist5_algo         ( gate ); break;
-    case ALGO_PENTABLAKE:    register_pentablake_algo    ( gate ); break;
-    case ALGO_PHI1612:       register_phi1612_algo       ( gate ); break;
-    case ALGO_PHI2:          register_phi2_algo          ( gate ); break;
-    case ALGO_POLYTIMOS:     register_polytimos_algo     ( gate ); break;
-    case ALGO_POWER2B:       register_power2b_algo       ( gate ); break;
-    case ALGO_QUARK:         register_quark_algo         ( gate ); break;
-    case ALGO_QUBIT:         register_qubit_algo         ( gate ); break;
-    case ALGO_SCRYPT:        register_scrypt_algo        ( gate ); break;
-    case ALGO_SHA256D:       register_sha256d_algo       ( gate ); break;
-    case ALGO_SHA256Q:       register_sha256q_algo       ( gate ); break;
-    case ALGO_SHA256T:       register_sha256t_algo       ( gate ); break;
-    case ALGO_SHA3D:         register_sha3d_algo         ( gate ); break;
-    case ALGO_SHAVITE3:      register_shavite_algo       ( gate ); break;
-    case ALGO_SKEIN:         register_skein_algo         ( gate ); break;
-    case ALGO_SKEIN2:        register_skein2_algo        ( gate ); break;
-    case ALGO_SKUNK:         register_skunk_algo         ( gate ); break;
-    case ALGO_SONOA:         register_sonoa_algo         ( gate ); break;
-    case ALGO_TIMETRAVEL:    register_timetravel_algo    ( gate ); break;
-    case ALGO_TIMETRAVEL10:  register_timetravel10_algo  ( gate ); break;
-    case ALGO_TRIBUS:        register_tribus_algo        ( gate ); break;
-    case ALGO_VANILLA:       register_vanilla_algo       ( gate ); break;
-    case ALGO_VELTOR:        register_veltor_algo        ( gate ); break;
-    case ALGO_WHIRLPOOL:     register_whirlpool_algo     ( gate ); break;
-    case ALGO_WHIRLPOOLX:    register_whirlpoolx_algo    ( gate ); break;
-    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_X13BCD:        register_x13bcd_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_X16RV2:        register_x16rv2_algo        ( gate ); break;
-    case ALGO_X16RT:         register_x16rt_algo         ( gate ); break;
-    case ALGO_X16RT_VEIL:    register_x16rt_veil_algo    ( gate ); break;
-    case ALGO_X16S:          register_x16s_algo          ( gate ); break;
-    case ALGO_X17:           register_x17_algo           ( gate ); break;
-    case ALGO_X21S:          register_x21s_algo          ( gate ); break;
-    case ALGO_X22I:          register_x22i_algo          ( gate ); break;
-    case ALGO_X25X:          register_x25x_algo          ( gate ); break;
-    case ALGO_XEVAN:         register_xevan_algo         ( gate ); break;
-    case ALGO_YESCRYPT:      register_yescrypt_algo      ( gate ); break;
-    case ALGO_YESCRYPTR8:    register_yescryptr8_algo    ( gate ); break;
-    case ALGO_YESCRYPTR8G:   register_yescryptr8g_algo   ( gate ); break;
-    case ALGO_YESCRYPTR16:   register_yescryptr16_algo   ( gate ); break;
-    case ALGO_YESCRYPTR32:   register_yescryptr32_algo   ( gate ); break;
-    case ALGO_YESPOWER:      register_yespower_algo      ( gate ); break;
-    case ALGO_YESPOWERR16:   register_yespowerr16_algo   ( gate ); break;
-    case ALGO_YESPOWER_B2B:  register_yespower_b2b_algo  ( gate ); break;
-    case ALGO_ZR5:           register_zr5_algo           ( gate ); break;
+    case ALGO_ALLIUM:       rc = register_allium_algo        ( gate ); break;
+    case ALGO_ANIME:        rc = register_anime_algo         ( gate ); break;
+    case ALGO_ARGON2:       rc = register_argon2_algo        ( gate ); break;
+    case ALGO_ARGON2D250:   rc = register_argon2d_crds_algo  ( gate ); break;
+    case ALGO_ARGON2D500:   rc = register_argon2d_dyn_algo   ( gate ); break;
+    case ALGO_ARGON2D4096:  rc = register_argon2d4096_algo   ( gate ); break;
+    case ALGO_AXIOM:        rc = register_axiom_algo         ( gate ); break;
+    case ALGO_BLAKE:        rc = register_blake_algo         ( gate ); break;
+    case ALGO_BLAKE2B:      rc = register_blake2b_algo       ( gate ); break;
+    case ALGO_BLAKE2S:      rc = register_blake2s_algo       ( gate ); break;
+    case ALGO_BLAKECOIN:    rc = register_blakecoin_algo     ( gate ); break;
+    case ALGO_BMW512:       rc = register_bmw512_algo        ( gate ); break;
+    case ALGO_C11:          rc = register_c11_algo           ( gate ); break;
+    case ALGO_DECRED:       rc = register_decred_algo        ( gate ); break;
+    case ALGO_DEEP:         rc = register_deep_algo          ( gate ); break;
+    case ALGO_DMD_GR:       rc = register_dmd_gr_algo        ( gate ); break;
+    case ALGO_GROESTL:      rc = register_groestl_algo       ( gate ); break;
+    case ALGO_HEX:          rc = register_hex_algo           ( gate ); break;
+    case ALGO_HMQ1725:      rc = register_hmq1725_algo       ( gate ); break;
+    case ALGO_HODL:         rc = register_hodl_algo          ( gate ); break;
+    case ALGO_JHA:          rc = register_jha_algo           ( gate ); break;
+    case ALGO_KECCAK:       rc = register_keccak_algo        ( gate ); break;
+    case ALGO_KECCAKC:      rc = register_keccakc_algo       ( gate ); break;
+    case ALGO_LBRY:         rc = register_lbry_algo          ( gate ); break;
+    case ALGO_LYRA2H:       rc = register_lyra2h_algo        ( gate ); break;
+    case ALGO_LYRA2RE:      rc = register_lyra2re_algo       ( gate ); break;
+    case ALGO_LYRA2REV2:    rc = register_lyra2rev2_algo     ( gate ); break;
+    case ALGO_LYRA2REV3:    rc = register_lyra2rev3_algo     ( gate ); break;
+    case ALGO_LYRA2Z:       rc = register_lyra2z_algo        ( gate ); break;
+    case ALGO_LYRA2Z330:    rc = register_lyra2z330_algo     ( gate ); break;
+    case ALGO_M7M:          rc = register_m7m_algo           ( gate ); break;
+    case ALGO_MINOTAUR:     rc = register_minotaur_algo      ( gate ); break;
+    case ALGO_MYR_GR:       rc = register_myriad_algo        ( gate ); break;
+    case ALGO_NEOSCRYPT:    rc = register_neoscrypt_algo     ( gate ); break;
+    case ALGO_NIST5:        rc = register_nist5_algo         ( gate ); break;
+    case ALGO_PENTABLAKE:   rc = register_pentablake_algo    ( gate ); break;
+    case ALGO_PHI1612:      rc = register_phi1612_algo       ( gate ); break;
+    case ALGO_PHI2:         rc = register_phi2_algo          ( gate ); break;
+    case ALGO_POLYTIMOS:    rc = register_polytimos_algo     ( gate ); break;
+    case ALGO_POWER2B:      rc = register_power2b_algo       ( gate ); break;
+    case ALGO_QUARK:        rc = register_quark_algo         ( gate ); break;
+    case ALGO_QUBIT:        rc = register_qubit_algo         ( gate ); break;
+    case ALGO_SCRYPT:       rc = register_scrypt_algo        ( gate ); break;
+    case ALGO_SHA256D:      rc = register_sha256d_algo       ( gate ); break;
+    case ALGO_SHA256Q:      rc = register_sha256q_algo       ( gate ); break;
+    case ALGO_SHA256T:      rc = register_sha256t_algo       ( gate ); break;
+    case ALGO_SHA3D:        rc = register_sha3d_algo         ( gate ); break;
+    case ALGO_SHAVITE3:     rc = register_shavite_algo       ( gate ); break;
+    case ALGO_SKEIN:        rc = register_skein_algo         ( gate ); break;
+    case ALGO_SKEIN2:       rc = register_skein2_algo        ( gate ); break;
+    case ALGO_SKUNK:        rc = register_skunk_algo         ( gate ); break;
+    case ALGO_SONOA:        rc = register_sonoa_algo         ( gate ); break;
+    case ALGO_TIMETRAVEL:   rc = register_timetravel_algo    ( gate ); break;
+    case ALGO_TIMETRAVEL10: rc = register_timetravel10_algo  ( gate ); break;
+    case ALGO_TRIBUS:       rc = register_tribus_algo        ( gate ); break;
+    case ALGO_VANILLA:      rc = register_vanilla_algo       ( gate ); break;
+    case ALGO_VELTOR:       rc = register_veltor_algo        ( gate ); break;
+    case ALGO_VERTHASH:     rc = register_verthash_algo      ( gate ); break;
+    case ALGO_WHIRLPOOL:    rc = register_whirlpool_algo     ( gate ); break;
+    case ALGO_WHIRLPOOLX:   rc = register_whirlpoolx_algo    ( gate ); break;
+    case ALGO_X11:          rc = register_x11_algo           ( gate ); break;
+    case ALGO_X11EVO:       rc = register_x11evo_algo        ( gate ); break;
+    case ALGO_X11GOST:      rc = register_x11gost_algo       ( gate ); break;
+    case ALGO_X12:          rc = register_x12_algo           ( gate ); break;
+    case ALGO_X13:          rc = register_x13_algo           ( gate ); break;
+    case ALGO_X13BCD:       rc = register_x13bcd_algo        ( gate ); break;
+    case ALGO_X13SM3:       rc = register_x13sm3_algo        ( gate ); break;
+    case ALGO_X14:          rc = register_x14_algo           ( gate ); break;
+    case ALGO_X15:          rc = register_x15_algo           ( gate ); break;
+    case ALGO_X16R:         rc = register_x16r_algo          ( gate ); break;
+    case ALGO_X16RV2:       rc = register_x16rv2_algo        ( gate ); break;
+    case ALGO_X16RT:        rc = register_x16rt_algo         ( gate ); break;
+    case ALGO_X16RT_VEIL:   rc = register_x16rt_veil_algo    ( gate ); break;
+    case ALGO_X16S:         rc = register_x16s_algo          ( gate ); break;
+    case ALGO_X17:          rc = register_x17_algo           ( gate ); break;
+    case ALGO_X21S:         rc = register_x21s_algo          ( gate ); break;
+    case ALGO_X22I:         rc = register_x22i_algo          ( gate ); break;
+    case ALGO_X25X:         rc = register_x25x_algo          ( gate ); break;
+    case ALGO_XEVAN:        rc = register_xevan_algo         ( gate ); break;
+    case ALGO_YESCRYPT:     rc = register_yescrypt_05_algo   ( gate ); break;
+//    case ALGO_YESCRYPT:      register_yescrypt_algo      ( gate ); break;
+    case ALGO_YESCRYPTR8:   rc = register_yescryptr8_05_algo ( gate ); break;
+//    case ALGO_YESCRYPTR8:    register_yescryptr8_algo    ( gate ); break;
+    case ALGO_YESCRYPTR8G:  rc = register_yescryptr8g_algo   ( gate ); break;
+    case ALGO_YESCRYPTR16:  rc = register_yescryptr16_05_algo( gate ); break;
+//    case ALGO_YESCRYPTR16:   register_yescryptr16_algo   ( gate ); break;
+    case ALGO_YESCRYPTR32:  rc = register_yescryptr32_05_algo( gate ); break;
+//    case ALGO_YESCRYPTR32:   register_yescryptr32_algo   ( gate ); break;
+    case ALGO_YESPOWER:     rc = register_yespower_algo      ( gate ); break;
+    case ALGO_YESPOWERR16:  rc = register_yespowerr16_algo   ( gate ); break;
+    case ALGO_YESPOWER_B2B: rc = register_yespower_b2b_algo  ( gate ); break;
+    case ALGO_ZR5:          rc = register_zr5_algo           ( gate ); break;
   default:
-      applog(LOG_ERR,"FAIL: algo_gate registration failed, unknown algo %s.\n", algo_names[opt_algo] );
+      applog(LOG_ERR,"BUG: unregistered algorithm %s.\n", algo_names[opt_algo] );
      return false;
  } // switch

- // ensure required functions were defined.
-  if (  gate->scanhash == (void*)&null_scanhash )
+  if ( !rc )
  {
-    applog(LOG_ERR, "FAIL: Required algo_gate functions undefined\n");
+    applog(LOG_ERR, "FAIL: %s algorithm failed to initialize\n", algo_names[opt_algo] );
    return false;
  }
  return true;
@@ -415,7 +419,6 @@ void exec_hash_function( int algo, void *output, const void *pdata )
 const char* const algo_alias_map[][2] =
 {
 //   alias                proper
-  { "argon2d-crds",      "argon2d250"     },
  { "argon2d-dyn",       "argon2d500"     },
  { "argon2d-uis",       "argon2d4096"    },
  { "bcd",               "x13bcd"         },
@@ -430,7 +433,6 @@ const char* const algo_alias_map[][2] =
  { "flax",              "c11"            },
  { "hsr",               "x13sm3"         },
  { "jackpot",           "jha"            },
-  { "jane",              "scryptjane"     }, 
  { "lyra2",             "lyra2re"        },
  { "lyra2v2",           "lyra2rev2"      },
  { "lyra2v3",           "lyra2rev3"      },
--- a/algo-gate-api.h
+++ b/algo-gate-api.h
@@ -1,3 +1,6 @@
+#ifndef __ALGO_GATE_API_H__
+#define __ALGO_GATE_API_H__ 1
+
 #include <stdlib.h>
 #include <stdbool.h>
 #include <stdint.h>
@@ -90,10 +93,11 @@ typedef  uint32_t set_t;
 #define AES_OPT          2  
 #define SSE42_OPT        4
 #define AVX_OPT          8   // Sandybridge
-#define AVX2_OPT      0x10   // Haswell
-#define SHA_OPT       0x20   // sha256 (Ryzen, Ice Lake)
-#define AVX512_OPT    0x40   // AVX512- F, VL, DQ, BW (Skylake-X)
-#define VAES_OPT      0x80   // VAES (Ice Lake)
+#define AVX2_OPT      0x10   // Haswell, Zen1
+#define SHA_OPT       0x20   // Zen1, Icelake (sha256)
+#define AVX512_OPT    0x40   // Skylake-X (AVX512[F,VL,DQ,BW])
+#define VAES_OPT      0x80   // Icelake (VAES & AVX512)
+#define VAES256_OPT   0x100  // Zen3 (VAES without AVX512)


 // return set containing all elements from sets a & b
@@ -111,17 +115,17 @@ inline bool set_excl ( set_t a, set_t b ) { return (a & b) == 0; }
 typedef struct
 {
 // Mandatory functions, one of these is mandatory. If a generic scanhash
-// is used a custom hash function must be registered, with a custom scanhash
-// the custom hash function can be called directly and doesn't need to be
-// registered in the gate. 
+// is used a custom target hash function must be registered, with a custom
+// scanhash the target hash function can be called directly and doesn't need
+// to be registered with the gate. 
 int ( *scanhash ) ( struct work*, uint32_t, uint64_t*, struct thr_info* );

 int ( *hash )     ( void*, const void*, int );

 //optional, safe to use default in most cases

-// Allocate thread local buffers and other initialization specific to miner
-// threads.
+// Called once by each miner thread to allocate thread local buffers and
+// other initialization specific to miner threads.
 bool ( *miner_thread_init )     ( int );

 // Get thread local copy of blockheader with unique nonce.
@@ -149,7 +153,7 @@ void ( *build_stratum_request ) ( char*, struct work*, struct stratum_ctx* );

 char* ( *malloc_txs_request )   ( struct work* );

-// Big or little
+// Big endian or little endian
 void ( *set_work_data_endian )  ( struct work* );

 double ( *calc_network_diff )   ( struct work* );
@@ -161,7 +165,7 @@ bool ( *ready_to_mine )         ( struct work*, struct stratum_ctx*, int );
 bool ( *do_this_thread )        ( int );

 // After do_this_thread
-void ( *resync_threads )        ( struct work* );
+void ( *resync_threads )        ( int, struct work* );

 // No longer needed
 json_t* (*longpoll_rpc_call)      ( CURL*, int*, char* );
@@ -259,7 +263,7 @@ int scanhash_8way_64in_32out( struct work *work, uint32_t max_nonce,
 #endif

 // displays warning
-int null_hash    ();
+int null_hash();

 // optional safe targets, default listed first unless noted.

@@ -280,7 +284,7 @@ void std_be_build_stratum_request( char *req, struct work *work );

 char* std_malloc_txs_request( struct work *work );

-// Default is do_nothing (assumed LE)
+// Default is do_nothing, little endian is assumed
 void set_work_data_big_endian( struct work *work );

 double std_calc_network_diff( struct work *work );
@@ -318,3 +322,4 @@ void exec_hash_function( int algo, void *output, const void *pdata );
 // algo name if valid alias, NULL if invalid alias or algo.
 void get_algo_alias( char **algo_or_alias );

+#endif
--- a/algo/argon2/argon2d/blake2/blamka-round-opt.h
+++ b/algo/argon2/argon2d/blake2/blamka-round-opt.h
@@ -328,7 +328,7 @@ static BLAKE2_INLINE __m128i fBlaMka(__m128i x, __m128i y) {

 #include <immintrin.h>

-#define ror64(x, n) _mm512_ror_epi64((x), (n))
+#define ROR64(x, n) _mm512_ror_epi64((x), (n))

 static __m512i muladd(__m512i x, __m512i y)
 {
@@ -344,8 +344,8 @@ static __m512i muladd(__m512i x, __m512i y)
        D0 = _mm512_xor_si512(D0, A0); \
        D1 = _mm512_xor_si512(D1, A1); \
 \
-        D0 = ror64(D0, 32); \
-        D1 = ror64(D1, 32); \
+        D0 = ROR64(D0, 32); \
+        D1 = ROR64(D1, 32); \
 \
        C0 = muladd(C0, D0); \
        C1 = muladd(C1, D1); \
@@ -353,8 +353,8 @@ static __m512i muladd(__m512i x, __m512i y)
        B0 = _mm512_xor_si512(B0, C0); \
        B1 = _mm512_xor_si512(B1, C1); \
 \
-        B0 = ror64(B0, 24); \
-        B1 = ror64(B1, 24); \
+        B0 = ROR64(B0, 24); \
+        B1 = ROR64(B1, 24); \
    } while ((void)0, 0)

 #define G2(A0, B0, C0, D0, A1, B1, C1, D1) \
@@ -365,8 +365,8 @@ static __m512i muladd(__m512i x, __m512i y)
        D0 = _mm512_xor_si512(D0, A0); \
        D1 = _mm512_xor_si512(D1, A1); \
 \
-        D0 = ror64(D0, 16); \
-        D1 = ror64(D1, 16); \
+        D0 = ROR64(D0, 16); \
+        D1 = ROR64(D1, 16); \
 \
        C0 = muladd(C0, D0); \
        C1 = muladd(C1, D1); \
@@ -374,8 +374,8 @@ static __m512i muladd(__m512i x, __m512i y)
        B0 = _mm512_xor_si512(B0, C0); \
        B1 = _mm512_xor_si512(B1, C1); \
 \
-        B0 = ror64(B0, 63); \
-        B1 = ror64(B1, 63); \
+        B0 = ROR64(B0, 63); \
+        B1 = ROR64(B1, 63); \
    } while ((void)0, 0)

 #define DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
--- a/algo/blake/blake-hash-4way.h
+++ b/algo/blake/blake-hash-4way.h
@@ -180,6 +180,7 @@ void blake512_8way_update( void *cc, const void *data, size_t len );
 void blake512_8way_close( void *cc, void *dst );
 void blake512_8way_full( blake_8way_big_context *sc, void * dst,
                        const void *data, size_t len );
+void blake512_8way_hash_le80( void *hash, const void *data );

 #endif  // AVX512
 #endif  // AVX2
--- a/algo/blake/blake256-hash-4way.c
+++ b/algo/blake/blake256-hash-4way.c
@@ -669,14 +669,14 @@ do { \
      ROUND_S_8WAY(2); \
      ROUND_S_8WAY(3); \
   } \
-   H0 = _mm256_xor_si256( _mm256_xor_si256( V8, V0 ), H0 ); \
-   H1 = _mm256_xor_si256( _mm256_xor_si256( V9, V1 ), H1 ); \
-   H2 = _mm256_xor_si256( _mm256_xor_si256( VA, V2 ), H2 ); \
-   H3 = _mm256_xor_si256( _mm256_xor_si256( VB, V3 ), H3 ); \
-   H4 = _mm256_xor_si256( _mm256_xor_si256( VC, V4 ), H4 ); \
-   H5 = _mm256_xor_si256( _mm256_xor_si256( VD, V5 ), H5 ); \
-   H6 = _mm256_xor_si256( _mm256_xor_si256( VE, V6 ), H6 ); \
-   H7 = _mm256_xor_si256( _mm256_xor_si256( VF, V7 ), H7 ); \
+   H0 = mm256_xor3( V8, V0, H0 ); \
+   H1 = mm256_xor3( V9, V1, H1 ); \
+   H2 = mm256_xor3( VA, V2, H2 ); \
+   H3 = mm256_xor3( VB, V3, H3 ); \
+   H4 = mm256_xor3( VC, V4, H4 ); \
+   H5 = mm256_xor3( VD, V5, H5 ); \
+   H6 = mm256_xor3( VE, V6, H6 ); \
+   H7 = mm256_xor3( VF, V7, H7 ); \
 } while (0)


@@ -808,14 +808,14 @@ do { \
      ROUND_S_16WAY(2); \
      ROUND_S_16WAY(3); \
   } \
-   H0 = _mm512_xor_si512( _mm512_xor_si512( V8, V0 ), H0 ); \
-   H1 = _mm512_xor_si512( _mm512_xor_si512( V9, V1 ), H1 ); \
-   H2 = _mm512_xor_si512( _mm512_xor_si512( VA, V2 ), H2 ); \
-   H3 = _mm512_xor_si512( _mm512_xor_si512( VB, V3 ), H3 ); \
-   H4 = _mm512_xor_si512( _mm512_xor_si512( VC, V4 ), H4 ); \
-   H5 = _mm512_xor_si512( _mm512_xor_si512( VD, V5 ), H5 ); \
-   H6 = _mm512_xor_si512( _mm512_xor_si512( VE, V6 ), H6 ); \
-   H7 = _mm512_xor_si512( _mm512_xor_si512( VF, V7 ), H7 ); \
+   H0 = mm512_xor3( V8, V0, H0 ); \
+   H1 = mm512_xor3( V9, V1, H1 ); \
+   H2 = mm512_xor3( VA, V2, H2 ); \
+   H3 = mm512_xor3( VB, V3, H3 ); \
+   H4 = mm512_xor3( VC, V4, H4 ); \
+   H5 = mm512_xor3( VD, V5, H5 ); \
+   H6 = mm512_xor3( VE, V6, H6 ); \
+   H7 = mm512_xor3( VF, V7, H7 ); \
 } while (0)

 #endif
--- a/algo/blake/blake2b-hash-4way.c
+++ b/algo/blake/blake2b-hash-4way.c
@@ -122,14 +122,14 @@ static void blake2b_8way_compress( blake2b_8way_ctx *ctx, int last )
      B2B8W_G( 3, 4,  9, 14, m[ sigma[i][14] ], m[ sigma[i][15] ] );
   }

-   ctx->h[0] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[0], v[0] ), v[ 8] );
-   ctx->h[1] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[1], v[1] ), v[ 9] );
-   ctx->h[2] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[2], v[2] ), v[10] );
-   ctx->h[3] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[3], v[3] ), v[11] );
-   ctx->h[4] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[4], v[4] ), v[12] );
-   ctx->h[5] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[5], v[5] ), v[13] );
-   ctx->h[6] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[6], v[6] ), v[14] );
-   ctx->h[7] = _mm512_xor_si512( _mm512_xor_si512( ctx->h[7], v[7] ), v[15] );
+   ctx->h[0] = mm512_xor3( ctx->h[0], v[0], v[ 8] );
+   ctx->h[1] = mm512_xor3( ctx->h[1], v[1], v[ 9] );
+   ctx->h[2] = mm512_xor3( ctx->h[2], v[2], v[10] );
+   ctx->h[3] = mm512_xor3( ctx->h[3], v[3], v[11] );
+   ctx->h[4] = mm512_xor3( ctx->h[4], v[4], v[12] );
+   ctx->h[5] = mm512_xor3( ctx->h[5], v[5], v[13] );
+   ctx->h[6] = mm512_xor3( ctx->h[6], v[6], v[14] );
+   ctx->h[7] = mm512_xor3( ctx->h[7], v[7], v[15] );
 }

 int blake2b_8way_init( blake2b_8way_ctx *ctx )
--- a/algo/blake/blake2b-hash-4way.h
+++ b/algo/blake/blake2b-hash-4way.h
@@ -17,7 +17,7 @@

 #if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

-ALIGN(128) typedef struct {
+typedef struct ALIGN( 64 ) {
   __m512i b[16]; // input buffer
   __m512i h[8];  // chained state
   uint64_t t[2];  // total number of bytes
@@ -35,7 +35,7 @@ void blake2b_8way_final( blake2b_8way_ctx *ctx, void *out );
 #if defined(__AVX2__)

 // state context
-ALIGN(128) typedef struct {
+typedef struct ALIGN( 64 ) {
 	__m256i b[16]; // input buffer
 	__m256i h[8];  // chained state
 	uint64_t t[2];  // total number of bytes
--- a/algo/blake/blake2s-gate.h
+++ b/algo/blake/blake2s-gate.h
@@ -4,7 +4,6 @@
 #include <stdint.h>
 #include "algo-gate-api.h"

-//#if defined(__SSE4_2__)
 #if defined(__SSE2__)
  #define BLAKE2S_4WAY
 #endif
@@ -27,8 +26,6 @@ int scanhash_blake2s_16way( struct work *work, uint32_t max_nonce,

 #elif defined (BLAKE2S_8WAY)

-//#if defined(BLAKE2S_8WAY)
-
 void blake2s_8way_hash( void *state, const void *input );
 int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
                         uint64_t *hashes_done, struct thr_info *mythr );
--- a/algo/blake/blake2s-hash-4way.c
+++ b/algo/blake/blake2s-hash-4way.c
@@ -368,7 +368,7 @@ do { \
   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] );
+      S->h[i] = mm256_xor3( S->h[i], v[i], v[i + 8] );

 #undef G8W
 #undef ROUND8W
@@ -566,7 +566,7 @@ do { \
   ROUND16W( 9 );

   for( size_t i = 0; i < 8; ++i )
-      S->h[i] = _mm512_xor_si512( _mm512_xor_si512( S->h[i], v[i] ), v[i + 8] );
+      S->h[i] = mm512_xor3( S->h[i], v[i], v[i + 8] );

 #undef G16W
 #undef ROUND16W
--- a/algo/blake/blake2s-hash-4way.h
+++ b/algo/blake/blake2s-hash-4way.h
@@ -60,7 +60,7 @@ typedef struct __blake2s_nway_param
 } blake2s_nway_param;
 #pragma pack(pop)

-ALIGN( 64 ) typedef struct __blake2s_4way_state
+typedef struct ALIGN( 64 ) __blake2s_4way_state
 {
   __m128i h[8];
   uint8_t  buf[ BLAKE2S_BLOCKBYTES * 4 ];
@@ -80,7 +80,7 @@ int blake2s_4way_full_blocks( blake2s_4way_state *S, void *out,

 #if defined(__AVX2__)

-ALIGN( 64 ) typedef struct __blake2s_8way_state
+typedef struct ALIGN( 64 ) __blake2s_8way_state
 {
   __m256i h[8];
   uint8_t  buf[ BLAKE2S_BLOCKBYTES * 8 ];
@@ -101,7 +101,7 @@ int blake2s_8way_full_blocks( blake2s_8way_state *S, void *out,

 #if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

-ALIGN( 128 ) typedef struct __blake2s_16way_state
+typedef struct ALIGN( 64 ) __blake2s_16way_state
 {
   __m512i h[8];
   uint8_t  buf[ BLAKE2S_BLOCKBYTES * 16 ];
--- a/algo/blake/blake512-hash-4way.c
+++ b/algo/blake/blake512-hash-4way.c
@@ -293,10 +293,6 @@ static const sph_u64 CB[16] = {
      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)
@@ -310,10 +306,6 @@ static const sph_u64 CB[16] = {
      (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)
@@ -348,7 +340,6 @@ static const sph_u64 CB[16] = {

 #define DECL_STATE64_8WAY \
   __m512i H0, H1, H2, H3, H4, H5, H6, H7; \
-        __m512i S0, S1, S2, S3; \
   uint64_t T0, T1;

 #define COMPRESS64_8WAY( buf )   do \
@@ -366,10 +357,10 @@ static const sph_u64 CB[16] = {
  V5 = H5; \
  V6 = H6; \
  V7 = H7; \
-  V8 = _mm512_xor_si512( S0, m512_const1_64( CB0 ) );  \
-  V9 = _mm512_xor_si512( S1, m512_const1_64( CB1 ) );  \
-  VA = _mm512_xor_si512( S2, m512_const1_64( CB2 ) );  \
-  VB = _mm512_xor_si512( S3, m512_const1_64( CB3 ) );  \
+  V8 = m512_const1_64( CB0 );  \
+  V9 = m512_const1_64( CB1 );  \
+  VA = m512_const1_64( CB2 );  \
+  VB = m512_const1_64( CB3 );  \
  VC = _mm512_xor_si512( _mm512_set1_epi64( T0 ), \
                         m512_const1_64( CB4 ) );  \
  VD = _mm512_xor_si512( _mm512_set1_epi64( T0 ), \
@@ -414,14 +405,14 @@ static const sph_u64 CB[16] = {
  ROUND_B_8WAY(3); \
  ROUND_B_8WAY(4); \
  ROUND_B_8WAY(5); \
-  H0 = mm512_xor4( V8, V0, S0, H0 ); \
-  H1 = mm512_xor4( V9, V1, S1, H1 ); \
-  H2 = mm512_xor4( VA, V2, S2, H2 ); \
-  H3 = mm512_xor4( VB, V3, S3, H3 ); \
-  H4 = mm512_xor4( VC, V4, S0, H4 ); \
-  H5 = mm512_xor4( VD, V5, S1, H5 ); \
-  H6 = mm512_xor4( VE, V6, S2, H6 ); \
-  H7 = mm512_xor4( VF, V7, S3, H7 ); \
+  H0 = mm512_xor3( V8, V0, H0 ); \
+  H1 = mm512_xor3( V9, V1, H1 ); \
+  H2 = mm512_xor3( VA, V2, H2 ); \
+  H3 = mm512_xor3( VB, V3, H3 ); \
+  H4 = mm512_xor3( VC, V4, H4 ); \
+  H5 = mm512_xor3( VD, V5, H5 ); \
+  H6 = mm512_xor3( VE, V6, H6 ); \
+  H7 = mm512_xor3( VF, V7, H7 ); \
 } while (0)

 void blake512_8way_compress( blake_8way_big_context *sc )
@@ -440,10 +431,10 @@ void blake512_8way_compress( blake_8way_big_context *sc )
  V5 = sc->H[5];
  V6 = sc->H[6];
  V7 = sc->H[7];
-  V8 = _mm512_xor_si512( sc->S[0], m512_const1_64( CB0 ) );
-  V9 = _mm512_xor_si512( sc->S[1], m512_const1_64( CB1 ) );
-  VA = _mm512_xor_si512( sc->S[2], m512_const1_64( CB2 ) );
-  VB = _mm512_xor_si512( sc->S[3], m512_const1_64( CB3 ) );
+  V8 = m512_const1_64( CB0 );
+  V9 = m512_const1_64( CB1 );
+  VA = m512_const1_64( CB2 );
+  VB = m512_const1_64( CB3 );
  VC = _mm512_xor_si512( _mm512_set1_epi64( sc->T0 ),
                            m512_const1_64( CB4 ) );
  VD = _mm512_xor_si512( _mm512_set1_epi64( sc->T0 ),
@@ -492,19 +483,18 @@ void blake512_8way_compress( blake_8way_big_context *sc )
  ROUND_B_8WAY(4);
  ROUND_B_8WAY(5);

-  sc->H[0] = mm512_xor4( V8, V0, sc->S[0], sc->H[0] );
-  sc->H[1] = mm512_xor4( V9, V1, sc->S[1], sc->H[1] );
-  sc->H[2] = mm512_xor4( VA, V2, sc->S[2], sc->H[2] );
-  sc->H[3] = mm512_xor4( VB, V3, sc->S[3], sc->H[3] );
-  sc->H[4] = mm512_xor4( VC, V4, sc->S[0], sc->H[4] );
-  sc->H[5] = mm512_xor4( VD, V5, sc->S[1], sc->H[5] );
-  sc->H[6] = mm512_xor4( VE, V6, sc->S[2], sc->H[6] );
-  sc->H[7] = mm512_xor4( VF, V7, sc->S[3], sc->H[7] );
+  sc->H[0] = mm512_xor3( V8, V0, sc->H[0] );
+  sc->H[1] = mm512_xor3( V9, V1, sc->H[1] );
+  sc->H[2] = mm512_xor3( VA, V2, sc->H[2] );
+  sc->H[3] = mm512_xor3( VB, V3, sc->H[3] );
+  sc->H[4] = mm512_xor3( VC, V4, sc->H[4] );
+  sc->H[5] = mm512_xor3( VD, V5, sc->H[5] );
+  sc->H[6] = mm512_xor3( VE, V6, sc->H[6] );
+  sc->H[7] = mm512_xor3( VF, V7, sc->H[7] );
 }

 void blake512_8way_init( blake_8way_big_context *sc )
 {
-   __m512i zero = m512_zero;
   casti_m512i( sc->H, 0 ) = m512_const1_64( 0x6A09E667F3BCC908 );
   casti_m512i( sc->H, 1 ) = m512_const1_64( 0xBB67AE8584CAA73B );
   casti_m512i( sc->H, 2 ) = m512_const1_64( 0x3C6EF372FE94F82B );
@@ -514,11 +504,6 @@ void blake512_8way_init( blake_8way_big_context *sc )
   casti_m512i( sc->H, 6 ) = m512_const1_64( 0x1F83D9ABFB41BD6B );
   casti_m512i( sc->H, 7 ) = m512_const1_64( 0x5BE0CD19137E2179 );

-   casti_m512i( sc->S, 0 ) = zero;
-   casti_m512i( sc->S, 1 ) = zero;
-   casti_m512i( sc->S, 2 ) = zero;
-   casti_m512i( sc->S, 3 ) = zero;
-
   sc->T0 = sc->T1 = 0;
   sc->ptr = 0;
 }
@@ -641,11 +626,6 @@ void blake512_8way_full( blake_8way_big_context *sc, void * dst,
   casti_m512i( sc->H, 6 ) = m512_const1_64( 0x1F83D9ABFB41BD6B );
   casti_m512i( sc->H, 7 ) = m512_const1_64( 0x5BE0CD19137E2179 );

-   casti_m512i( sc->S, 0 ) = m512_zero;
-   casti_m512i( sc->S, 1 ) = m512_zero;
-   casti_m512i( sc->S, 2 ) = m512_zero;
-   casti_m512i( sc->S, 3 ) = m512_zero;
-
   sc->T0 = sc->T1 = 0;
   sc->ptr = 0;

@@ -740,7 +720,6 @@ blake512_8way_close(void *cc, void *dst)

 #define DECL_STATE64_4WAY \
 	__m256i H0, H1, H2, H3, H4, H5, H6, H7; \
-        __m256i S0, S1, S2, S3; \
 	uint64_t T0, T1;

 #define COMPRESS64_4WAY   do \
@@ -758,10 +737,10 @@ blake512_8way_close(void *cc, void *dst)
  V5 = H5; \
  V6 = H6; \
  V7 = H7; \
-  V8 = _mm256_xor_si256( S0, m256_const1_64( CB0 ) );  \
-  V9 = _mm256_xor_si256( S1, m256_const1_64( CB1 ) );  \
-  VA = _mm256_xor_si256( S2, m256_const1_64( CB2 ) );  \
-  VB = _mm256_xor_si256( S3, m256_const1_64( CB3 ) );  \
+  V8 = m256_const1_64( CB0 );  \
+  V9 = m256_const1_64( CB1 );  \
+  VA = m256_const1_64( CB2 );  \
+  VB = m256_const1_64( CB3 );  \
  VC = _mm256_xor_si256( _mm256_set1_epi64x( T0 ), \
                         m256_const1_64( CB4 ) );  \
  VD = _mm256_xor_si256( _mm256_set1_epi64x( T0 ), \
@@ -804,14 +783,14 @@ blake512_8way_close(void *cc, void *dst)
  ROUND_B_4WAY(3); \
  ROUND_B_4WAY(4); \
  ROUND_B_4WAY(5); \
-  H0 = mm256_xor4( V8, V0, S0, H0 ); \
-  H1 = mm256_xor4( V9, V1, S1, H1 ); \
-  H2 = mm256_xor4( VA, V2, S2, H2 ); \
-  H3 = mm256_xor4( VB, V3, S3, H3 ); \
-  H4 = mm256_xor4( VC, V4, S0, H4 ); \
-  H5 = mm256_xor4( VD, V5, S1, H5 ); \
-  H6 = mm256_xor4( VE, V6, S2, H6 ); \
-  H7 = mm256_xor4( VF, V7, S3, H7 ); \
+  H0 = mm256_xor3( V8, V0, H0 ); \
+  H1 = mm256_xor3( V9, V1, H1 ); \
+  H2 = mm256_xor3( VA, V2, H2 ); \
+  H3 = mm256_xor3( VB, V3, H3 ); \
+  H4 = mm256_xor3( VC, V4, H4 ); \
+  H5 = mm256_xor3( VD, V5, H5 ); \
+  H6 = mm256_xor3( VE, V6, H6 ); \
+  H7 = mm256_xor3( VF, V7, H7 ); \
 } while (0)


@@ -831,10 +810,10 @@ void blake512_4way_compress( blake_4way_big_context *sc )
  V5 = sc->H[5];
  V6 = sc->H[6];
  V7 = sc->H[7];
-  V8 = _mm256_xor_si256( sc->S[0], m256_const1_64( CB0 ) );
-  V9 = _mm256_xor_si256( sc->S[1], m256_const1_64( CB1 ) );
-  VA = _mm256_xor_si256( sc->S[2], m256_const1_64( CB2 ) );
-  VB = _mm256_xor_si256( sc->S[3], m256_const1_64( CB3 ) );
+  V8 = m256_const1_64( CB0 );
+  V9 = m256_const1_64( CB1 );
+  VA = m256_const1_64( CB2 );
+  VB = m256_const1_64( CB3 );
  VC = _mm256_xor_si256( _mm256_set1_epi64x( sc->T0 ),
                             m256_const1_64( CB4 ) );
  VD = _mm256_xor_si256( _mm256_set1_epi64x( sc->T0 ),
@@ -880,19 +859,18 @@ void blake512_4way_compress( blake_4way_big_context *sc )
  ROUND_B_4WAY(4);
  ROUND_B_4WAY(5);

-  sc->H[0] = mm256_xor4( V8, V0, sc->S[0], sc->H[0] );
-  sc->H[1] = mm256_xor4( V9, V1, sc->S[1], sc->H[1] );
-  sc->H[2] = mm256_xor4( VA, V2, sc->S[2], sc->H[2] );
-  sc->H[3] = mm256_xor4( VB, V3, sc->S[3], sc->H[3] );
-  sc->H[4] = mm256_xor4( VC, V4, sc->S[0], sc->H[4] );
-  sc->H[5] = mm256_xor4( VD, V5, sc->S[1], sc->H[5] );
-  sc->H[6] = mm256_xor4( VE, V6, sc->S[2], sc->H[6] );
-  sc->H[7] = mm256_xor4( VF, V7, sc->S[3], sc->H[7] );
+  sc->H[0] = mm256_xor3( V8, V0, sc->H[0] );
+  sc->H[1] = mm256_xor3( V9, V1, sc->H[1] );
+  sc->H[2] = mm256_xor3( VA, V2, sc->H[2] );
+  sc->H[3] = mm256_xor3( VB, V3, sc->H[3] );
+  sc->H[4] = mm256_xor3( VC, V4, sc->H[4] );
+  sc->H[5] = mm256_xor3( VD, V5, sc->H[5] );
+  sc->H[6] = mm256_xor3( VE, V6, sc->H[6] );
+  sc->H[7] = mm256_xor3( VF, V7, sc->H[7] );
 }

 void blake512_4way_init( blake_4way_big_context *sc )
 {
-   __m256i zero = m256_zero;
   casti_m256i( sc->H, 0 ) = m256_const1_64( 0x6A09E667F3BCC908 );
   casti_m256i( sc->H, 1 ) = m256_const1_64( 0xBB67AE8584CAA73B );
   casti_m256i( sc->H, 2 ) = m256_const1_64( 0x3C6EF372FE94F82B );
@@ -902,11 +880,6 @@ void blake512_4way_init( blake_4way_big_context *sc )
   casti_m256i( sc->H, 6 ) = m256_const1_64( 0x1F83D9ABFB41BD6B );
   casti_m256i( sc->H, 7 ) = m256_const1_64( 0x5BE0CD19137E2179 );

-   casti_m256i( sc->S, 0 ) = zero;
-   casti_m256i( sc->S, 1 ) = zero;
-   casti_m256i( sc->S, 2 ) = zero;
-   casti_m256i( sc->S, 3 ) = zero;
-
   sc->T0 = sc->T1 = 0;
   sc->ptr = 0;
 }
@@ -1026,11 +999,6 @@ void blake512_4way_full( blake_4way_big_context *sc, void * dst,
   casti_m256i( sc->H, 6 ) = m256_const1_64( 0x1F83D9ABFB41BD6B );
   casti_m256i( sc->H, 7 ) = m256_const1_64( 0x5BE0CD19137E2179 );

-   casti_m256i( sc->S, 0 ) = m256_zero;
-   casti_m256i( sc->S, 1 ) = m256_zero;
-   casti_m256i( sc->S, 2 ) = m256_zero;
-   casti_m256i( sc->S, 3 ) = m256_zero;
-
   sc->T0 = sc->T1 = 0;
   sc->ptr = 0;

--- a/algo/blake/sph-blake2s.c
+++ b/algo/blake/sph-blake2s.c
@@ -323,7 +323,7 @@ int blake2s_final( blake2s_state *S, uint8_t *out, uint8_t outlen )

 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];
+	blake2s_state S;

 	/* Verify parameters */
 	if ( NULL == in ) return -1;
@@ -334,15 +334,15 @@ int blake2s( uint8_t *out, const void *in, const void *key, const uint8_t outlen

 	if( keylen > 0 )
 	{
-		if( blake2s_init_key( S, outlen, key, keylen ) < 0 ) return -1;
+		if( blake2s_init_key( &S, outlen, key, keylen ) < 0 ) return -1;
 	}
 	else
 	{
-		if( blake2s_init( S, outlen ) < 0 ) return -1;
+		if( blake2s_init( &S, outlen ) < 0 ) return -1;
 	}

-	blake2s_update( S, ( uint8_t * )in, inlen );
-	blake2s_final( S, out, outlen );
+	blake2s_update( &S, ( uint8_t * )in, inlen );
+	blake2s_final( &S, out, outlen );
 	return 0;
 }

--- a/algo/blake/sph-blake2s.h
+++ b/algo/blake/sph-blake2s.h
@@ -116,7 +116,7 @@ extern "C" {
 		uint8_t  personal[BLAKE2S_PERSONALBYTES];  // 32
 	} blake2s_param;

-	ALIGN( 64 ) typedef struct __blake2s_state
+	typedef struct ALIGN( 64 ) __blake2s_state
 	{
 		uint32_t h[8];
 		uint32_t t[2];
--- a/algo/blake/sph_blake2b.h
+++ b/algo/blake/sph_blake2b.h
@@ -18,7 +18,7 @@
 #endif

 // state context
-ALIGN(64) typedef struct {
+typedef ALIGN(64) struct {
 	uint8_t b[128]; // input buffer
 	uint64_t h[8];  // chained state
 	uint64_t t[2];  // total number of bytes
--- a/algo/bmw/bmw256-hash-4way.c
+++ b/algo/bmw/bmw256-hash-4way.c
@@ -867,40 +867,35 @@ void compress_small_8way( const __m256i *M, const __m256i H[16],
   qt[30] = expand2s8( qt, M, H, 30 );
   qt[31] = expand2s8( qt, M, H, 31 );

-   xl = _mm256_xor_si256(
-              mm256_xor4( qt[16], qt[17], qt[18], qt[19] ),
-              mm256_xor4( qt[20], qt[21], qt[22], qt[23] ) );
-   xh = _mm256_xor_si256( xl,  _mm256_xor_si256(
-                 mm256_xor4( qt[24], qt[25], qt[26], qt[27] ),
-                 mm256_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
+   xl = mm256_xor3( mm256_xor3( qt[16], qt[17], qt[18] ),
+                    mm256_xor3( qt[19], qt[20], qt[21] ),
+                    _mm256_xor_si256( qt[22], qt[23] ) );
+
+   xh = mm256_xor3( mm256_xor3( xl,     qt[24], qt[25] ),
+                    mm256_xor3( qt[26], qt[27], qt[28] ),
+                    mm256_xor3( qt[29], qt[30], qt[31] ) );

 #define DH1L( m, sl, sr, a, b, c ) \
-   _mm256_add_epi32( \
-               _mm256_xor_si256( M[m], \
-                  _mm256_xor_si256( _mm256_slli_epi32( xh, sl ), \
-                                    _mm256_srli_epi32( qt[a], sr ) ) ), \
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[b] ), qt[c] ) )
+   _mm256_add_epi32( mm256_xor3( M[m], _mm256_slli_epi32( xh, sl ), \
+                                       _mm256_srli_epi32( qt[a], sr ) ), \
+                     mm256_xor3( xl, qt[b], qt[c] ) )

 #define DH1R( m, sl, sr, a, b, c ) \
-   _mm256_add_epi32( \
-               _mm256_xor_si256( M[m], \
-                  _mm256_xor_si256( _mm256_srli_epi32( xh, sl ), \
-                                    _mm256_slli_epi32( qt[a], sr ) ) ), \
-               _mm256_xor_si256( _mm256_xor_si256( xl, qt[b] ), qt[c] ) )
+   _mm256_add_epi32( mm256_xor3( M[m], _mm256_srli_epi32( xh, sl ), \
+                                       _mm256_slli_epi32( qt[a], sr ) ), \
+                     mm256_xor3( xl, qt[b], qt[c] ) )

 #define DH2L( m, rl, sl, h, a, b, c ) \
   _mm256_add_epi32( _mm256_add_epi32( \
-       mm256_rol_32( dH[h], rl ), \
-          _mm256_xor_si256( _mm256_xor_si256( xh, qt[a] ), M[m] )), \
-                 _mm256_xor_si256( _mm256_slli_epi32( xl, sl ), \
-                                   _mm256_xor_si256( qt[b], qt[c] ) ) );
+                        mm256_rol_32( dH[h], rl ), \
+                        mm256_xor3( xh, qt[a], M[m] ) ), \
+                     mm256_xor3( _mm256_slli_epi32( xl, sl ), qt[b], qt[c] ) ) 

 #define DH2R( m, rl, sr, h, a, b, c ) \
   _mm256_add_epi32( _mm256_add_epi32( \
-       mm256_rol_32( dH[h], rl ), \
-          _mm256_xor_si256( _mm256_xor_si256( xh, qt[a] ), M[m] )), \
-                 _mm256_xor_si256( _mm256_srli_epi32( xl, sr ), \
-                                   _mm256_xor_si256( qt[b], qt[c] ) ) );
+                        mm256_rol_32( dH[h], rl ), \
+                        mm256_xor3( xh, qt[a], M[m] ) ), \
+                     mm256_xor3( _mm256_srli_epi32( xl, sr ), qt[b], qt[c] ) )

   dH[ 0] = DH1L(  0,  5,  5, 16, 24, 0 );
   dH[ 1] = DH1R(  1,  7,  8, 17, 25, 1 );
@@ -924,88 +919,6 @@ void compress_small_8way( const __m256i *M, const __m256i H[16],
 #undef DH2L
 #undef DH2R

-/*   
-   dH[ 0] = _mm256_add_epi32(
-                 _mm256_xor_si256( M[0],
-                      _mm256_xor_si256( _mm256_slli_epi32( xh, 5 ),
-                                        _mm256_srli_epi32( qt[16], 5 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[24] ), qt[ 0] ));
-   dH[ 1] = _mm256_add_epi32(
-                 _mm256_xor_si256( M[1],
-                      _mm256_xor_si256( _mm256_srli_epi32( xh, 7 ),
-                                        _mm256_slli_epi32( qt[17], 8 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[25] ), qt[ 1] ));
-   dH[ 2] = _mm256_add_epi32(
-                 _mm256_xor_si256( M[2],
-                      _mm256_xor_si256( _mm256_srli_epi32( xh, 5 ),
-                                        _mm256_slli_epi32( qt[18], 5 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[26] ), qt[ 2] ));
-   dH[ 3] = _mm256_add_epi32(
-                 _mm256_xor_si256( M[3],
-                      _mm256_xor_si256( _mm256_srli_epi32( xh, 1 ),
-                                        _mm256_slli_epi32( qt[19], 5 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[27] ), qt[ 3] ));
-   dH[ 4] = _mm256_add_epi32(
-                 _mm256_xor_si256( M[4],
-                      _mm256_xor_si256( _mm256_srli_epi32( xh, 3 ),
-                                        _mm256_slli_epi32( qt[20], 0 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[28] ), qt[ 4] ));
-   dH[ 5] = _mm256_add_epi32(
-                 _mm256_xor_si256( M[5],
-                      _mm256_xor_si256( _mm256_slli_epi32( xh, 6 ),
-                                        _mm256_srli_epi32( qt[21], 6 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[29] ), qt[ 5] ));
-   dH[ 6] = _mm256_add_epi32(
-                 _mm256_xor_si256( M[6],
-                      _mm256_xor_si256( _mm256_srli_epi32( xh, 4 ),
-                                        _mm256_slli_epi32( qt[22], 6 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[30] ), qt[ 6] ));
-   dH[ 7] = _mm256_add_epi32(
-                 _mm256_xor_si256( M[7],
-                      _mm256_xor_si256( _mm256_srli_epi32( xh, 11 ),
-                                        _mm256_slli_epi32( qt[23], 2 ) ) ),
-                 _mm256_xor_si256( _mm256_xor_si256( xl, qt[31] ), qt[ 7] ));
-   dH[ 8] = _mm256_add_epi32( _mm256_add_epi32(
-                 mm256_rol_32( dH[4], 9 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[24] ), M[ 8] )),
-                 _mm256_xor_si256( _mm256_slli_epi32( xl, 8 ),
-                                   _mm256_xor_si256( qt[23], qt[ 8] ) ) );
-   dH[ 9] = _mm256_add_epi32( _mm256_add_epi32(
-                 mm256_rol_32( dH[5], 10 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[25] ), M[ 9] )),
-                 _mm256_xor_si256( _mm256_srli_epi32( xl, 6 ),
-                                   _mm256_xor_si256( qt[16], qt[ 9] ) ) );
-   dH[10] = _mm256_add_epi32( _mm256_add_epi32(
-                 mm256_rol_32( dH[6], 11 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[26] ), M[10] )),
-                 _mm256_xor_si256( _mm256_slli_epi32( xl, 6 ),
-                                   _mm256_xor_si256( qt[17], qt[10] ) ) );
-   dH[11] = _mm256_add_epi32( _mm256_add_epi32(
-                 mm256_rol_32( dH[7], 12 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[27] ), M[11] )),
-                 _mm256_xor_si256( _mm256_slli_epi32( xl, 4 ),
-                                   _mm256_xor_si256( qt[18], qt[11] ) ) );
-   dH[12] = _mm256_add_epi32( _mm256_add_epi32(
-                 mm256_rol_32( dH[0], 13 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[28] ), M[12] )),
-                 _mm256_xor_si256( _mm256_srli_epi32( xl, 3 ),
-                                   _mm256_xor_si256( qt[19], qt[12] ) ) );
-   dH[13] = _mm256_add_epi32( _mm256_add_epi32(
-                 mm256_rol_32( dH[1], 14 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[29] ), M[13] )),
-                 _mm256_xor_si256( _mm256_srli_epi32( xl, 4 ),
-                                   _mm256_xor_si256( qt[20], qt[13] ) ) );
-   dH[14] = _mm256_add_epi32( _mm256_add_epi32(
-                 mm256_rol_32( dH[2], 15 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[30] ), M[14] )),
-                 _mm256_xor_si256( _mm256_srli_epi32( xl, 7 ),
-                                   _mm256_xor_si256( qt[21], qt[14] ) ) );
-   dH[15] = _mm256_add_epi32( _mm256_add_epi32(
-                 mm256_rol_32( dH[3], 16 ),
-                 _mm256_xor_si256( _mm256_xor_si256( xh, qt[31] ), M[15] )),
-                 _mm256_xor_si256( _mm256_srli_epi32( xl, 2 ),
-                                   _mm256_xor_si256( qt[22], qt[15] ) ) );
-*/
 }

 static const __m256i final_s8[16] =
@@ -1422,40 +1335,35 @@ void compress_small_16way( const __m512i *M, const __m512i H[16],
   qt[30] = expand2s16( qt, M, H, 30 );
   qt[31] = expand2s16( qt, M, H, 31 );

-   xl = _mm512_xor_si512(
-              mm512_xor4( qt[16], qt[17], qt[18], qt[19] ),
-              mm512_xor4( qt[20], qt[21], qt[22], qt[23] ) );
-   xh = _mm512_xor_si512( xl,  _mm512_xor_si512(
-                 mm512_xor4( qt[24], qt[25], qt[26], qt[27] ),
-                 mm512_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
+   xl = mm512_xor3( mm512_xor3( qt[16], qt[17], qt[18] ),
+                    mm512_xor3( qt[19], qt[20], qt[21] ),
+                    _mm512_xor_si512( qt[22], qt[23] ) );
+
+   xh = mm512_xor3( mm512_xor3( xl,     qt[24], qt[25] ),
+                    mm512_xor3( qt[26], qt[27], qt[28] ),
+                    mm512_xor3( qt[29], qt[30], qt[31] ) );

 #define DH1L( m, sl, sr, a, b, c ) \
-   _mm512_add_epi32( \
-               _mm512_xor_si512( M[m], \
-                  _mm512_xor_si512( _mm512_slli_epi32( xh, sl ), \
-                                    _mm512_srli_epi32( qt[a], sr ) ) ), \
-               _mm512_xor_si512( _mm512_xor_si512( xl, qt[b] ), qt[c] ) )
+   _mm512_add_epi32( mm512_xor3( M[m], _mm512_slli_epi32( xh, sl ), \
+                                       _mm512_srli_epi32( qt[a], sr ) ), \
+                     mm512_xor3( xl, qt[b], qt[c] ) )

 #define DH1R( m, sl, sr, a, b, c ) \
-   _mm512_add_epi32( \
-               _mm512_xor_si512( M[m], \
-                  _mm512_xor_si512( _mm512_srli_epi32( xh, sl ), \
-                                    _mm512_slli_epi32( qt[a], sr ) ) ), \
-               _mm512_xor_si512( _mm512_xor_si512( xl, qt[b] ), qt[c] ) )
+   _mm512_add_epi32( mm512_xor3( M[m], _mm512_srli_epi32( xh, sl ), \
+                                       _mm512_slli_epi32( qt[a], sr ) ), \
+                     mm512_xor3( xl, qt[b], qt[c] ) )

 #define DH2L( m, rl, sl, h, a, b, c ) \
   _mm512_add_epi32( _mm512_add_epi32( \
-       mm512_rol_32( dH[h], rl ), \
-          _mm512_xor_si512( _mm512_xor_si512( xh, qt[a] ), M[m] )), \
-                 _mm512_xor_si512( _mm512_slli_epi32( xl, sl ), \
-                                   _mm512_xor_si512( qt[b], qt[c] ) ) );
+                        mm512_rol_32( dH[h], rl ), \
+                        mm512_xor3( xh, qt[a], M[m] ) ), \
+                     mm512_xor3( _mm512_slli_epi32( xl, sl ), qt[b], qt[c] ) ) 

 #define DH2R( m, rl, sr, h, a, b, c ) \
   _mm512_add_epi32( _mm512_add_epi32( \
-       mm512_rol_32( dH[h], rl ), \
-          _mm512_xor_si512( _mm512_xor_si512( xh, qt[a] ), M[m] )), \
-                 _mm512_xor_si512( _mm512_srli_epi32( xl, sr ), \
-                                   _mm512_xor_si512( qt[b], qt[c] ) ) );
+                        mm512_rol_32( dH[h], rl ), \
+                        mm512_xor3( xh, qt[a], M[m] ) ), \
+                     mm512_xor3( _mm512_srli_epi32( xl, sr ), qt[b], qt[c] ) )

   dH[ 0] = DH1L(  0,  5,  5, 16, 24, 0 );
   dH[ 1] = DH1R(  1,  7,  8, 17, 25, 1 );
--- a/algo/bmw/bmw512-hash-4way.c
+++ b/algo/bmw/bmw512-hash-4way.c
@@ -594,22 +594,15 @@ void bmw512_2way_close( bmw_2way_big_context *ctx, void *dst )
 #define rb6(x)    mm256_rol_64( x, 43 ) 
 #define rb7(x)    mm256_rol_64( x, 53 ) 

-#define rol_off_64( M, j, off ) \
-   mm256_rol_64( M[ ( (j) + (off) ) & 0xF ] , \
-                  ( ( (j) + (off) ) & 0xF ) + 1 )
+#define rol_off_64( M, j ) \
+   mm256_rol_64( M[ (j) & 0xF ], ( (j) & 0xF ) + 1 )

-#define add_elt_b( M, H, j ) \
-   _mm256_xor_si256( \
-      _mm256_add_epi64( \
-            _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 ) & 0xF ] )
+#define add_elt_b( mj0, mj3, mj10, h, K ) \
+  _mm256_xor_si256( h, _mm256_add_epi64( K, \
+              _mm256_sub_epi64( _mm256_add_epi64( mj0, mj3 ), mj10 ) ) )

-
-#define expand1b( qt, M, H, i ) \
-   _mm256_add_epi64( mm256_add4_64( \
+#define expand1_b( qt, i ) \
+   mm256_add4_64( \
      mm256_add4_64( sb1( qt[ (i)-16 ] ), sb2( qt[ (i)-15 ] ), \
                     sb3( qt[ (i)-14 ] ), sb0( qt[ (i)-13 ] )), \
      mm256_add4_64( sb1( qt[ (i)-12 ] ), sb2( qt[ (i)-11 ] ), \
@@ -617,11 +610,10 @@ void bmw512_2way_close( bmw_2way_big_context *ctx, void *dst )
      mm256_add4_64( sb1( qt[ (i)- 8 ] ), sb2( qt[ (i)- 7 ] ), \
                     sb3( qt[ (i)- 6 ] ), sb0( qt[ (i)- 5 ] )), \
      mm256_add4_64( sb1( qt[ (i)- 4 ] ), sb2( qt[ (i)- 3 ] ), \
-                     sb3( qt[ (i)- 2 ] ), sb0( qt[ (i)- 1 ] ) ) ), \
-      add_elt_b( M, H, (i)-16 ) )
+                     sb3( qt[ (i)- 2 ] ), sb0( qt[ (i)- 1 ] ) ) )

-#define expand2b( qt, M, H, i) \
-   _mm256_add_epi64( mm256_add4_64( \
+#define expand2_b( qt, i) \
+   mm256_add4_64( \
      mm256_add4_64( qt[ (i)-16 ], rb1( qt[ (i)-15 ] ), \
                     qt[ (i)-14 ], rb2( qt[ (i)-13 ] ) ), \
      mm256_add4_64( qt[ (i)-12 ], rb3( qt[ (i)-11 ] ), \
@@ -629,159 +621,98 @@ void bmw512_2way_close( bmw_2way_big_context *ctx, void *dst )
      mm256_add4_64( qt[ (i)- 8 ], rb5( qt[ (i)- 7 ] ), \
                     qt[ (i)- 6 ], rb6( qt[ (i)- 5 ] ) ), \
      mm256_add4_64( qt[ (i)- 4 ], rb7( qt[ (i)- 3 ] ), \
-                     sb4( qt[ (i)- 2 ] ), sb5( qt[ (i)- 1 ] ) ) ), \
-      add_elt_b( M, H, (i)-16 ) )
-
-
+                     sb4( qt[ (i)- 2 ] ), sb5( qt[ (i)- 1 ] ) ) )

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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


 void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
 {
   __m256i qt[32], xl, xh;
+   __m256i mh[16];
+   int i;
+
+   for ( i = 0; i < 16; i++ )
+      mh[i] = _mm256_xor_si256( M[i], H[i] );

   qt[ 0] = _mm256_add_epi64( sb0( Wb0 ), H[ 1] ); 
   qt[ 1] = _mm256_add_epi64( sb1( Wb1 ), H[ 2] ); 
@@ -799,22 +730,60 @@ void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
   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 ); 
-   qt[19] = expand2b( qt, M, H, 19 ); 
-   qt[20] = expand2b( qt, M, H, 20 ); 
-   qt[21] = expand2b( qt, M, H, 21 ); 
-   qt[22] = expand2b( qt, M, H, 22 ); 
-   qt[23] = expand2b( qt, M, H, 23 ); 
-   qt[24] = expand2b( qt, M, H, 24 ); 
-   qt[25] = expand2b( qt, M, H, 25 ); 
-   qt[26] = expand2b( qt, M, H, 26 ); 
-   qt[27] = expand2b( qt, M, H, 27 ); 
-   qt[28] = expand2b( qt, M, H, 28 ); 
-   qt[29] = expand2b( qt, M, H, 29 ); 
-   qt[30] = expand2b( qt, M, H, 30 ); 
-   qt[31] = expand2b( qt, M, H, 31 ); 
+
+   __m256i mj[16];
+   for ( i = 0; i < 16; i++ )
+      mj[i] = rol_off_64( M, i );
+
+   qt[16] = add_elt_b( mj[ 0], mj[ 3], mj[10], H[ 7],
+              (const __m256i)_mm256_set1_epi64x( 16 * 0x0555555555555555ULL ) );
+   qt[17] = add_elt_b( mj[ 1], mj[ 4], mj[11], H[ 8],
+              (const __m256i)_mm256_set1_epi64x( 17 * 0x0555555555555555ULL ) );
+   qt[18] = add_elt_b( mj[ 2], mj[ 5], mj[12], H[ 9],
+              (const __m256i)_mm256_set1_epi64x( 18 * 0x0555555555555555ULL ) );
+   qt[19] = add_elt_b( mj[ 3], mj[ 6], mj[13], H[10],
+              (const __m256i)_mm256_set1_epi64x( 19 * 0x0555555555555555ULL ) );
+   qt[20] = add_elt_b( mj[ 4], mj[ 7], mj[14], H[11],
+              (const __m256i)_mm256_set1_epi64x( 20 * 0x0555555555555555ULL ) );
+   qt[21] = add_elt_b( mj[ 5], mj[ 8], mj[15], H[12],
+              (const __m256i)_mm256_set1_epi64x( 21 * 0x0555555555555555ULL ) );
+   qt[22] = add_elt_b( mj[ 6], mj[ 9], mj[ 0], H[13],
+              (const __m256i)_mm256_set1_epi64x( 22 * 0x0555555555555555ULL ) );
+   qt[23] = add_elt_b( mj[ 7], mj[10], mj[ 1], H[14],
+              (const __m256i)_mm256_set1_epi64x( 23 * 0x0555555555555555ULL ) );
+   qt[24] = add_elt_b( mj[ 8], mj[11], mj[ 2], H[15],
+              (const __m256i)_mm256_set1_epi64x( 24 * 0x0555555555555555ULL ) );
+   qt[25] = add_elt_b( mj[ 9], mj[12], mj[ 3], H[ 0],
+              (const __m256i)_mm256_set1_epi64x( 25 * 0x0555555555555555ULL ) );
+   qt[26] = add_elt_b( mj[10], mj[13], mj[ 4], H[ 1],
+              (const __m256i)_mm256_set1_epi64x( 26 * 0x0555555555555555ULL ) );
+   qt[27] = add_elt_b( mj[11], mj[14], mj[ 5], H[ 2],
+              (const __m256i)_mm256_set1_epi64x( 27 * 0x0555555555555555ULL ) );
+   qt[28] = add_elt_b( mj[12], mj[15], mj[ 6], H[ 3],
+              (const __m256i)_mm256_set1_epi64x( 28 * 0x0555555555555555ULL ) );
+   qt[29] = add_elt_b( mj[13], mj[ 0], mj[ 7], H[ 4],
+              (const __m256i)_mm256_set1_epi64x( 29 * 0x0555555555555555ULL ) );
+   qt[30] = add_elt_b( mj[14], mj[ 1], mj[ 8], H[ 5],
+              (const __m256i)_mm256_set1_epi64x( 30 * 0x0555555555555555ULL ) );
+   qt[31] = add_elt_b( mj[15], mj[ 2], mj[ 9], H[ 6],
+              (const __m256i)_mm256_set1_epi64x( 31 * 0x0555555555555555ULL ) );
+
+   qt[16] = _mm256_add_epi64( qt[16], expand1_b( qt, 16 ) );
+   qt[17] = _mm256_add_epi64( qt[17], expand1_b( qt, 17 ) );
+   qt[18] = _mm256_add_epi64( qt[18], expand2_b( qt, 18 ) );
+   qt[19] = _mm256_add_epi64( qt[19], expand2_b( qt, 19 ) );
+   qt[20] = _mm256_add_epi64( qt[20], expand2_b( qt, 20 ) );
+   qt[21] = _mm256_add_epi64( qt[21], expand2_b( qt, 21 ) );
+   qt[22] = _mm256_add_epi64( qt[22], expand2_b( qt, 22 ) );
+   qt[23] = _mm256_add_epi64( qt[23], expand2_b( qt, 23 ) );
+   qt[24] = _mm256_add_epi64( qt[24], expand2_b( qt, 24 ) );
+   qt[25] = _mm256_add_epi64( qt[25], expand2_b( qt, 25 ) );
+   qt[26] = _mm256_add_epi64( qt[26], expand2_b( qt, 26 ) );
+   qt[27] = _mm256_add_epi64( qt[27], expand2_b( qt, 27 ) );
+   qt[28] = _mm256_add_epi64( qt[28], expand2_b( qt, 28 ) );
+   qt[29] = _mm256_add_epi64( qt[29], expand2_b( qt, 29 ) );
+   qt[30] = _mm256_add_epi64( qt[30], expand2_b( qt, 30 ) );
+   qt[31] = _mm256_add_epi64( qt[31], expand2_b( qt, 31 ) );

   xl = _mm256_xor_si256(
           mm256_xor4( qt[16], qt[17], qt[18], qt[19] ), 
@@ -823,7 +792,6 @@ void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
           mm256_xor4( qt[24], qt[25], qt[26], qt[27] ),
           mm256_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );

-
 #define DH1L( m, sl, sr, a, b, c ) \
   _mm256_add_epi64( \
               _mm256_xor_si256( M[m], \
@@ -1066,21 +1034,15 @@ bmw512_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
 #define r8b6(x)    mm512_rol_64( x, 43 )
 #define r8b7(x)    mm512_rol_64( x, 53 )

-#define rol8w_off_64( M, j, off ) \
-   mm512_rol_64( M[ ( (j) + (off) ) & 0xF ] , \
-                  ( ( (j) + (off) ) & 0xF ) + 1 )
+#define rol8w_off_64( M, j ) \
+   mm512_rol_64( M[ (j) & 0xF ], ( (j) & 0xF ) + 1 )

-#define add_elt_b8( M, H, j ) \
-   _mm512_xor_si512( \
-      _mm512_add_epi64( \
-            _mm512_sub_epi64( _mm512_add_epi64( rol8w_off_64( M, j, 0 ), \
-                                                rol8w_off_64( M, j, 3 ) ), \
-                             rol8w_off_64( M, j, 10 ) ), \
-            _mm512_set1_epi64( ( (j) + 16 ) * 0x0555555555555555ULL ) ), \
-       H[ ( (j)+7 ) & 0xF ] )
+#define add_elt_b8( mj0, mj3, mj10, h, K ) \
+  _mm512_xor_si512( h, _mm512_add_epi64( K, \
+              _mm512_sub_epi64( _mm512_add_epi64( mj0, mj3 ), mj10 ) ) )

-#define expand1b8( qt, M, H, i ) \
-   _mm512_add_epi64( mm512_add4_64( \
+#define expand1_b8( qt, i ) \
+   mm512_add4_64( \
      mm512_add4_64( s8b1( qt[ (i)-16 ] ), s8b2( qt[ (i)-15 ] ), \
                     s8b3( qt[ (i)-14 ] ), s8b0( qt[ (i)-13 ] )), \
      mm512_add4_64( s8b1( qt[ (i)-12 ] ), s8b2( qt[ (i)-11 ] ), \
@@ -1088,11 +1050,10 @@ bmw512_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
      mm512_add4_64( s8b1( qt[ (i)- 8 ] ), s8b2( qt[ (i)- 7 ] ), \
                     s8b3( qt[ (i)- 6 ] ), s8b0( qt[ (i)- 5 ] )), \
      mm512_add4_64( s8b1( qt[ (i)- 4 ] ), s8b2( qt[ (i)- 3 ] ), \
-                     s8b3( qt[ (i)- 2 ] ), s8b0( qt[ (i)- 1 ] ) ) ), \
-      add_elt_b8( M, H, (i)-16 ) )
+                     s8b3( qt[ (i)- 2 ] ), s8b0( qt[ (i)- 1 ] ) ) )

-#define expand2b8( qt, M, H, i) \
-   _mm512_add_epi64( mm512_add4_64( \
+#define expand2_b8( qt, i) \
+   mm512_add4_64( \
      mm512_add4_64( qt[ (i)-16 ], r8b1( qt[ (i)-15 ] ), \
                     qt[ (i)-14 ], r8b2( qt[ (i)-13 ] ) ), \
      mm512_add4_64( qt[ (i)-12 ], r8b3( qt[ (i)-11 ] ), \
@@ -1100,157 +1061,97 @@ bmw512_4way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
      mm512_add4_64( qt[ (i)- 8 ], r8b5( qt[ (i)- 7 ] ), \
                     qt[ (i)- 6 ], r8b6( qt[ (i)- 5 ] ) ), \
      mm512_add4_64( qt[ (i)- 4 ], r8b7( qt[ (i)- 3 ] ), \
-                     s8b4( qt[ (i)- 2 ] ), s8b5( qt[ (i)- 1 ] ) ) ), \
-      add_elt_b8( M, H, (i)-16 ) )
+                     s8b4( qt[ (i)- 2 ] ), s8b5( qt[ (i)- 1 ] ) ) )

 #define W8b0 \
   _mm512_add_epi64( \
-      _mm512_add_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 5], H[ 5] ), \
-                           _mm512_xor_si512( M[ 7], H[ 7] ) ), \
-         _mm512_xor_si512( M[10], H[10] ) ), \
-      _mm512_add_epi64( _mm512_xor_si512( M[13], H[13] ), \
-                        _mm512_xor_si512( M[14], H[14] ) ) )
+      _mm512_add_epi64( _mm512_sub_epi64( mh[ 5], mh[ 7] ), mh[10] ), \
+      _mm512_add_epi64( mh[13], mh[14] ) )

 #define W8b1 \
   _mm512_add_epi64( \
-       _mm512_add_epi64( \
-          _mm512_sub_epi64( _mm512_xor_si512( M[ 6], H[ 6] ), \
-                            _mm512_xor_si512( M[ 8], H[ 8] ) ), \
-          _mm512_xor_si512( M[11], H[11] ) ), \
-       _mm512_sub_epi64( _mm512_xor_si512( M[14], H[14] ), \
-                         _mm512_xor_si512( M[15], H[15] ) ) )
+         _mm512_add_epi64( _mm512_sub_epi64( mh[ 6], mh[ 8] ), mh[11] ), \
+         _mm512_sub_epi64( mh[14], mh[15] ) )

 #define W8b2 \
   _mm512_sub_epi64( \
-      _mm512_add_epi64( \
-         _mm512_add_epi64( _mm512_xor_si512( M[ 0], H[ 0] ), \
-                           _mm512_xor_si512( M[ 7], H[ 7] ) ), \
-         _mm512_xor_si512( M[ 9], H[ 9] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[12], H[12] ), \
-                        _mm512_xor_si512( M[15], H[15] ) ) )
+      _mm512_add_epi64( _mm512_add_epi64( mh[ 0], mh[ 7] ), mh[ 9] ), \
+      _mm512_sub_epi64( mh[12], mh[15] ) )

 #define W8b3 \
   _mm512_sub_epi64( \
-      _mm512_add_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 0], H[ 0] ), \
-                           _mm512_xor_si512( M[ 1], H[ 1] ) ), \
-         _mm512_xor_si512( M[ 8], H[ 8] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[10], H[10] ), \
-                        _mm512_xor_si512( M[13], H[13] ) ) )
+      _mm512_add_epi64( _mm512_sub_epi64( mh[ 0], mh[ 1] ), mh[ 8] ), \
+      _mm512_sub_epi64( mh[10], mh[13] ) )

 #define W8b4 \
   _mm512_sub_epi64( \
-      _mm512_add_epi64( \
-         _mm512_add_epi64( _mm512_xor_si512( M[ 1], H[ 1] ), \
-                           _mm512_xor_si512( M[ 2], H[ 2] ) ), \
-         _mm512_xor_si512( M[ 9], H[ 9] ) ), \
-      _mm512_add_epi64( _mm512_xor_si512( M[11], H[11] ), \
-                        _mm512_xor_si512( M[14], H[14] ) ) )
+      _mm512_add_epi64( _mm512_add_epi64( mh[ 1], mh[ 2] ), mh[ 9] ), \
+      _mm512_add_epi64( mh[11], mh[14] ) )

 #define W8b5 \
   _mm512_sub_epi64( \
-      _mm512_add_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 3], H[ 3] ), \
-                           _mm512_xor_si512( M[ 2], H[ 2] ) ), \
-         _mm512_xor_si512( M[10], H[10] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[12], H[12] ), \
-                        _mm512_xor_si512( M[15], H[15] ) ) )
+      _mm512_add_epi64( _mm512_sub_epi64( mh[ 3], mh[ 2] ), mh[10] ), \
+      _mm512_sub_epi64( mh[12], mh[15] ) )

 #define W8b6 \
   _mm512_sub_epi64( \
-      _mm512_sub_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 4], H[ 4] ), \
-                           _mm512_xor_si512( M[ 0], H[ 0] ) ), \
-         _mm512_xor_si512( M[ 3], H[ 3] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[11], H[11] ), \
-                        _mm512_xor_si512( M[13], H[13] ) ) )
+         _mm512_sub_epi64( _mm512_sub_epi64( mh[ 4], mh[ 0] ), mh[ 3] ), \
+      _mm512_sub_epi64( mh[11], mh[13] ) )

 #define W8b7 \
   _mm512_sub_epi64( \
-      _mm512_sub_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 1], H[ 1] ), \
-                           _mm512_xor_si512( M[ 4], H[ 4] ) ), \
-         _mm512_xor_si512( M[ 5], H[ 5] ) ), \
-      _mm512_add_epi64( _mm512_xor_si512( M[12], H[12] ), \
-                        _mm512_xor_si512( M[14], H[14] ) ) )
+      _mm512_sub_epi64( _mm512_sub_epi64( mh[ 1], mh[ 4] ), mh[ 5] ), \
+      _mm512_add_epi64( mh[12], mh[14] ) )

 #define W8b8 \
   _mm512_add_epi64( \
-      _mm512_sub_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 2], H[ 2] ), \
-                           _mm512_xor_si512( M[ 5], H[ 5] ) ), \
-         _mm512_xor_si512( M[ 6], H[ 6] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[13], H[13] ), \
-                        _mm512_xor_si512( M[15], H[15] ) ) )
+      _mm512_sub_epi64( _mm512_sub_epi64( mh[ 2], mh[ 5] ), mh[ 6] ), \
+      _mm512_sub_epi64( mh[13], mh[15] ) )

 #define W8b9 \
   _mm512_sub_epi64( \
-      _mm512_add_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 0], H[ 0] ), \
-                           _mm512_xor_si512( M[ 3], H[ 3] ) ), \
-         _mm512_xor_si512( M[ 6], H[ 6] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[ 7], H[ 7] ), \
-                        _mm512_xor_si512( M[14], H[14] ) ) )
+      _mm512_add_epi64( _mm512_sub_epi64( mh[ 0], mh[ 3] ), mh[ 6] ), \
+      _mm512_sub_epi64( mh[ 7], mh[14] ) )

 #define W8b10 \
   _mm512_sub_epi64( \
-      _mm512_sub_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 8], H[ 8] ), \
-                           _mm512_xor_si512( M[ 1], H[ 1] ) ), \
-         _mm512_xor_si512( M[ 4], H[ 4] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[ 7], H[ 7] ), \
-                        _mm512_xor_si512( M[15], H[15] ) ) )
+      _mm512_sub_epi64( _mm512_sub_epi64( mh[ 8], mh[ 1] ), mh[ 4] ), \
+      _mm512_sub_epi64( mh[ 7], mh[15] ) )

 #define W8b11 \
   _mm512_sub_epi64( \
-      _mm512_sub_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 8], H[ 8] ), \
-                           _mm512_xor_si512( M[ 0], H[ 0] ) ), \
-         _mm512_xor_si512( M[ 2], H[ 2] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[ 5], H[ 5] ), \
-                        _mm512_xor_si512( M[ 9], H[ 9] ) ) )
+      _mm512_sub_epi64( _mm512_sub_epi64( mh[ 8], mh[ 0] ), mh[ 2] ), \
+      _mm512_sub_epi64( mh[ 5], mh[ 9] ) )

 #define W8b12 \
   _mm512_sub_epi64( \
-      _mm512_sub_epi64( \
-         _mm512_add_epi64( _mm512_xor_si512( M[ 1], H[ 1] ), \
-                           _mm512_xor_si512( M[ 3], H[ 3] ) ), \
-         _mm512_xor_si512( M[ 6], H[ 6] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[ 9], H[ 9] ), \
-                        _mm512_xor_si512( M[10], H[10] ) ) )
+      _mm512_sub_epi64( _mm512_add_epi64( mh[ 1], mh[ 3] ), mh[ 6] ), \
+      _mm512_sub_epi64( mh[ 9], mh[10] ) )

 #define W8b13 \
   _mm512_add_epi64( \
-      _mm512_add_epi64( \
-         _mm512_add_epi64( _mm512_xor_si512( M[ 2], H[ 2] ), \
-                           _mm512_xor_si512( M[ 4], H[ 4] ) ), \
-         _mm512_xor_si512( M[ 7], H[ 7] ) ), \
-      _mm512_add_epi64( _mm512_xor_si512( M[10], H[10] ), \
-                        _mm512_xor_si512( M[11], H[11] ) ) )
+      _mm512_add_epi64( _mm512_add_epi64( mh[ 2], mh[ 4] ), mh[ 7] ), \
+      _mm512_add_epi64( mh[10], mh[11] ) )

 #define W8b14 \
   _mm512_sub_epi64( \
-      _mm512_add_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[ 3], H[ 3] ), \
-                           _mm512_xor_si512( M[ 5], H[ 5] ) ), \
-         _mm512_xor_si512( M[ 8], H[ 8] ) ), \
-      _mm512_add_epi64( _mm512_xor_si512( M[11], H[11] ), \
-                        _mm512_xor_si512( M[12], H[12] ) ) )
+      _mm512_add_epi64( _mm512_sub_epi64( mh[ 3], mh[ 5] ), mh[ 8] ), \
+      _mm512_add_epi64( mh[11], mh[12] ) )

 #define W8b15 \
   _mm512_sub_epi64( \
-      _mm512_sub_epi64( \
-         _mm512_sub_epi64( _mm512_xor_si512( M[12], H[12] ), \
-                           _mm512_xor_si512( M[ 4], H[4] ) ), \
-         _mm512_xor_si512( M[ 6], H[ 6] ) ), \
-      _mm512_sub_epi64( _mm512_xor_si512( M[ 9], H[ 9] ), \
-                        _mm512_xor_si512( M[13], H[13] ) ) )
+      _mm512_sub_epi64( _mm512_sub_epi64( mh[12], mh[ 4] ), mh[ 6] ), \
+      _mm512_sub_epi64( mh[ 9], mh[13] ) )

 void compress_big_8way( const __m512i *M, const __m512i H[16],
                        __m512i dH[16] )
 {
   __m512i qt[32], xl, xh;
+   __m512i mh[16];
+   int i;
+
+   for ( i = 0; i < 16; i++ )
+      mh[i] = _mm512_xor_si512( M[i], H[i] );

   qt[ 0] = _mm512_add_epi64( s8b0( W8b0 ), H[ 1] );
   qt[ 1] = _mm512_add_epi64( s8b1( W8b1 ), H[ 2] );
@@ -1268,57 +1169,90 @@ void compress_big_8way( const __m512i *M, const __m512i H[16],
   qt[13] = _mm512_add_epi64( s8b3( W8b13), H[14] );
   qt[14] = _mm512_add_epi64( s8b4( W8b14), H[15] );
   qt[15] = _mm512_add_epi64( s8b0( W8b15), H[ 0] );
-   qt[16] = expand1b8( qt, M, H, 16 );
-   qt[17] = expand1b8( qt, M, H, 17 );
-   qt[18] = expand2b8( qt, M, H, 18 );
-   qt[19] = expand2b8( qt, M, H, 19 );
-   qt[20] = expand2b8( qt, M, H, 20 );
-   qt[21] = expand2b8( qt, M, H, 21 );
-   qt[22] = expand2b8( qt, M, H, 22 );
-   qt[23] = expand2b8( qt, M, H, 23 );
-   qt[24] = expand2b8( qt, M, H, 24 );
-   qt[25] = expand2b8( qt, M, H, 25 );
-   qt[26] = expand2b8( qt, M, H, 26 );
-   qt[27] = expand2b8( qt, M, H, 27 );
-   qt[28] = expand2b8( qt, M, H, 28 );
-   qt[29] = expand2b8( qt, M, H, 29 );
-   qt[30] = expand2b8( qt, M, H, 30 );
-   qt[31] = expand2b8( qt, M, H, 31 );

-   xl = _mm512_xor_si512(
-           mm512_xor4( qt[16], qt[17], qt[18], qt[19] ),
-           mm512_xor4( qt[20], qt[21], qt[22], qt[23] ) );
-   xh = _mm512_xor_si512( xl, _mm512_xor_si512(
-           mm512_xor4( qt[24], qt[25], qt[26], qt[27] ),
-           mm512_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
+   __m512i mj[16];
+   for ( i = 0; i < 16; i++ )
+      mj[i] = rol8w_off_64( M, i );
+
+   qt[16] = add_elt_b8( mj[ 0], mj[ 3], mj[10], H[ 7],
+              (const __m512i)_mm512_set1_epi64( 16 * 0x0555555555555555ULL ) );
+   qt[17] = add_elt_b8( mj[ 1], mj[ 4], mj[11], H[ 8],
+              (const __m512i)_mm512_set1_epi64( 17 * 0x0555555555555555ULL ) );
+   qt[18] = add_elt_b8( mj[ 2], mj[ 5], mj[12], H[ 9],
+              (const __m512i)_mm512_set1_epi64( 18 * 0x0555555555555555ULL ) );
+   qt[19] = add_elt_b8( mj[ 3], mj[ 6], mj[13], H[10],
+              (const __m512i)_mm512_set1_epi64( 19 * 0x0555555555555555ULL ) );
+   qt[20] = add_elt_b8( mj[ 4], mj[ 7], mj[14], H[11],
+              (const __m512i)_mm512_set1_epi64( 20 * 0x0555555555555555ULL ) );
+   qt[21] = add_elt_b8( mj[ 5], mj[ 8], mj[15], H[12],
+              (const __m512i)_mm512_set1_epi64( 21 * 0x0555555555555555ULL ) );
+   qt[22] = add_elt_b8( mj[ 6], mj[ 9], mj[ 0], H[13],
+              (const __m512i)_mm512_set1_epi64( 22 * 0x0555555555555555ULL ) );
+   qt[23] = add_elt_b8( mj[ 7], mj[10], mj[ 1], H[14],
+              (const __m512i)_mm512_set1_epi64( 23 * 0x0555555555555555ULL ) );
+   qt[24] = add_elt_b8( mj[ 8], mj[11], mj[ 2], H[15],
+              (const __m512i)_mm512_set1_epi64( 24 * 0x0555555555555555ULL ) );
+   qt[25] = add_elt_b8( mj[ 9], mj[12], mj[ 3], H[ 0],
+              (const __m512i)_mm512_set1_epi64( 25 * 0x0555555555555555ULL ) );
+   qt[26] = add_elt_b8( mj[10], mj[13], mj[ 4], H[ 1],
+              (const __m512i)_mm512_set1_epi64( 26 * 0x0555555555555555ULL ) );
+   qt[27] = add_elt_b8( mj[11], mj[14], mj[ 5], H[ 2],
+              (const __m512i)_mm512_set1_epi64( 27 * 0x0555555555555555ULL ) );
+   qt[28] = add_elt_b8( mj[12], mj[15], mj[ 6], H[ 3],
+              (const __m512i)_mm512_set1_epi64( 28 * 0x0555555555555555ULL ) );
+   qt[29] = add_elt_b8( mj[13], mj[ 0], mj[ 7], H[ 4],
+              (const __m512i)_mm512_set1_epi64( 29 * 0x0555555555555555ULL ) );
+   qt[30] = add_elt_b8( mj[14], mj[ 1], mj[ 8], H[ 5],
+              (const __m512i)_mm512_set1_epi64( 30 * 0x0555555555555555ULL ) );
+   qt[31] = add_elt_b8( mj[15], mj[ 2], mj[ 9], H[ 6],
+              (const __m512i)_mm512_set1_epi64( 31 * 0x0555555555555555ULL ) );
+
+   qt[16] = _mm512_add_epi64( qt[16], expand1_b8( qt, 16 ) );
+   qt[17] = _mm512_add_epi64( qt[17], expand1_b8( qt, 17 ) );
+   qt[18] = _mm512_add_epi64( qt[18], expand2_b8( qt, 18 ) );
+   qt[19] = _mm512_add_epi64( qt[19], expand2_b8( qt, 19 ) );
+   qt[20] = _mm512_add_epi64( qt[20], expand2_b8( qt, 20 ) );
+   qt[21] = _mm512_add_epi64( qt[21], expand2_b8( qt, 21 ) );
+   qt[22] = _mm512_add_epi64( qt[22], expand2_b8( qt, 22 ) );
+   qt[23] = _mm512_add_epi64( qt[23], expand2_b8( qt, 23 ) );
+   qt[24] = _mm512_add_epi64( qt[24], expand2_b8( qt, 24 ) );
+   qt[25] = _mm512_add_epi64( qt[25], expand2_b8( qt, 25 ) );
+   qt[26] = _mm512_add_epi64( qt[26], expand2_b8( qt, 26 ) );
+   qt[27] = _mm512_add_epi64( qt[27], expand2_b8( qt, 27 ) );
+   qt[28] = _mm512_add_epi64( qt[28], expand2_b8( qt, 28 ) );
+   qt[29] = _mm512_add_epi64( qt[29], expand2_b8( qt, 29 ) );
+   qt[30] = _mm512_add_epi64( qt[30], expand2_b8( qt, 30 ) );
+   qt[31] = _mm512_add_epi64( qt[31], expand2_b8( qt, 31 ) );
+
+   xl = mm512_xor3( mm512_xor3( qt[16], qt[17], qt[18] ),
+                    mm512_xor3( qt[19], qt[20], qt[21] ),
+                    _mm512_xor_si512( qt[22], qt[23] ) );
+
+   xh = mm512_xor3( mm512_xor3( xl,     qt[24], qt[25] ),
+                    mm512_xor3( qt[26], qt[27], qt[28] ),
+                    mm512_xor3( qt[29], qt[30], qt[31] ) );

 #define DH1L( m, sl, sr, a, b, c ) \
-   _mm512_add_epi64( \
-               _mm512_xor_si512( M[m], \
-                  _mm512_xor_si512( _mm512_slli_epi64( xh, sl ), \
-                                    _mm512_srli_epi64( qt[a], sr ) ) ), \
-               _mm512_xor_si512( _mm512_xor_si512( xl, qt[b] ), qt[c] ) )
+   _mm512_add_epi64( mm512_xor3( M[m], _mm512_slli_epi64( xh, sl ), \
+                                       _mm512_srli_epi64( qt[a], sr ) ), \
+                     mm512_xor3( xl, qt[b], qt[c] ) )

 #define DH1R( m, sl, sr, a, b, c ) \
-   _mm512_add_epi64( \
-               _mm512_xor_si512( M[m], \
-                  _mm512_xor_si512( _mm512_srli_epi64( xh, sl ), \
-                                    _mm512_slli_epi64( qt[a], sr ) ) ), \
-               _mm512_xor_si512( _mm512_xor_si512( xl, qt[b] ), qt[c] ) )
+   _mm512_add_epi64( mm512_xor3( M[m], _mm512_srli_epi64( xh, sl ), \
+                                       _mm512_slli_epi64( qt[a], sr ) ), \
+                     mm512_xor3( xl, qt[b], qt[c] ) )

 #define DH2L( m, rl, sl, h, a, b, c ) \
   _mm512_add_epi64( _mm512_add_epi64( \
-       mm512_rol_64( dH[h], rl ), \
-          _mm512_xor_si512( _mm512_xor_si512( xh, qt[a] ), M[m] )), \
-                 _mm512_xor_si512( _mm512_slli_epi64( xl, sl ), \
-                                   _mm512_xor_si512( qt[b], qt[c] ) ) );
-   
+                        mm512_rol_64( dH[h], rl ), \
+                        mm512_xor3( xh, qt[a], M[m] ) ), \
+                     mm512_xor3( _mm512_slli_epi64( xl, sl ), qt[b], qt[c] ) ) 
+
 #define DH2R( m, rl, sr, h, a, b, c ) \
   _mm512_add_epi64( _mm512_add_epi64( \
-       mm512_rol_64( dH[h], rl ), \
-          _mm512_xor_si512( _mm512_xor_si512( xh, qt[a] ), M[m] )), \
-                 _mm512_xor_si512( _mm512_srli_epi64( xl, sr ), \
-                                   _mm512_xor_si512( qt[b], qt[c] ) ) );
+                        mm512_rol_64( dH[h], rl ), \
+                        mm512_xor3( xh, qt[a], M[m] ) ), \
+                     mm512_xor3( _mm512_srli_epi64( xl, sr ), qt[b], qt[c] ) )


   dH[ 0] = DH1L(  0,  5,  5, 16, 24, 0 );
--- a/algo/cubehash/cube-hash-2way.c
+++ b/algo/cubehash/cube-hash-2way.c
@@ -98,6 +98,138 @@ static void transform_4way( cube_4way_context *sp )
    _mm512_store_si512( (__m512i*)sp->h + 7, x7 );
 }

+// 8 ways, 4 way parallel double buffered
+static void transform_4way_2buf( cube_4way_2buf_context *sp )
+{
+    int r;
+    const int rounds = sp->rounds;
+
+    __m512i x0, x1, x2, x3, x4, x5, x6, x7;
+    __m512i y0, y1, y2, y3, y4, y5, y6, y7;
+    __m512i tx0, tx1, ty0, ty1;
+
+    x0 = _mm512_load_si512( (__m512i*)sp->h0     );
+    x1 = _mm512_load_si512( (__m512i*)sp->h0 + 1 );
+    x2 = _mm512_load_si512( (__m512i*)sp->h0 + 2 );
+    x3 = _mm512_load_si512( (__m512i*)sp->h0 + 3 );
+    x4 = _mm512_load_si512( (__m512i*)sp->h0 + 4 );
+    x5 = _mm512_load_si512( (__m512i*)sp->h0 + 5 );
+    x6 = _mm512_load_si512( (__m512i*)sp->h0 + 6 );
+    x7 = _mm512_load_si512( (__m512i*)sp->h0 + 7 );
+
+    y0 = _mm512_load_si512( (__m512i*)sp->h1     );
+    y1 = _mm512_load_si512( (__m512i*)sp->h1 + 1 );
+    y2 = _mm512_load_si512( (__m512i*)sp->h1 + 2 );
+    y3 = _mm512_load_si512( (__m512i*)sp->h1 + 3 );
+    y4 = _mm512_load_si512( (__m512i*)sp->h1 + 4 );
+    y5 = _mm512_load_si512( (__m512i*)sp->h1 + 5 );
+    y6 = _mm512_load_si512( (__m512i*)sp->h1 + 6 );
+    y7 = _mm512_load_si512( (__m512i*)sp->h1 + 7 );
+
+
+    for ( r = 0; r < rounds; ++r )
+    {
+        x4 = _mm512_add_epi32( x0, x4 );
+        y4 = _mm512_add_epi32( y0, y4 );
+        tx0 = x0;
+        ty0 = y0;
+        x5 = _mm512_add_epi32( x1, x5 );
+        y5 = _mm512_add_epi32( y1, y5 );
+        tx1 = x1;
+        ty1 = y1;
+        x0 = mm512_rol_32( x2, 7 );
+        y0 = mm512_rol_32( y2, 7 );
+        x6 = _mm512_add_epi32( x2, x6 );
+        y6 = _mm512_add_epi32( y2, y6 );
+        x1 = mm512_rol_32( x3, 7 );
+        y1 = mm512_rol_32( y3, 7 );
+        x7 = _mm512_add_epi32( x3, x7 );
+        y7 = _mm512_add_epi32( y3, y7 );
+
+
+        x2 = mm512_rol_32( tx0, 7 );
+        y2 = mm512_rol_32( ty0, 7 );
+        x0 = _mm512_xor_si512( x0, x4 );
+        y0 = _mm512_xor_si512( y0, y4 );
+        x4 = mm512_swap128_64( x4 );
+        x3 = mm512_rol_32( tx1, 7 );
+        y3 = mm512_rol_32( ty1, 7 );
+        y4 = mm512_swap128_64( y4 );
+
+        x1 = _mm512_xor_si512( x1, x5 );
+        y1 = _mm512_xor_si512( y1, y5 );
+        x5 = mm512_swap128_64( x5 );
+        x2 = _mm512_xor_si512( x2, x6 );
+        y2 = _mm512_xor_si512( y2, y6 );
+        y5 = mm512_swap128_64( y5 );
+        x3 = _mm512_xor_si512( x3, x7 );
+        y3 = _mm512_xor_si512( y3, y7 );
+
+        x6 = mm512_swap128_64( x6 );
+        x4 = _mm512_add_epi32( x0, x4 );
+        y4 = _mm512_add_epi32( y0, y4 );
+        y6 = mm512_swap128_64( y6 );
+        x5 = _mm512_add_epi32( x1, x5 );
+        y5 = _mm512_add_epi32( y1, y5 );
+        x7 = mm512_swap128_64( x7 );
+        x6 = _mm512_add_epi32( x2, x6 );
+        y6 = _mm512_add_epi32( y2, y6 );
+        tx0 = x0;
+        ty0 = y0;
+        y7 = mm512_swap128_64( y7 );
+        tx1 = x2;
+        ty1 = y2;
+        x0 = mm512_rol_32( x1, 11 );
+        y0 = mm512_rol_32( y1, 11 );
+
+        x7 = _mm512_add_epi32( x3, x7 );
+        y7 = _mm512_add_epi32( y3, y7 );
+
+        x1 = mm512_rol_32( tx0, 11 );
+        y1 = mm512_rol_32( ty0, 11 );
+        x0 = _mm512_xor_si512( x0, x4 );
+        x4 = mm512_swap64_32( x4 );
+        y0 = _mm512_xor_si512( y0, y4 );
+        x2 = mm512_rol_32( x3, 11 );
+        y4 = mm512_swap64_32( y4 );
+        y2 = mm512_rol_32( y3, 11 );
+        x1 = _mm512_xor_si512( x1, x5 );
+        x5 = mm512_swap64_32( x5 );
+        y1 = _mm512_xor_si512( y1, y5 );
+        x3 = mm512_rol_32( tx1, 11 );
+        y5 = mm512_swap64_32( y5 );
+        y3 = mm512_rol_32( ty1, 11 );
+
+        x2 = _mm512_xor_si512( x2, x6 );
+        x6 = mm512_swap64_32( x6 );
+        y2 = _mm512_xor_si512( y2, y6 );
+        y6 = mm512_swap64_32( y6 );
+        x3 = _mm512_xor_si512( x3, x7 );
+        x7 = mm512_swap64_32( x7 );
+        y3 = _mm512_xor_si512( y3, y7 );
+
+        y7 = mm512_swap64_32( y7 );
+    }
+
+    _mm512_store_si512( (__m512i*)sp->h0,     x0 );
+    _mm512_store_si512( (__m512i*)sp->h0 + 1, x1 );
+    _mm512_store_si512( (__m512i*)sp->h0 + 2, x2 );
+    _mm512_store_si512( (__m512i*)sp->h0 + 3, x3 );
+    _mm512_store_si512( (__m512i*)sp->h0 + 4, x4 );
+    _mm512_store_si512( (__m512i*)sp->h0 + 5, x5 );
+    _mm512_store_si512( (__m512i*)sp->h0 + 6, x6 );
+    _mm512_store_si512( (__m512i*)sp->h0 + 7, x7 );
+
+    _mm512_store_si512( (__m512i*)sp->h1,     y0 );
+    _mm512_store_si512( (__m512i*)sp->h1 + 1, y1 );
+    _mm512_store_si512( (__m512i*)sp->h1 + 2, y2 );
+    _mm512_store_si512( (__m512i*)sp->h1 + 3, y3 );
+    _mm512_store_si512( (__m512i*)sp->h1 + 4, y4 );
+    _mm512_store_si512( (__m512i*)sp->h1 + 5, y5 );
+    _mm512_store_si512( (__m512i*)sp->h1 + 6, y6 );
+    _mm512_store_si512( (__m512i*)sp->h1 + 7, y7 );
+}
+
 int cube_4way_init( cube_4way_context *sp, int hashbitlen, int rounds,
                    int blockbytes )
 {
@@ -219,6 +351,67 @@ int cube_4way_full( cube_4way_context *sp, void *output,  int hashbitlen,
    return 0;
 }

+int cube_4way_2buf_full( cube_4way_2buf_context *sp,
+                         void *output0, void *output1, int hashbitlen,
+                         const void *data0, const void *data1, size_t size )
+{
+    __m512i *h0 = (__m512i*)sp->h0;
+    __m512i *h1 = (__m512i*)sp->h1;
+    __m128i *iv = (__m128i*)( hashbitlen == 512 ? (__m128i*)IV512
+                                                : (__m128i*)IV256 );
+    sp->hashlen   = hashbitlen/128;
+    sp->blocksize = 32/16;
+    sp->rounds    = 16;
+    sp->pos       = 0;
+
+    h1[0] = h0[0] = m512_const1_128( iv[0] );
+    h1[1] = h0[1] = m512_const1_128( iv[1] );
+    h1[2] = h0[2] = m512_const1_128( iv[2] );
+    h1[3] = h0[3] = m512_const1_128( iv[3] );
+    h1[4] = h0[4] = m512_const1_128( iv[4] );
+    h1[5] = h0[5] = m512_const1_128( iv[5] );
+    h1[6] = h0[6] = m512_const1_128( iv[6] );
+    h1[7] = h0[7] = m512_const1_128( iv[7] );
+
+    const int len = size >> 4;
+    const __m512i *in0 = (__m512i*)data0;
+    const __m512i *in1 = (__m512i*)data1;
+    __m512i *hash0 = (__m512i*)output0;
+    __m512i *hash1 = (__m512i*)output1;
+    int i;
+
+    for ( i = 0; i < len; i++ )
+    {
+        sp->h0[ sp->pos ] = _mm512_xor_si512( sp->h0[ sp->pos ], in0[i] );
+        sp->h1[ sp->pos ] = _mm512_xor_si512( sp->h1[ sp->pos ], in1[i] );
+        sp->pos++;
+        if ( sp->pos == sp->blocksize )
+        {
+           transform_4way_2buf( sp );
+           sp->pos = 0;
+        }
+    }
+
+    // pos is zero for 64 byte data, 1 for 80 byte data.
+    __m512i tmp = m512_const2_64( 0, 0x0000000000000080 );
+    sp->h0[ sp->pos ] = _mm512_xor_si512( sp->h0[ sp->pos ], tmp );
+    sp->h1[ sp->pos ] = _mm512_xor_si512( sp->h1[ sp->pos ], tmp );
+
+    transform_4way_2buf( sp );
+
+    tmp = m512_const2_64( 0x0000000100000000, 0 );
+    sp->h0[7] = _mm512_xor_si512( sp->h0[7], tmp );
+    sp->h1[7] = _mm512_xor_si512( sp->h1[7], tmp );
+
+    for ( i = 0; i < 10; ++i )
+       transform_4way_2buf( sp );
+
+    memcpy( hash0, sp->h0, sp->hashlen<<6);
+    memcpy( hash1, sp->h1, sp->hashlen<<6);
+
+    return 0;
+}
+

 int cube_4way_update_close( cube_4way_context *sp, void *output,
                               const void *data, size_t size )
@@ -259,6 +452,21 @@ int cube_4way_update_close( cube_4way_context *sp, void *output,

 // 2 way 128 

+// This isn't expected to be used with AVX512 so HW rotate intruction
+// is assumed not avaiable.
+// Use double buffering to optimize serial bit rotations. Full double
+// buffering isn't practical because it needs twice as many registers
+// with AVX2 having only half as many as AVX512.
+#define ROL2( out0, out1, in0, in1, c ) \
+{ \
+ __m256i t0 = _mm256_slli_epi32( in0, c ); \
+ __m256i t1 = _mm256_slli_epi32( in1, c ); \
+ out0 = _mm256_srli_epi32( in0, 32-(c) ); \
+ out1 = _mm256_srli_epi32( in1, 32-(c) ); \
+ out0 = _mm256_or_si256( out0, t0 ); \
+ out1 = _mm256_or_si256( out1, t1 ); \
+}
+
 static void transform_2way( cube_2way_context *sp )
 {
    int r;
@@ -283,35 +491,31 @@ static void transform_2way( cube_2way_context *sp )
        x7 = _mm256_add_epi32( x3, x7 );
        y0 = x0;
        y1 = x1;
-        x0 = mm256_rol_32( x2, 7 );
-        x1 = mm256_rol_32( x3, 7 );
-        x2 = mm256_rol_32( y0, 7 );
-        x3 = mm256_rol_32( y1, 7 );
+        ROL2( x0, x1, x2, x3, 7 );
+        ROL2( x2, x3, y0, y1, 7 );
        x0 = _mm256_xor_si256( x0, x4 );
-        x1 = _mm256_xor_si256( x1, x5 );
-        x2 = _mm256_xor_si256( x2, x6 );
-        x3 = _mm256_xor_si256( x3, x7 );
        x4 = mm256_swap128_64( x4 );
-        x5 = mm256_swap128_64( x5 );
-        x6 = mm256_swap128_64( x6 );
-        x7 = mm256_swap128_64( x7 );
-        x4 = _mm256_add_epi32( x0, x4 );
-        x5 = _mm256_add_epi32( x1, x5 );
-        x6 = _mm256_add_epi32( x2, x6 );
-        x7 = _mm256_add_epi32( x3, x7 );
-        y0 = x0;
-        y1 = x2;
-        x0 = mm256_rol_32( x1, 11 );
-        x1 = mm256_rol_32( y0, 11 );
-        x2 = mm256_rol_32( x3, 11 );
-        x3 = mm256_rol_32( y1, 11 );
-        x0 = _mm256_xor_si256( x0, x4 );
        x1 = _mm256_xor_si256( x1, x5 );
        x2 = _mm256_xor_si256( x2, x6 );
+        x5 = mm256_swap128_64( x5 );
        x3 = _mm256_xor_si256( x3, x7 );
+        x4 = _mm256_add_epi32( x0, x4 );
+        x6 = mm256_swap128_64( x6 );
+        y0 = x0;
+        x5 = _mm256_add_epi32( x1, x5 );
+        x7 = mm256_swap128_64( x7 );
+        x6 = _mm256_add_epi32( x2, x6 );
+        y1 = x2;
+        ROL2( x0, x1, x1, y0, 11 );
+        x7 = _mm256_add_epi32( x3, x7 );
+        ROL2( x2, x3, x3, y1, 11 );
+        x0 = _mm256_xor_si256( x0, x4 );
        x4 = mm256_swap64_32( x4 );
+        x1 = _mm256_xor_si256( x1, x5 );
        x5 = mm256_swap64_32( x5 );
+        x2 = _mm256_xor_si256( x2, x6 );
        x6 = mm256_swap64_32( x6 );
+        x3 = _mm256_xor_si256( x3, x7 );
        x7 = mm256_swap64_32( x7 );
    }

--- a/algo/cubehash/cube-hash-2way.h
+++ b/algo/cubehash/cube-hash-2way.h
@@ -17,41 +17,41 @@ struct _cube_4way_context
    int pos; 
 } __attribute__ ((aligned (128)));

+struct _cube_4way_2buf_context
+{
+    __m512i h0[8];
+    __m512i h1[8];
+    int hashlen;
+    int rounds;
+    int blocksize;
+    int pos;
+} __attribute__ ((aligned (128)));
+
+
 typedef struct _cube_4way_context cube_4way_context;

+typedef struct _cube_4way_2buf_context cube_4way_2buf_context;
+
 int cube_4way_init( cube_4way_context* sp, int hashbitlen, int rounds,
-                       int blockbytes );
+                    int blockbytes );
+
 int cube_4way_update( cube_4way_context *sp, const void *data, size_t size );
+
 int cube_4way_close( cube_4way_context *sp, void *output );
+
 int cube_4way_update_close( cube_4way_context *sp, void *output,
                            const void *data, size_t size );
+
 int cube_4way_full( cube_4way_context *sp, void *output, int hashbitlen,
                    const void *data, size_t size );

-int cube_4x256_full( cube_4way_context *sp, void *output, int hashbitlen,
-                     const void *data, size_t size );
-
-#define cube512_4way_init( sp ) cube_4way_update( sp, 512 )
-#define cube512_4way_update cube_4way_update
-#define cube512_4way_update_close cube_4way_update
-#define cube512_4way_close cube_4way_update
-#define cube512_4way_full( sp, output, data, size ) \
-           cube_4way_full( sp, output, 512, data, size )
-#define cube512_4x256_full( sp, output, data, size ) \
-           cube_4x256_full( sp, output, 512, data, size )
-
-#define cube256_4way_init( sp ) cube_4way_update( sp, 256 )
-#define cube256_4way_update cube_4way_update
-#define cube256_4way_update_close cube_4way_update
-#define cube256_4way_close cube_4way_update
-#define cube256_4way_full( sp, output, data, size ) \
-           cube_4way_full( sp, output, 256, data, size )
-#define cube256_4x256_full( sp, output, data, size ) \
-           cube_4x256_full( sp, output, 256, data, size )
+int cube_4way_2buf_full( cube_4way_2buf_context *sp,
+                         void *output0, void *output1, int hashbitlen,
+                         const void *data0, const void *data1, size_t size );

 #endif

-// 2x128, 2 way parallel SSE2
+// 2x128, 2 way parallel AVX2

 struct _cube_2way_context
 {
--- a/algo/cubehash/cubehash_sse2.c
+++ b/algo/cubehash/cubehash_sse2.c
@@ -31,10 +31,14 @@ static void transform( cubehashParam *sp )
    for ( r = 0; r < rounds; ++r )
    { 
        x1 = _mm512_add_epi32( x0, x1 );
-        x0 = _mm512_xor_si512( mm512_rol_32( mm512_swap_256( x0 ), 7 ), x1 );
-        x1 = _mm512_add_epi32( x0, mm512_swap128_64( x1 ) );
-        x0 = _mm512_xor_si512( mm512_rol_32(
-                                         mm512_swap256_128( x0 ), 11 ), x1 );
+        x0 = mm512_swap_256( x0 );
+        x0 = mm512_rol_32(  x0, 7 );
+        x0 = _mm512_xor_si512( x0, x1 );
+        x1 = mm512_swap128_64( x1 );
+        x1 = _mm512_add_epi32( x0, x1 );
+        x0 = mm512_swap256_128( x0 );
+        x0 = mm512_rol_32( x0, 11 );
+        x0 = _mm512_xor_si512( x0, x1 );
        x1 = mm512_swap64_32( x1 );
    }

--- a/algo/echo/aes_ni/hash.c
+++ b/algo/echo/aes_ni/hash.c
@@ -53,10 +53,24 @@ MYALIGN const unsigned int	zero[]			= {0x00000000, 0x00000000, 0x00000000, 0x000
 MYALIGN const unsigned int	mul2ipt[]		= {0x728efc00, 0x6894e61a, 0x3fc3b14d, 0x25d9ab57, 0xfd5ba600, 0x2a8c71d7, 0x1eb845e3, 0xc96f9234};


+#define ECHO_SUBBYTES4(state, j) \
+   state[0][j] = _mm_aesenc_si128(state[0][j], k1);\
+   k1 = _mm_add_epi32(k1, M128(const1));\
+   state[1][j] = _mm_aesenc_si128(state[1][j], k1);\
+   k1 = _mm_add_epi32(k1, M128(const1));\
+   state[2][j] = _mm_aesenc_si128(state[2][j], k1);\
+   k1 = _mm_add_epi32(k1, M128(const1));\
+   state[3][j] = _mm_aesenc_si128(state[3][j], k1);\
+   k1 = _mm_add_epi32(k1, M128(const1));\
+   state[0][j] = _mm_aesenc_si128(state[0][j], m128_zero ); \
+   state[1][j] = _mm_aesenc_si128(state[1][j], m128_zero ); \
+   state[2][j] = _mm_aesenc_si128(state[2][j], m128_zero ); \
+   state[3][j] = _mm_aesenc_si128(state[3][j], m128_zero )
+
 #define ECHO_SUBBYTES(state, i, j) \
 	state[i][j] = _mm_aesenc_si128(state[i][j], k1);\
-	state[i][j] = _mm_aesenc_si128(state[i][j], M128(zero));\
-	k1 = _mm_add_epi32(k1, M128(const1))
+   k1 = _mm_add_epi32(k1, M128(const1));\
+	state[i][j] = _mm_aesenc_si128(state[i][j], M128(zero))

 #define ECHO_MIXBYTES(state1, state2, j, t1, t2, s2) \
 	s2 = _mm_add_epi8(state1[0][j], state1[0][j]);\
@@ -73,7 +87,7 @@ MYALIGN const unsigned int	mul2ipt[]		= {0x728efc00, 0x6894e61a, 0x3fc3b14d, 0x2
 	t1 = _mm_and_si128(t1, M128(lsbmask));\
 	t2 = _mm_shuffle_epi8(M128(mul2mask), t1);\
 	s2 = _mm_xor_si128(s2, t2);\
-	state2[0][j] = _mm_xor_si128(state2[0][j], _mm_xor_si128(s2, state1[1][(j + 1) & 3]));\
+	state2[0][j] = mm128_xor3(state2[0][j], s2, state1[1][(j + 1) & 3] );\
 	state2[1][j] = _mm_xor_si128(state2[1][j], s2);\
 	state2[2][j] = _mm_xor_si128(state2[2][j], state1[1][(j + 1) & 3]);\
 	state2[3][j] = _mm_xor_si128(state2[3][j], state1[1][(j + 1) & 3]);\
@@ -83,7 +97,7 @@ MYALIGN const unsigned int	mul2ipt[]		= {0x728efc00, 0x6894e61a, 0x3fc3b14d, 0x2
 	t2 = _mm_shuffle_epi8(M128(mul2mask), t1);\
 	s2 = _mm_xor_si128(s2, t2);\
 	state2[0][j] = _mm_xor_si128(state2[0][j], state1[2][(j + 2) & 3]);\
-	state2[1][j] = _mm_xor_si128(state2[1][j], _mm_xor_si128(s2, state1[2][(j + 2) & 3]));\
+	state2[1][j] = mm128_xor3(state2[1][j], s2, state1[2][(j + 2) & 3] );\
 	state2[2][j] = _mm_xor_si128(state2[2][j], s2);\
 	state2[3][j] = _mm_xor_si128(state2[3][j], state1[2][(j + 2) & 3]);\
 	s2 = _mm_add_epi8(state1[3][(j + 3) & 3], state1[3][(j + 3) & 3]);\
@@ -93,10 +107,29 @@ MYALIGN const unsigned int	mul2ipt[]		= {0x728efc00, 0x6894e61a, 0x3fc3b14d, 0x2
 	s2 = _mm_xor_si128(s2, t2);\
 	state2[0][j] = _mm_xor_si128(state2[0][j], state1[3][(j + 3) & 3]);\
 	state2[1][j] = _mm_xor_si128(state2[1][j], state1[3][(j + 3) & 3]);\
-	state2[2][j] = _mm_xor_si128(state2[2][j], _mm_xor_si128(s2, state1[3][(j + 3) & 3]));\
+	state2[2][j] = mm128_xor3(state2[2][j], s2, state1[3][(j + 3) & 3] );\
 	state2[3][j] = _mm_xor_si128(state2[3][j], s2)


+#define ECHO_ROUND_UNROLL2 \
+   ECHO_SUBBYTES4(_state, 0);\
+   ECHO_SUBBYTES4(_state, 1);\
+   ECHO_SUBBYTES4(_state, 2);\
+   ECHO_SUBBYTES4(_state, 3);\
+   ECHO_MIXBYTES(_state, _state2, 0, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 1, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 2, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 3, t1, t2, s2);\
+   ECHO_SUBBYTES4(_state2, 0);\
+   ECHO_SUBBYTES4(_state2, 1);\
+   ECHO_SUBBYTES4(_state2, 2);\
+   ECHO_SUBBYTES4(_state2, 3);\
+   ECHO_MIXBYTES(_state2, _state, 0, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 1, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 2, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 3, t1, t2, s2)
+
+/*
 #define ECHO_ROUND_UNROLL2 \
 	ECHO_SUBBYTES(_state, 0, 0);\
 	ECHO_SUBBYTES(_state, 1, 0);\
@@ -138,7 +171,7 @@ MYALIGN const unsigned int	mul2ipt[]		= {0x728efc00, 0x6894e61a, 0x3fc3b14d, 0x2
 	ECHO_MIXBYTES(_state2, _state, 1, t1, t2, s2);\
 	ECHO_MIXBYTES(_state2, _state, 2, t1, t2, s2);\
 	ECHO_MIXBYTES(_state2, _state, 3, t1, t2, s2)
-
+*/


 #define SAVESTATE(dst, src)\
--- a/algo/echo/echo-hash-4way.c
+++ b/algo/echo/echo-hash-4way.c
@@ -1,5 +1,4 @@
-//#if 0
-#if defined(__VAES__) && defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(__VAES__)

 #include "simd-utils.h"
 #include "echo-hash-4way.h"
@@ -11,18 +10,27 @@ static const unsigned int mul2ipt[] __attribute__ ((aligned (64))) =
   0xfd5ba600, 0x2a8c71d7, 0x1eb845e3, 0xc96f9234
 };
 */
-// do these need to be reversed?

-#define mul2mask \
-   _mm512_set4_epi32( 0, 0, 0, 0x00001b00 ) 
-//   _mm512_set4_epi32( 0x00001b00, 0, 0, 0 )  
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

-#define lsbmask    m512_const1_32( 0x01010101 ) 
+#define ECHO_SUBBYTES4(state, j) \
+   state[0][j] = _mm512_aesenc_epi128( state[0][j], k1 ); \
+   k1 = _mm512_add_epi32( k1, one ); \
+   state[1][j] = _mm512_aesenc_epi128( state[1][j], k1 ); \
+   k1 = _mm512_add_epi32( k1, one ); \
+   state[2][j] = _mm512_aesenc_epi128( state[2][j], k1 ); \
+   k1 = _mm512_add_epi32( k1, one ); \
+   state[3][j] = _mm512_aesenc_epi128( state[3][j], k1 ); \
+   k1 = _mm512_add_epi32( k1, one ); \
+   state[0][j] = _mm512_aesenc_epi128( state[0][j], m512_zero ); \
+   state[1][j] = _mm512_aesenc_epi128( state[1][j], m512_zero ); \
+   state[2][j] = _mm512_aesenc_epi128( state[2][j], m512_zero ); \
+   state[3][j] = _mm512_aesenc_epi128( state[3][j], m512_zero )

 #define ECHO_SUBBYTES( state, i, j ) \
 	state[i][j] = _mm512_aesenc_epi128( state[i][j], k1 ); \
-	state[i][j] = _mm512_aesenc_epi128( state[i][j], m512_zero ); \
-	k1 = _mm512_add_epi32( k1, m512_one_128 );
+   k1 = _mm512_add_epi32( k1, one ); \
+	state[i][j] = _mm512_aesenc_epi128( state[i][j], m512_zero );

 #define ECHO_MIXBYTES( state1, state2, j, t1, t2, s2 ) do \
 { \
@@ -30,87 +38,104 @@ static const unsigned int mul2ipt[] __attribute__ ((aligned (64))) =
   const int j2 = ( (j)+2 ) & 3; \
   const int j3 = ( (j)+3 ) & 3; \
   s2 = _mm512_add_epi8( state1[ 0 ] [j ], state1[ 0 ][ j ] ); \
-	t1 = _mm512_srli_epi16( state1[ 0 ][ j ], 7 ); \
-	t1 = _mm512_and_si512( t1, lsbmask );\
-	t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
-	s2 = _mm512_xor_si512( s2, t2 ); \
-	state2[ 0 ] [j ] = s2; \
-	state2[ 1 ] [j ] = state1[ 0 ][ j ]; \
-	state2[ 2 ] [j ] = state1[ 0 ][ j ]; \
-	state2[ 3 ] [j ] = _mm512_xor_si512( s2, state1[ 0 ][ j ] );\
-	s2 = _mm512_add_epi8( state1[ 1 ][ j1 ], state1[ 1 ][ j1 ] ); \
-	t1 = _mm512_srli_epi16( state1[ 1 ][ j1 ], 7 ); \
-	t1 = _mm512_and_si512( t1, lsbmask ); \
-	t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
-	s2 = _mm512_xor_si512( s2, t2 );\
-	state2[ 0 ][ j ] = _mm512_xor_si512( state2[ 0 ][ j ], \
-                            _mm512_xor_si512( s2, state1[ 1 ][ j1 ] ) ); \
-	state2[ 1 ][ j ] = _mm512_xor_si512( state2[ 1 ][ j ], s2 ); \
-	state2[ 2 ][ j ] = _mm512_xor_si512( state2[ 2 ][ j ], state1[ 1 ][ j1 ] ); \
-	state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3 ][ j ], state1[ 1 ][ j1 ] ); \
-	s2 = _mm512_add_epi8( state1[ 2 ][ j2 ], state1[ 2 ][ j2 ] ); \
-	t1 = _mm512_srli_epi16( state1[ 2 ][ j2 ], 7 ); \
-	t1 = _mm512_and_si512( t1, lsbmask ); \
-	t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
-	s2 = _mm512_xor_si512( s2, t2 ); \
-	state2[ 0 ][ j ] = _mm512_xor_si512( state2[ 0 ][ j ], state1[ 2 ][ j2 ] ); \
-	state2[ 1 ][ j ] = _mm512_xor_si512( state2[ 1 ][ j ], \
-                            _mm512_xor_si512( s2, state1[ 2 ][ j2 ] ) ); \
-	state2[ 2 ][ j ] = _mm512_xor_si512( state2[ 2 ][ j ], s2 ); \
-	state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3][ j ], state1[ 2 ][ j2 ] ); \
-	s2 = _mm512_add_epi8( state1[ 3 ][ j3 ], state1[ 3 ][ j3 ] ); \
-	t1 = _mm512_srli_epi16( state1[ 3 ][ j3 ], 7 ); \
-	t1 = _mm512_and_si512( t1, lsbmask ); \
-	t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
-	s2 = _mm512_xor_si512( s2, t2 ); \
-	state2[ 0 ][ j ] = _mm512_xor_si512( state2[ 0 ][ j ], state1[ 3 ][ j3 ] ); \
-	state2[ 1 ][ j ] = _mm512_xor_si512( state2[ 1 ][ j ], state1[ 3 ][ j3 ] ); \
-	state2[ 2 ][ j ] = _mm512_xor_si512( state2[ 2 ][ j ], \
-                            _mm512_xor_si512( s2, state1[ 3 ][ j3] ) ); \
-	state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3 ][ j ], s2 ); \
+   t1 = _mm512_srli_epi16( state1[ 0 ][ j ], 7 ); \
+   t1 = _mm512_and_si512( t1, lsbmask );\
+   t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
+   s2 = _mm512_xor_si512( s2, t2 ); \
+   state2[ 0 ] [j ] = s2; \
+   state2[ 1 ] [j ] = state1[ 0 ][ j ]; \
+   state2[ 2 ] [j ] = state1[ 0 ][ j ]; \
+   state2[ 3 ] [j ] = _mm512_xor_si512( s2, state1[ 0 ][ j ] );\
+   s2 = _mm512_add_epi8( state1[ 1 ][ j1 ], state1[ 1 ][ j1 ] ); \
+   t1 = _mm512_srli_epi16( state1[ 1 ][ j1 ], 7 ); \
+   t1 = _mm512_and_si512( t1, lsbmask ); \
+   t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
+   s2 = _mm512_xor_si512( s2, t2 );\
+   state2[ 0 ][ j ] = mm512_xor3( state2[ 0 ][ j ], s2, state1[ 1 ][ j1 ] ); \
+   state2[ 1 ][ j ] = _mm512_xor_si512( state2[ 1 ][ j ], s2 ); \
+   state2[ 2 ][ j ] = _mm512_xor_si512( state2[ 2 ][ j ], state1[ 1 ][ j1 ] ); \
+   state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3 ][ j ], state1[ 1 ][ j1 ] ); \
+   s2 = _mm512_add_epi8( state1[ 2 ][ j2 ], state1[ 2 ][ j2 ] ); \
+   t1 = _mm512_srli_epi16( state1[ 2 ][ j2 ], 7 ); \
+   t1 = _mm512_and_si512( t1, lsbmask ); \
+   t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
+   s2 = _mm512_xor_si512( s2, t2 ); \
+   state2[ 0 ][ j ] = _mm512_xor_si512( state2[ 0 ][ j ], state1[ 2 ][ j2 ] ); \
+   state2[ 1 ][ j ] = mm512_xor3( state2[ 1 ][ j ], s2, state1[ 2 ][ j2 ] ); \
+   state2[ 2 ][ j ] = _mm512_xor_si512( state2[ 2 ][ j ], s2 ); \
+   state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3][ j ], state1[ 2 ][ j2 ] ); \
+   s2 = _mm512_add_epi8( state1[ 3 ][ j3 ], state1[ 3 ][ j3 ] ); \
+   t1 = _mm512_srli_epi16( state1[ 3 ][ j3 ], 7 ); \
+   t1 = _mm512_and_si512( t1, lsbmask ); \
+   t2 = _mm512_shuffle_epi8( mul2mask, t1 ); \
+   s2 = _mm512_xor_si512( s2, t2 ); \
+   state2[ 0 ][ j ] = _mm512_xor_si512( state2[ 0 ][ j ], state1[ 3 ][ j3 ] ); \
+   state2[ 1 ][ j ] = _mm512_xor_si512( state2[ 1 ][ j ], state1[ 3 ][ j3 ] ); \
+   state2[ 2 ][ j ] = mm512_xor3( state2[ 2 ][ j ], s2, state1[ 3 ][ j3] ); \
+   state2[ 3 ][ j ] = _mm512_xor_si512( state2[ 3 ][ j ], s2 ); \
 } while(0)

 #define ECHO_ROUND_UNROLL2 \
-	ECHO_SUBBYTES(_state, 0, 0);\
+   ECHO_SUBBYTES4(_state, 0);\
+   ECHO_SUBBYTES4(_state, 1);\
+   ECHO_SUBBYTES4(_state, 2);\
+   ECHO_SUBBYTES4(_state, 3);\
+   ECHO_MIXBYTES(_state, _state2, 0, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 1, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 2, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 3, t1, t2, s2);\
+   ECHO_SUBBYTES4(_state2, 0);\
+   ECHO_SUBBYTES4(_state2, 1);\
+   ECHO_SUBBYTES4(_state2, 2);\
+   ECHO_SUBBYTES4(_state2, 3);\
+   ECHO_MIXBYTES(_state2, _state, 0, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 1, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 2, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 3, t1, t2, s2)
+
+/*
+#define ECHO_ROUND_UNROLL2 \
+   ECHO_SUBBYTES(_state, 0, 0);\
   ECHO_SUBBYTES(_state, 1, 0);\
-	ECHO_SUBBYTES(_state, 2, 0);\
-	ECHO_SUBBYTES(_state, 3, 0);\
-	ECHO_SUBBYTES(_state, 0, 1);\
-	ECHO_SUBBYTES(_state, 1, 1);\
-	ECHO_SUBBYTES(_state, 2, 1);\
-	ECHO_SUBBYTES(_state, 3, 1);\
-	ECHO_SUBBYTES(_state, 0, 2);\
-	ECHO_SUBBYTES(_state, 1, 2);\
-	ECHO_SUBBYTES(_state, 2, 2);\
-	ECHO_SUBBYTES(_state, 3, 2);\
-	ECHO_SUBBYTES(_state, 0, 3);\
-	ECHO_SUBBYTES(_state, 1, 3);\
-	ECHO_SUBBYTES(_state, 2, 3);\
-	ECHO_SUBBYTES(_state, 3, 3);\
-	ECHO_MIXBYTES(_state, _state2, 0, t1, t2, s2);\
-	ECHO_MIXBYTES(_state, _state2, 1, t1, t2, s2);\
-	ECHO_MIXBYTES(_state, _state2, 2, t1, t2, s2);\
-	ECHO_MIXBYTES(_state, _state2, 3, t1, t2, s2);\
-	ECHO_SUBBYTES(_state2, 0, 0);\
-	ECHO_SUBBYTES(_state2, 1, 0);\
-	ECHO_SUBBYTES(_state2, 2, 0);\
-	ECHO_SUBBYTES(_state2, 3, 0);\
-	ECHO_SUBBYTES(_state2, 0, 1);\
-	ECHO_SUBBYTES(_state2, 1, 1);\
-	ECHO_SUBBYTES(_state2, 2, 1);\
-	ECHO_SUBBYTES(_state2, 3, 1);\
-	ECHO_SUBBYTES(_state2, 0, 2);\
-	ECHO_SUBBYTES(_state2, 1, 2);\
-	ECHO_SUBBYTES(_state2, 2, 2);\
-	ECHO_SUBBYTES(_state2, 3, 2);\
-	ECHO_SUBBYTES(_state2, 0, 3);\
-	ECHO_SUBBYTES(_state2, 1, 3);\
-	ECHO_SUBBYTES(_state2, 2, 3);\
-	ECHO_SUBBYTES(_state2, 3, 3);\
-	ECHO_MIXBYTES(_state2, _state, 0, t1, t2, s2);\
-	ECHO_MIXBYTES(_state2, _state, 1, t1, t2, s2);\
-	ECHO_MIXBYTES(_state2, _state, 2, t1, t2, s2);\
-	ECHO_MIXBYTES(_state2, _state, 3, t1, t2, s2)
+   ECHO_SUBBYTES(_state, 2, 0);\
+   ECHO_SUBBYTES(_state, 3, 0);\
+   ECHO_SUBBYTES(_state, 0, 1);\
+   ECHO_SUBBYTES(_state, 1, 1);\
+   ECHO_SUBBYTES(_state, 2, 1);\
+   ECHO_SUBBYTES(_state, 3, 1);\
+   ECHO_SUBBYTES(_state, 0, 2);\
+   ECHO_SUBBYTES(_state, 1, 2);\
+   ECHO_SUBBYTES(_state, 2, 2);\
+   ECHO_SUBBYTES(_state, 3, 2);\
+   ECHO_SUBBYTES(_state, 0, 3);\
+   ECHO_SUBBYTES(_state, 1, 3);\
+   ECHO_SUBBYTES(_state, 2, 3);\
+   ECHO_SUBBYTES(_state, 3, 3);\
+   ECHO_MIXBYTES(_state, _state2, 0, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 1, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 2, t1, t2, s2);\
+   ECHO_MIXBYTES(_state, _state2, 3, t1, t2, s2);\
+   ECHO_SUBBYTES(_state2, 0, 0);\
+   ECHO_SUBBYTES(_state2, 1, 0);\
+   ECHO_SUBBYTES(_state2, 2, 0);\
+   ECHO_SUBBYTES(_state2, 3, 0);\
+   ECHO_SUBBYTES(_state2, 0, 1);\
+   ECHO_SUBBYTES(_state2, 1, 1);\
+   ECHO_SUBBYTES(_state2, 2, 1);\
+   ECHO_SUBBYTES(_state2, 3, 1);\
+   ECHO_SUBBYTES(_state2, 0, 2);\
+   ECHO_SUBBYTES(_state2, 1, 2);\
+   ECHO_SUBBYTES(_state2, 2, 2);\
+   ECHO_SUBBYTES(_state2, 3, 2);\
+   ECHO_SUBBYTES(_state2, 0, 3);\
+   ECHO_SUBBYTES(_state2, 1, 3);\
+   ECHO_SUBBYTES(_state2, 2, 3);\
+   ECHO_SUBBYTES(_state2, 3, 3);\
+   ECHO_MIXBYTES(_state2, _state, 0, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 1, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 2, t1, t2, s2);\
+   ECHO_MIXBYTES(_state2, _state, 3, t1, t2, s2)
+*/

 #define SAVESTATE(dst, src)\
 	dst[0][0] = src[0][0];\
@@ -137,6 +162,9 @@ void echo_4way_compress( echo_4way_context *ctx, const __m512i *pmsg,
  unsigned int r, b, i, j;
  __m512i t1, t2, s2, k1;
  __m512i _state[4][4], _state2[4][4], _statebackup[4][4]; 
+  __m512i one = m512_one_128;
+  __m512i mul2mask = m512_const2_64( 0, 0x00001b00 );
+  __m512i lsbmask  = m512_const1_32( 0x01010101 ); 

  _state[ 0 ][ 0 ] = ctx->state[ 0 ][ 0 ];
  _state[ 0 ][ 1 ] = ctx->state[ 0 ][ 1 ];
@@ -224,43 +252,43 @@ void echo_4way_compress( echo_4way_context *ctx, const __m512i *pmsg,

 int echo_4way_init( echo_4way_context *ctx, int nHashSize )
 {
-	int i, j;
+   int i, j;

   ctx->k = m512_zero; 
-	ctx->processed_bits = 0;
-	ctx->uBufferBytes = 0;
+   ctx->processed_bits = 0;
+   ctx->uBufferBytes = 0;

-	switch( nHashSize )
-	{
-		case 256:
-			ctx->uHashSize = 256;
-			ctx->uBlockLength = 192;
-			ctx->uRounds = 8;
-			ctx->hashsize = _mm512_set4_epi32( 0, 0, 0, 0x100 );
-			ctx->const1536 = _mm512_set4_epi32( 0, 0, 0, 0x600 );
-			break;
+   switch( nHashSize )
+   {
+	case 256:
+		ctx->uHashSize = 256;
+		ctx->uBlockLength = 192;
+		ctx->uRounds = 8;
+		ctx->hashsize = m512_const2_64( 0, 0x100 );
+		ctx->const1536 = m512_const2_64( 0, 0x600 );
+		break;

-		case 512:
-			ctx->uHashSize = 512;
-			ctx->uBlockLength = 128;
-			ctx->uRounds = 10;
-			ctx->hashsize = _mm512_set4_epi32( 0, 0, 0, 0x200 );
-			ctx->const1536 = _mm512_set4_epi32( 0, 0, 0, 0x400);
-			break;
+	case 512:
+		ctx->uHashSize = 512;
+		ctx->uBlockLength = 128;
+		ctx->uRounds = 10;
+		ctx->hashsize = m512_const2_64( 0, 0x200 );
+		ctx->const1536 = m512_const2_64( 0, 0x400);
+		break;

-		default:
-			return 1;
-	}
+	default:
+        	return 1;
+   }

-	for( i = 0; i < 4; i++ )
-		for( j = 0; j < nHashSize / 256; j++ )
-			ctx->state[ i ][ j ] = ctx->hashsize;
+   for( i = 0; i < 4; i++ )
+	for( j = 0; j < nHashSize / 256; j++ )
+		ctx->state[ i ][ j ] = ctx->hashsize;

-	for( i = 0; i < 4; i++ )
-		for( j = nHashSize / 256; j < 4; j++ )
-			ctx->state[ i ][ j ] = m512_zero;
+   for( i = 0; i < 4; i++ )
+	for( j = nHashSize / 256; j < 4; j++ )
+		ctx->state[ i ][ j ] = m512_zero;

-	return 0;
+   return 0;
 }

 int echo_4way_update_close( echo_4way_context *state, void *hashval,
@@ -285,17 +313,13 @@ int echo_4way_update_close( echo_4way_context *state, void *hashval,
      vlen = databitlen / 128;  // * 4 lanes / 128 bits per lane
      memcpy_512( state->buffer, data, vlen );
      state->processed_bits += (unsigned int)( databitlen );
-      remainingbits = _mm512_set4_epi32( 0, 0, 0, databitlen );
-
+      remainingbits = m512_const2_64( 0, (uint64_t)databitlen );
   }

-   state->buffer[ vlen ] = _mm512_set4_epi32( 0, 0, 0, 0x80 );
+   state->buffer[ vlen ] = m512_const2_64( 0, 0x80 );
   memset_zero_512( state->buffer + vlen + 1, vblen - vlen - 2 );
-   state->buffer[ vblen-2 ] =
-                _mm512_set4_epi32( (uint32_t)state->uHashSize << 16, 0, 0, 0 );
-   state->buffer[ vblen-1 ] =
-                   _mm512_set4_epi64( 0, state->processed_bits,
-                                      0, state->processed_bits );  
+   state->buffer[ vblen-2 ] = m512_const2_64( (uint64_t)state->uHashSize << 48, 0 );
+   state->buffer[ vblen-1 ] = m512_const2_64( 0, state->processed_bits);

   state->k = _mm512_add_epi64( state->k, remainingbits );
   state->k = _mm512_sub_epi64( state->k, state->const1536 );
@@ -328,16 +352,16 @@ int echo_4way_full( echo_4way_context *ctx, void *hashval, int nHashSize,
         ctx->uHashSize = 256;
         ctx->uBlockLength = 192;
         ctx->uRounds = 8;
-         ctx->hashsize = _mm512_set4_epi32( 0, 0, 0, 0x100 );
-         ctx->const1536 = _mm512_set4_epi32( 0, 0, 0, 0x600 );
+         ctx->hashsize = m512_const2_64( 0, 0x100 );
+         ctx->const1536 = m512_const2_64( 0, 0x600 );
         break;

      case 512:
         ctx->uHashSize = 512;
         ctx->uBlockLength = 128;
         ctx->uRounds = 10;
-         ctx->hashsize = _mm512_set4_epi32( 0, 0, 0, 0x200 );
-         ctx->const1536 = _mm512_set4_epi32( 0, 0, 0, 0x400);
+         ctx->hashsize = m512_const2_64( 0, 0x200 );
+         ctx->const1536 = m512_const2_64( 0, 0x400 );
         break;

      default:
@@ -372,17 +396,14 @@ int echo_4way_full( echo_4way_context *ctx, void *hashval, int nHashSize,
      vlen = databitlen / 128;  // * 4 lanes / 128 bits per lane
      memcpy_512( ctx->buffer, data, vlen );
      ctx->processed_bits += (unsigned int)( databitlen );
-      remainingbits = _mm512_set4_epi32( 0, 0, 0, databitlen );
-
+      remainingbits = m512_const2_64( 0, databitlen );
   }

-   ctx->buffer[ vlen ] = _mm512_set4_epi32( 0, 0, 0, 0x80 );
+   ctx->buffer[ vlen ] = m512_const2_64( 0, 0x80 );
   memset_zero_512( ctx->buffer + vlen + 1, vblen - vlen - 2 );
   ctx->buffer[ vblen-2 ] =
-                _mm512_set4_epi32( (uint32_t)ctx->uHashSize << 16, 0, 0, 0 );
-   ctx->buffer[ vblen-1 ] =
-                   _mm512_set4_epi64( 0, ctx->processed_bits,
-                                      0, ctx->processed_bits );
+                     m512_const2_64( (uint64_t)ctx->uHashSize << 48, 0 );
+   ctx->buffer[ vblen-1 ] = m512_const2_64( 0, ctx->processed_bits);

   ctx->k = _mm512_add_epi64( ctx->k, remainingbits );
   ctx->k = _mm512_sub_epi64( ctx->k, ctx->const1536 );
@@ -400,5 +421,414 @@ int echo_4way_full( echo_4way_context *ctx, void *hashval, int nHashSize,
   return 0;
 }

+#endif  // AVX512

-#endif
+// AVX2 + VAES
+
+#define mul2mask_2way   m256_const2_64( 0, 0x0000000000001b00 ) 
+
+#define lsbmask_2way    m256_const1_32( 0x01010101 ) 
+
+#define ECHO_SUBBYTES4_2WAY( state, j ) \
+   state[0][j] = _mm256_aesenc_epi128( state[0][j], k1 ); \
+   k1 = _mm256_add_epi32( k1, m256_one_128 ); \
+   state[1][j] = _mm256_aesenc_epi128( state[1][j], k1 ); \
+   k1 = _mm256_add_epi32( k1, m256_one_128 ); \
+   state[2][j] = _mm256_aesenc_epi128( state[2][j], k1 ); \
+   k1 = _mm256_add_epi32( k1, m256_one_128 ); \
+   state[3][j] = _mm256_aesenc_epi128( state[3][j], k1 ); \
+   k1 = _mm256_add_epi32( k1, m256_one_128 ); \
+   state[0][j] = _mm256_aesenc_epi128( state[0][j], m256_zero ); \
+   state[1][j] = _mm256_aesenc_epi128( state[1][j], m256_zero ); \
+   state[2][j] = _mm256_aesenc_epi128( state[2][j], m256_zero ); \
+   state[3][j] = _mm256_aesenc_epi128( state[3][j], m256_zero )
+
+#define ECHO_SUBBYTES_2WAY( state, i, j ) \
+        state[i][j] = _mm256_aesenc_epi128( state[i][j], k1 ); \
+        k1 = _mm256_add_epi32( k1, m256_one_128 ); \
+        state[i][j] = _mm256_aesenc_epi128( state[i][j], m256_zero ); \
+
+#define ECHO_MIXBYTES_2WAY( state1, state2, j, t1, t2, s2 ) do \
+{ \
+   const int j1 = ( (j)+1 ) & 3; \
+   const int j2 = ( (j)+2 ) & 3; \
+   const int j3 = ( (j)+3 ) & 3; \
+   s2 = _mm256_add_epi8( state1[ 0 ] [j ], state1[ 0 ][ j ] ); \
+   t1 = _mm256_srli_epi16( state1[ 0 ][ j ], 7 ); \
+   t1 = _mm256_and_si256( t1, lsbmask_2way );\
+   t2 = _mm256_shuffle_epi8( mul2mask_2way, t1 ); \
+   s2 = _mm256_xor_si256( s2, t2 ); \
+   state2[ 0 ] [j ] = s2; \
+   state2[ 1 ] [j ] = state1[ 0 ][ j ]; \
+   state2[ 2 ] [j ] = state1[ 0 ][ j ]; \
+   state2[ 3 ] [j ] = _mm256_xor_si256( s2, state1[ 0 ][ j ] );\
+   s2 = _mm256_add_epi8( state1[ 1 ][ j1 ], state1[ 1 ][ j1 ] ); \
+   t1 = _mm256_srli_epi16( state1[ 1 ][ j1 ], 7 ); \
+   t1 = _mm256_and_si256( t1, lsbmask_2way ); \
+   t2 = _mm256_shuffle_epi8( mul2mask_2way, t1 ); \
+   s2 = _mm256_xor_si256( s2, t2 );\
+   state2[ 0 ][ j ] = _mm256_xor_si256( state2[ 0 ][ j ], \
+                              _mm256_xor_si256( s2, state1[ 1 ][ j1 ] ) ); \
+   state2[ 1 ][ j ] = _mm256_xor_si256( state2[ 1 ][ j ], s2 ); \
+   state2[ 2 ][ j ] = _mm256_xor_si256( state2[ 2 ][ j ], state1[ 1 ][ j1 ] ); \
+   state2[ 3 ][ j ] = _mm256_xor_si256( state2[ 3 ][ j ], state1[ 1 ][ j1 ] ); \
+   s2 = _mm256_add_epi8( state1[ 2 ][ j2 ], state1[ 2 ][ j2 ] ); \
+   t1 = _mm256_srli_epi16( state1[ 2 ][ j2 ], 7 ); \
+   t1 = _mm256_and_si256( t1, lsbmask_2way ); \
+   t2 = _mm256_shuffle_epi8( mul2mask_2way, t1 ); \
+   s2 = _mm256_xor_si256( s2, t2 ); \
+   state2[ 0 ][ j ] = _mm256_xor_si256( state2[ 0 ][ j ], state1[ 2 ][ j2 ] ); \
+   state2[ 1 ][ j ] = _mm256_xor_si256( state2[ 1 ][ j ], \
+                            _mm256_xor_si256( s2, state1[ 2 ][ j2 ] ) ); \
+   state2[ 2 ][ j ] = _mm256_xor_si256( state2[ 2 ][ j ], s2 ); \
+   state2[ 3 ][ j ] = _mm256_xor_si256( state2[ 3][ j ], state1[ 2 ][ j2 ] ); \
+   s2 = _mm256_add_epi8( state1[ 3 ][ j3 ], state1[ 3 ][ j3 ] ); \
+   t1 = _mm256_srli_epi16( state1[ 3 ][ j3 ], 7 ); \
+   t1 = _mm256_and_si256( t1, lsbmask_2way ); \
+   t2 = _mm256_shuffle_epi8( mul2mask_2way, t1 ); \
+   s2 = _mm256_xor_si256( s2, t2 ); \
+   state2[ 0 ][ j ] = _mm256_xor_si256( state2[ 0 ][ j ], state1[ 3 ][ j3 ] ); \
+   state2[ 1 ][ j ] = _mm256_xor_si256( state2[ 1 ][ j ], state1[ 3 ][ j3 ] ); \
+   state2[ 2 ][ j ] = _mm256_xor_si256( state2[ 2 ][ j ], \
+                            _mm256_xor_si256( s2, state1[ 3 ][ j3] ) ); \
+   state2[ 3 ][ j ] = _mm256_xor_si256( state2[ 3 ][ j ], s2 ); \
+} while(0)
+
+#define ECHO_ROUND_UNROLL2_2WAY \
+   ECHO_SUBBYTES4_2WAY(_state, 0);\
+   ECHO_SUBBYTES4_2WAY(_state, 1);\
+   ECHO_SUBBYTES4_2WAY(_state, 2);\
+   ECHO_SUBBYTES4_2WAY(_state, 3);\
+   ECHO_MIXBYTES_2WAY(_state, _state2, 0, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state, _state2, 1, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state, _state2, 2, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state, _state2, 3, t1, t2, s2);\
+   ECHO_SUBBYTES4_2WAY(_state2, 0);\
+   ECHO_SUBBYTES4_2WAY(_state2, 1);\
+   ECHO_SUBBYTES4_2WAY(_state2, 2);\
+   ECHO_SUBBYTES4_2WAY(_state2, 3);\
+   ECHO_MIXBYTES_2WAY(_state2, _state, 0, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state2, _state, 1, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state2, _state, 2, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state2, _state, 3, t1, t2, s2)
+
+/*
+#define ECHO_ROUND_UNROLL2_2WAY \
+   ECHO_SUBBYTES_2WAY(_state, 0, 0);\
+   ECHO_SUBBYTES_2WAY(_state, 1, 0);\
+   ECHO_SUBBYTES_2WAY(_state, 2, 0);\
+   ECHO_SUBBYTES_2WAY(_state, 3, 0);\
+   ECHO_SUBBYTES_2WAY(_state, 0, 1);\
+   ECHO_SUBBYTES_2WAY(_state, 1, 1);\
+   ECHO_SUBBYTES_2WAY(_state, 2, 1);\
+   ECHO_SUBBYTES_2WAY(_state, 3, 1);\
+   ECHO_SUBBYTES_2WAY(_state, 0, 2);\
+   ECHO_SUBBYTES_2WAY(_state, 1, 2);\
+   ECHO_SUBBYTES_2WAY(_state, 2, 2);\
+   ECHO_SUBBYTES_2WAY(_state, 3, 2);\
+   ECHO_SUBBYTES_2WAY(_state, 0, 3);\
+   ECHO_SUBBYTES_2WAY(_state, 1, 3);\
+   ECHO_SUBBYTES_2WAY(_state, 2, 3);\
+   ECHO_SUBBYTES_2WAY(_state, 3, 3);\
+   ECHO_MIXBYTES_2WAY(_state, _state2, 0, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state, _state2, 1, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state, _state2, 2, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state, _state2, 3, t1, t2, s2);\
+   ECHO_SUBBYTES_2WAY(_state2, 0, 0);\
+   ECHO_SUBBYTES_2WAY(_state2, 1, 0);\
+   ECHO_SUBBYTES_2WAY(_state2, 2, 0);\
+   ECHO_SUBBYTES_2WAY(_state2, 3, 0);\
+   ECHO_SUBBYTES_2WAY(_state2, 0, 1);\
+   ECHO_SUBBYTES_2WAY(_state2, 1, 1);\
+   ECHO_SUBBYTES_2WAY(_state2, 2, 1);\
+   ECHO_SUBBYTES_2WAY(_state2, 3, 1);\
+   ECHO_SUBBYTES_2WAY(_state2, 0, 2);\
+   ECHO_SUBBYTES_2WAY(_state2, 1, 2);\
+   ECHO_SUBBYTES_2WAY(_state2, 2, 2);\
+   ECHO_SUBBYTES_2WAY(_state2, 3, 2);\
+   ECHO_SUBBYTES_2WAY(_state2, 0, 3);\
+   ECHO_SUBBYTES_2WAY(_state2, 1, 3);\
+   ECHO_SUBBYTES_2WAY(_state2, 2, 3);\
+   ECHO_SUBBYTES_2WAY(_state2, 3, 3);\
+   ECHO_MIXBYTES_2WAY(_state2, _state, 0, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state2, _state, 1, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state2, _state, 2, t1, t2, s2);\
+   ECHO_MIXBYTES_2WAY(_state2, _state, 3, t1, t2, s2)
+*/
+
+#define SAVESTATE_2WAY(dst, src)\
+        dst[0][0] = src[0][0];\
+        dst[0][1] = src[0][1];\
+        dst[0][2] = src[0][2];\
+        dst[0][3] = src[0][3];\
+        dst[1][0] = src[1][0];\
+        dst[1][1] = src[1][1];\
+        dst[1][2] = src[1][2];\
+        dst[1][3] = src[1][3];\
+        dst[2][0] = src[2][0];\
+        dst[2][1] = src[2][1];\
+        dst[2][2] = src[2][2];\
+        dst[2][3] = src[2][3];\
+        dst[3][0] = src[3][0];\
+        dst[3][1] = src[3][1];\
+        dst[3][2] = src[3][2];\
+        dst[3][3] = src[3][3]
+
+// blockcount always 1
+void echo_2way_compress( echo_2way_context *ctx, const __m256i *pmsg,
+               unsigned int uBlockCount )
+{
+  unsigned int r, b, i, j;
+  __m256i t1, t2, s2, k1;
+  __m256i _state[4][4], _state2[4][4], _statebackup[4][4];
+
+  _state[ 0 ][ 0 ] = ctx->state[ 0 ][ 0 ];
+  _state[ 0 ][ 1 ] = ctx->state[ 0 ][ 1 ];
+  _state[ 0 ][ 2 ] = ctx->state[ 0 ][ 2 ];
+  _state[ 0 ][ 3 ] = ctx->state[ 0 ][ 3 ];
+  _state[ 1 ][ 0 ] = ctx->state[ 1 ][ 0 ];
+  _state[ 1 ][ 1 ] = ctx->state[ 1 ][ 1 ];
+  _state[ 1 ][ 2 ] = ctx->state[ 1 ][ 2 ];
+  _state[ 1 ][ 3 ] = ctx->state[ 1 ][ 3 ];
+  _state[ 2 ][ 0 ] = ctx->state[ 2 ][ 0 ];
+  _state[ 2 ][ 1 ] = ctx->state[ 2 ][ 1 ];
+  _state[ 2 ][ 2 ] = ctx->state[ 2 ][ 2 ];
+  _state[ 2 ][ 3 ] = ctx->state[ 2 ][ 3 ];
+  _state[ 3 ][ 0 ] = ctx->state[ 3 ][ 0 ];
+  _state[ 3 ][ 1 ] = ctx->state[ 3 ][ 1 ];
+  _state[ 3 ][ 2 ] = ctx->state[ 3 ][ 2 ];
+  _state[ 3 ][ 3 ] = ctx->state[ 3 ][ 3 ];
+
+  for ( b = 0; b < uBlockCount; b++ )
+  {
+    ctx->k = _mm256_add_epi64( ctx->k, ctx->const1536 );
+
+    for( j = ctx->uHashSize / 256; j < 4; j++ )
+    {
+      for ( i = 0; i < 4; i++ )
+      {
+        _state[ i ][ j ] = _mm256_load_si256(
+                     pmsg + 4 * (j - (ctx->uHashSize / 256)) + i );
+      }
+   }
+
+   // save state
+   SAVESTATE_2WAY( _statebackup, _state );
+
+   k1 = ctx->k;
+
+   for ( r = 0; r < ctx->uRounds / 2; r++ )
+   {
+       ECHO_ROUND_UNROLL2_2WAY;
+   }
+
+   if ( ctx->uHashSize == 256 )
+   {
+      for ( i = 0; i < 4; i++ )
+      {
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _state[ i ][ 1 ] );
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _state[ i ][ 2 ] );
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _state[ i ][ 3 ] );
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 0 ] );
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 1 ] );
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 2 ] ) ;
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 3 ] );
+       }
+    }
+    else
+    {
+       for ( i = 0; i < 4; i++ )
+       {
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _state[ i ][ 2 ] );
+         _state[ i ][ 1 ] = _mm256_xor_si256( _state[ i ][ 1 ],
+                                              _state[ i ][ 3 ] );
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ][ 0 ],
+                                              _statebackup[ i ][ 0 ] );
+         _state[ i ][ 0 ] = _mm256_xor_si256( _state[ i ] [0 ],
+                                              _statebackup[ i ][ 2 ] );
+         _state[ i ][ 1 ] = _mm256_xor_si256( _state[ i ][ 1 ],
+                                              _statebackup[ i ][ 1 ] );
+         _state[ i ][ 1 ] = _mm256_xor_si256( _state[ i ][ 1 ],
+                                              _statebackup[ i ][ 3 ] );
+      }
+    }
+    pmsg += ctx->uBlockLength;
+  }
+  SAVESTATE_2WAY(ctx->state, _state);
+
+}
+int echo_2way_init( echo_2way_context *ctx, int nHashSize )
+{
+        int i, j;
+
+   ctx->k = m256_zero;
+   ctx->processed_bits = 0;
+   ctx->uBufferBytes = 0;
+
+   switch( nHashSize )
+   {
+                case 256:
+                        ctx->uHashSize = 256;
+                        ctx->uBlockLength = 192;
+                        ctx->uRounds = 8;
+                        ctx->hashsize = m256_const2_64( 0, 0x100 );
+                        ctx->const1536 = m256_const2_64( 0, 0x600 );
+                        break;
+
+                case 512:
+                        ctx->uHashSize = 512;
+                        ctx->uBlockLength = 128;
+                        ctx->uRounds = 10;
+                        ctx->hashsize = m256_const2_64( 0, 0x200 );
+                        ctx->const1536 = m256_const2_64( 0, 0x400 );
+                        break;
+
+                default:
+                        return 1;
+        }
+
+        for( i = 0; i < 4; i++ )
+                for( j = 0; j < nHashSize / 256; j++ )
+                        ctx->state[ i ][ j ] = ctx->hashsize;
+
+        for( i = 0; i < 4; i++ )
+                for( j = nHashSize / 256; j < 4; j++ )
+                        ctx->state[ i ][ j ] = m256_zero;
+
+        return 0;
+}
+
+int echo_2way_update_close( echo_2way_context *state, void *hashval,
+                              const void *data, int databitlen )
+{
+// bytelen is either 32 (maybe), 64 or 80 or 128!
+// all are less than full block.
+
+   int vlen = databitlen / 128;  // * 4 lanes / 128 bits per lane
+   const int vblen = state->uBlockLength / 16; //  16 bytes per lane
+   __m256i remainingbits;
+
+   if ( databitlen == 1024 )
+   {
+      echo_2way_compress( state, data, 1 );
+      state->processed_bits = 1024;
+      remainingbits = m256_const2_64( 0, -1024 );
+      vlen = 0;
+   }
+   else
+   {
+      memcpy_256( state->buffer, data, vlen );
+      state->processed_bits += (unsigned int)( databitlen );
+      remainingbits = m256_const2_64( 0, databitlen );
+   }
+
+   state->buffer[ vlen ] = m256_const2_64( 0, 0x80 );
+   memset_zero_256( state->buffer + vlen + 1, vblen - vlen - 2 );
+   state->buffer[ vblen-2 ] = m256_const2_64( (uint64_t)state->uHashSize << 48, 0 );
+   state->buffer[ vblen-1 ] = m256_const2_64( 0, state->processed_bits );
+
+   state->k = _mm256_add_epi64( state->k, remainingbits );
+   state->k = _mm256_sub_epi64( state->k, state->const1536 );
+
+   echo_2way_compress( state, state->buffer, 1 );
+
+   _mm256_store_si256( (__m256i*)hashval + 0, state->state[ 0 ][ 0] );
+   _mm256_store_si256( (__m256i*)hashval + 1, state->state[ 1 ][ 0] );
+
+   if ( state->uHashSize == 512 )
+   {
+      _mm256_store_si256( (__m256i*)hashval + 2, state->state[ 2 ][ 0 ] );
+      _mm256_store_si256( (__m256i*)hashval + 3, state->state[ 3 ][ 0 ] );
+   }
+   return 0;
+}
+
+int echo_2way_full( echo_2way_context *ctx, void *hashval, int nHashSize,
+                    const void *data, int datalen )
+{
+   int i, j;
+   int databitlen = datalen * 8;
+   ctx->k = m256_zero;
+   ctx->processed_bits = 0;
+   ctx->uBufferBytes = 0;
+
+   switch( nHashSize )
+   {
+      case 256:
+         ctx->uHashSize = 256;
+         ctx->uBlockLength = 192;
+         ctx->uRounds = 8;
+         ctx->hashsize = m256_const2_64( 0, 0x100 );
+         ctx->const1536 = m256_const2_64( 0, 0x600 );
+         break;
+
+      case 512:
+         ctx->uHashSize = 512;
+         ctx->uBlockLength = 128;
+         ctx->uRounds = 10;
+         ctx->hashsize = m256_const2_64( 0, 0x200 );
+         ctx->const1536 = m256_const2_64( 0, 0x400 );
+         break;
+
+      default:
+         return 1;
+   }
+
+   for( i = 0; i < 4; i++ )
+      for( j = 0; j < nHashSize / 256; j++ )
+         ctx->state[ i ][ j ] = ctx->hashsize;
+
+   for( i = 0; i < 4; i++ )
+      for( j = nHashSize / 256; j < 4; j++ )
+         ctx->state[ i ][ j ] = m256_zero;
+
+   int vlen = datalen / 32;
+   const int vblen = ctx->uBlockLength / 16; //  16 bytes per lane
+   __m256i remainingbits;
+
+   if ( databitlen == 1024 )
+   {
+      echo_2way_compress( ctx, data, 1 );
+      ctx->processed_bits = 1024;
+      remainingbits = m256_const2_64( 0, -1024 );
+      vlen = 0;
+   }
+   else
+   {
+      vlen = databitlen / 128;  // * 4 lanes / 128 bits per lane
+      memcpy_256( ctx->buffer, data, vlen );
+      ctx->processed_bits += (unsigned int)( databitlen );
+      remainingbits = m256_const2_64( 0, databitlen );
+   }
+
+   ctx->buffer[ vlen ] = m256_const2_64( 0, 0x80 );
+   memset_zero_256( ctx->buffer + vlen + 1, vblen - vlen - 2 );
+   ctx->buffer[ vblen-2 ] = m256_const2_64( (uint64_t)ctx->uHashSize << 48, 0 );
+   ctx->buffer[ vblen-1 ] = m256_const2_64( 0, ctx->processed_bits );
+
+   ctx->k = _mm256_add_epi64( ctx->k, remainingbits );
+   ctx->k = _mm256_sub_epi64( ctx->k, ctx->const1536 );
+
+   echo_2way_compress( ctx, ctx->buffer, 1 );
+
+   _mm256_store_si256( (__m256i*)hashval + 0, ctx->state[ 0 ][ 0] );
+   _mm256_store_si256( (__m256i*)hashval + 1, ctx->state[ 1 ][ 0] );
+
+   if ( ctx->uHashSize == 512 )
+   {
+      _mm256_store_si256( (__m256i*)hashval + 2, ctx->state[ 2 ][ 0 ] );
+      _mm256_store_si256( (__m256i*)hashval + 3, ctx->state[ 3 ][ 0 ] );
+   }
+   return 0;
+}
+
+
+#endif   // VAES
--- a/algo/echo/echo-hash-4way.h
+++ b/algo/echo/echo-hash-4way.h
@@ -1,10 +1,12 @@
 #if !defined(ECHO_HASH_4WAY_H__)
 #define ECHO_HASH_4WAY_H__ 1

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

 #include "simd-utils.h"

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 typedef struct
 {
   __m512i    state[4][4];
@@ -20,6 +22,7 @@ typedef struct
   unsigned int   processed_bits;

 } echo_4way_context __attribute__ ((aligned (64)));
+#define echo512_4way_context echo_4way_context

 int echo_4way_init( echo_4way_context *state, int hashbitlen );
 #define echo512_4way_init( state ) echo_4way_init( state, 512 )
@@ -29,8 +32,8 @@ int echo_4way_update( echo_4way_context *state, const void *data,
    unsigned int databitlen);
 #define echo512_4way_update echo_4way_update

-int echo_close( echo_4way_context *state, void *hashval );
-#define echo512_4way_close echo_4way_close
+// int echo_4way_close( echo_4way_context *state, void *hashval );
+// #define echo512_4way_close echo_4way_close

 int echo_4way_update_close( echo_4way_context *state, void *hashval,
                              const void *data, int databitlen );
@@ -43,5 +46,45 @@ int echo_4way_full( echo_4way_context *ctx, void *hashval, int nHashSize,
 #define echo256_4way_full( state, hashval, data, datalen ) \
           echo_4way_full( state, hashval, 256, data, datalen )

-#endif 
-#endif
+#endif   // AVX512
+
+typedef struct
+{
+   __m256i    state[4][4];
+   __m256i    buffer[ 4 * 192 / 16 ];  // 4x128 interleaved 192 bytes
+   __m256i    k;
+   __m256i    hashsize;
+   __m256i    const1536;
+
+   unsigned int   uRounds;
+   unsigned int   uHashSize;
+   unsigned int   uBlockLength;
+   unsigned int   uBufferBytes;
+   unsigned int   processed_bits;
+
+} echo_2way_context __attribute__ ((aligned (64)));
+#define echo512_2way_context echo_2way_context
+
+int echo_2way_init( echo_2way_context *state, int hashbitlen );
+#define echo512_2way_init( state ) echo_2way_init( state, 512 )
+#define echo256_2way_init( state ) echo_2way_init( state, 256 )
+
+int echo_2way_update( echo_2way_context *state, const void *data,
+    unsigned int databitlen);
+#define echo512_2way_update echo_2way_update
+
+int echo_2way_update_close( echo_2way_context *state, void *hashval,
+                              const void *data, int databitlen );
+#define echo512_2way_update_close echo_2way_update_close
+
+int echo_2way_full( echo_2way_context *ctx, void *hashval, int nHashSize,
+                    const void *data, int datalen );
+#define echo512_2way_full( state, hashval, data, datalen ) \
+           echo_2way_full( state, hashval, 512, data, datalen )
+#define echo256_2way_full( state, hashval, data, datalen ) \
+           echo_2way_full( state, hashval, 256, data, datalen )
+
+
+#endif  // VAES
+
+#endif   // ECHO_HASH_4WAY_H__
--- a/algo/fugue/fugue-aesni.c
+++ b/algo/fugue/fugue-aesni.c
@@ -34,13 +34,11 @@ MYALIGN const unsigned long long _supermix4c[]	= {0x0706050403020000, 0x03020000
 MYALIGN const unsigned long long _supermix7a[]	= {0x010c0b060d080702, 0x0904030e03000104};
 MYALIGN const unsigned long long _supermix7b[]	= {0x8080808080808080, 0x0504070605040f06};
 MYALIGN const unsigned long long _k_n[] = {0x4E4E4E4E4E4E4E4E, 0x1B1B1B1B0E0E0E0E};
-MYALIGN const unsigned int _maskd3n[] = {0xffffffff, 0xffffffff, 0xffffffff, 0x00000000};
 MYALIGN const unsigned char _shift_one_mask[]   = {7, 4, 5, 6, 11, 8, 9, 10, 15, 12, 13, 14, 3, 0, 1, 2};
 MYALIGN const unsigned char _shift_four_mask[]  = {13, 14, 15, 12, 1, 2, 3, 0, 5, 6, 7, 4, 9, 10, 11, 8};
 MYALIGN const unsigned char _shift_seven_mask[] = {10, 11, 8, 9, 14, 15, 12, 13, 2, 3, 0, 1, 6, 7, 4, 5};
 MYALIGN const unsigned char _aes_shift_rows[]   = {0, 5, 10, 15, 4, 9, 14, 3, 8, 13, 2, 7, 12, 1, 6, 11};
 MYALIGN const unsigned int _inv_shift_rows[] = {0x070a0d00, 0x0b0e0104, 0x0f020508, 0x0306090c};
-MYALIGN const unsigned int _zero[] = {0x00000000, 0x00000000, 0x00000000, 0x00000000};
 MYALIGN const unsigned int _mul2mask[] = {0x1b1b0000, 0x00000000, 0x00000000, 0x00000000};
 MYALIGN const unsigned int _mul4mask[] = {0x2d361b00, 0x00000000, 0x00000000, 0x00000000};
 MYALIGN const unsigned int _lsbmask2[] = {0x03030303, 0x03030303, 0x03030303, 0x03030303};
@@ -61,7 +59,7 @@ MYALIGN const unsigned int _IV512[] = {

 #define UNPACK_S0(s0, s1, t1)\
   s1 = _mm_castps_si128(_mm_insert_ps(_mm_castsi128_ps(s1), _mm_castsi128_ps(s0), 0xc0));\
-   s0 = _mm_and_si128(s0, M128(_maskd3n))
+   s0 = mm128_mask_32( s0, 8 )

 #define CMIX(s1, s2, r1, r2, t1, t2)\
   t1 = s1;\
@@ -78,7 +76,7 @@ MYALIGN const unsigned int _IV512[] = {
 #define UNPACK_S0(s0, s1, t1)\
   t1 = _mm_shuffle_epi32(s0, _MM_SHUFFLE(3, 3, 3, 3));\
   s1 = _mm_castps_si128(_mm_move_ss(_mm_castsi128_ps(s1), _mm_castsi128_ps(t1)));\
-   s0 = _mm_and_si128(s0, M128(_maskd3n))
+   s0 = mm128_mask_32( s0, 8 )

 #define CMIX(s1, s2, r1, r2, t1, t2)\
   t1 = _mm_shuffle_epi32(s1, 0xf9);\
@@ -126,7 +124,16 @@ MYALIGN const unsigned int _IV512[] = {
 	t1 = _mm_shuffle_epi32(s30, _MM_SHUFFLE(3, 3, 0, 3));\
 	s7 = _mm_xor_si128(s7, t1)

+#define PRESUPERMIX(t0, t1, t2, t3, t4)\
+   t2 = t0;\
+   t3 = _mm_add_epi8(t0, t0);\
+   t4 = _mm_add_epi8(t3, t3);\
+   t1 = _mm_srli_epi16(t0, 6);\
+   t1 = _mm_and_si128(t1, M128(_lsbmask2));\
+   t3 = _mm_xor_si128(t3, _mm_shuffle_epi8(M128(_mul2mask), t1));\
+   t0  = _mm_xor_si128(t4, _mm_shuffle_epi8(M128(_mul4mask), t1))

+/*
 #define PRESUPERMIX(x, t1, s1, s2, t2)\
 	s1 = x;\
 	s2 = _mm_add_epi8(x, x);\
@@ -135,37 +142,59 @@ MYALIGN const unsigned int _IV512[] = {
 	t1 = _mm_and_si128(t1, M128(_lsbmask2));\
 	s2 = _mm_xor_si128(s2, _mm_shuffle_epi8(M128(_mul2mask), t1));\
 	x  = _mm_xor_si128(t2, _mm_shuffle_epi8(M128(_mul4mask), t1))
+*/

-#define SUBSTITUTE(r0, _t1, _t2, _t3, _t0)\
+#define SUBSTITUTE(r0, _t2 )\
 	_t2 = _mm_shuffle_epi8(r0, M128(_inv_shift_rows));\
-	_t2 = _mm_aesenclast_si128(_t2, M128(_zero))
-	
+	_t2 = _mm_aesenclast_si128( _t2, m128_zero )
+
+#define SUPERMIX(t0, t1, t2, t3, t4)\
+   t2 = t0;\
+   t3 = _mm_add_epi8(t0, t0);\
+   t4 = _mm_add_epi8(t3, t3);\
+   t1 = _mm_srli_epi16(t0, 6);\
+   t1 = _mm_and_si128(t1, M128(_lsbmask2));\
+   t0 = _mm_xor_si128(t4, _mm_shuffle_epi8(M128(_mul4mask), t1)); \
+   t4 = _mm_shuffle_epi8(t2, M128(_supermix1b));\
+   t3 = _mm_xor_si128(t3, _mm_shuffle_epi8(M128(_mul2mask), t1));\
+   t1 = _mm_shuffle_epi8(t4, M128(_supermix1c));\
+   t4 = _mm_xor_si128(t4, t1);\
+   t1 = _mm_shuffle_epi8(t4, M128(_supermix1d));\
+   t4 = _mm_xor_si128(t4, t1);\
+   t1 = _mm_shuffle_epi8(t2, M128(_supermix1a));\
+   t2 = mm128_xor3(t2, t3, t0 );\
+   t2 = _mm_shuffle_epi8(t2, M128(_supermix7a));\
+   t4 = mm128_xor3( t4, t1, t2 ); \
+   t2 = _mm_shuffle_epi8(t2, M128(_supermix7b));\
+   t3 = _mm_shuffle_epi8(t3, M128(_supermix2a));\
+   t1 = _mm_shuffle_epi8(t0, M128(_supermix4a));\
+   t0 = _mm_shuffle_epi8(t0, M128(_supermix4b));\
+   t4 = mm128_xor3( t4, t2, t1 ); \
+   t0 = _mm_xor_si128(t0, t3);\
+   t4 = mm128_xor3(t4, t0, _mm_shuffle_epi8(t0, M128(_supermix4c)));
+
+/*
 #define SUPERMIX(t0, t1, t2, t3, t4)\
 	PRESUPERMIX(t0, t1, t2, t3, t4);\
 	POSTSUPERMIX(t0, t1, t2, t3, t4)
-
+*/

 #define POSTSUPERMIX(t0, t1, t2, t3, t4)\
-	t1 = t2;\
-	t1 = _mm_shuffle_epi8(t1, M128(_supermix1b));\
+	t1 = _mm_shuffle_epi8(t2, M128(_supermix1b));\
 	t4 = t1;\
 	t1 = _mm_shuffle_epi8(t1, M128(_supermix1c));\
 	t4 = _mm_xor_si128(t4, t1);\
-	t1 = t4;\
-	t1 = _mm_shuffle_epi8(t1, M128(_supermix1d));\
+	t1 = _mm_shuffle_epi8(t4, M128(_supermix1d));\
 	t4 = _mm_xor_si128(t4, t1);\
-	t1 = t2;\
-	t1 = _mm_shuffle_epi8(t1, M128(_supermix1a));\
+	t1 = _mm_shuffle_epi8(t2, M128(_supermix1a));\
 	t4 = _mm_xor_si128(t4, t1);\
-	t2 = _mm_xor_si128(t2, t3);\
-	t2 = _mm_xor_si128(t2, t0);\
+	t2 = mm128_xor3(t2, t3, t0 );\
 	t2 = _mm_shuffle_epi8(t2, M128(_supermix7a));\
 	t4 = _mm_xor_si128(t4, t2);\
 	t2 = _mm_shuffle_epi8(t2, M128(_supermix7b));\
 	t4 = _mm_xor_si128(t4, t2);\
 	t3 = _mm_shuffle_epi8(t3, M128(_supermix2a));\
-	t1 = t0;\
-	t1 = _mm_shuffle_epi8(t1, M128(_supermix4a));\
+	t1 = _mm_shuffle_epi8(t0, M128(_supermix4a));\
 	t4 = _mm_xor_si128(t4, t1);\
 	t0 = _mm_shuffle_epi8(t0, M128(_supermix4b));\
 	t0 = _mm_xor_si128(t0, t3);\
@@ -173,59 +202,55 @@ MYALIGN const unsigned int _IV512[] = {
 	t0 = _mm_shuffle_epi8(t0, M128(_supermix4c));\
 	t4 = _mm_xor_si128(t4, t0)

-
 #define SUBROUND512_3(r1a, r1b, r1c, r1d, r2a, r2b, r2c, r2d, r3a, r3b, r3c, r3d)\
 	CMIX(r1a, r1b, r1c, r1d, _t0, _t1);\
 	PACK_S0(r1c, r1a, _t0);\
-	SUBSTITUTE(r1c, _t1, _t2, _t3, _t0);\
+	SUBSTITUTE(r1c, _t2 );\
 	SUPERMIX(_t2, _t3, _t0, _t1, r1c);\
 	_t0 = _mm_shuffle_epi32(r1c, 0x39);\
 	r2c = _mm_xor_si128(r2c, _t0);\
-	_t0 = _mm_and_si128(_t0, M128(_maskd3n));\
+   _t0 = mm128_mask_32( _t0, 8 ); \
 	r2d = _mm_xor_si128(r2d, _t0);\
 	UNPACK_S0(r1c, r1a, _t3);\
-	SUBSTITUTE(r2c, _t1, _t2, _t3, _t0);\
+	SUBSTITUTE(r2c, _t2 );\
 	SUPERMIX(_t2, _t3, _t0, _t1, r2c);\
 	_t0 = _mm_shuffle_epi32(r2c, 0x39);\
 	r3c = _mm_xor_si128(r3c, _t0);\
-	_t0 = _mm_and_si128(_t0, M128(_maskd3n));\
+   _t0 = mm128_mask_32( _t0, 8 ); \
 	r3d = _mm_xor_si128(r3d, _t0);\
 	UNPACK_S0(r2c, r2a, _t3);\
-	SUBSTITUTE(r3c, _t1, _t2, _t3, _t0);\
+	SUBSTITUTE(r3c, _t2 );\
 	SUPERMIX(_t2, _t3, _t0, _t1, r3c);\
 	UNPACK_S0(r3c, r3a, _t3)

-
 #define SUBROUND512_4(r1a, r1b, r1c, r1d, r2a, r2b, r2c, r2d, r3a, r3b, r3c, r3d, r4a, r4b, r4c, r4d)\
 	CMIX(r1a, r1b, r1c, r1d, _t0, _t1);\
 	PACK_S0(r1c, r1a, _t0);\
-	SUBSTITUTE(r1c, _t1, _t2, _t3, _t0);\
+	SUBSTITUTE( r1c, _t2 );\
 	SUPERMIX(_t2, _t3, _t0, _t1, r1c);\
 	_t0 = _mm_shuffle_epi32(r1c, 0x39);\
 	r2c = _mm_xor_si128(r2c, _t0);\
-	_t0 = _mm_and_si128(_t0, M128(_maskd3n));\
+   _t0 = mm128_mask_32( _t0, 8 ); \
 	r2d = _mm_xor_si128(r2d, _t0);\
 	UNPACK_S0(r1c, r1a, _t3);\
-	SUBSTITUTE(r2c, _t1, _t2, _t3, _t0);\
+	SUBSTITUTE(r2c, _t2 );\
 	SUPERMIX(_t2, _t3, _t0, _t1, r2c);\
 	_t0 = _mm_shuffle_epi32(r2c, 0x39);\
 	r3c = _mm_xor_si128(r3c, _t0);\
-	_t0 = _mm_and_si128(_t0, M128(_maskd3n));\
+   _t0 = mm128_mask_32( _t0, 8 ); \
 	r3d = _mm_xor_si128(r3d, _t0);\
 	UNPACK_S0(r2c, r2a, _t3);\
-	SUBSTITUTE(r3c, _t1, _t2, _t3, _t0);\
+	SUBSTITUTE( r3c, _t2 );\
 	SUPERMIX(_t2, _t3, _t0, _t1, r3c);\
 	_t0 = _mm_shuffle_epi32(r3c, 0x39);\
 	r4c = _mm_xor_si128(r4c, _t0);\
-	_t0 = _mm_and_si128(_t0, M128(_maskd3n));\
+   _t0 = mm128_mask_32( _t0, 8 ); \
 	r4d = _mm_xor_si128(r4d, _t0);\
 	UNPACK_S0(r3c, r3a, _t3);\
-	SUBSTITUTE(r4c, _t1, _t2, _t3, _t0);\
+	SUBSTITUTE( r4c, _t2 );\
 	SUPERMIX(_t2, _t3, _t0, _t1, r4c);\
 	UNPACK_S0(r4c, r4a, _t3)

-
-
 #define LOADCOLUMN(x, s, a)\
 	block[0] = col[(base + a + 0) % s];\
 	block[1] = col[(base + a + 1) % s];\
@@ -249,14 +274,14 @@ void Compress512(hashState_fugue *ctx, const unsigned char *pmsg, unsigned int u
      case 1:
         TIX512( pmsg, ctx->state[3], ctx->state[10], ctx->state[4],
                       ctx->state[5], ctx->state[ 6], ctx->state[8],
-		       ctx->state[9], ctx->state[10], _t0, _t1, _t2 );
+                       ctx->state[9], ctx->state[10], _t0, _t1, _t2 );

-	 SUBROUND512_4( ctx->state[8], ctx->state[9], ctx->state[7],
+	      SUBROUND512_4( ctx->state[8], ctx->state[9], ctx->state[7],
                        ctx->state[1], ctx->state[7], ctx->state[8],
-		       	ctx->state[6], ctx->state[0], ctx->state[6],
-		       	ctx->state[7], ctx->state[5], ctx->state[11],
-		       	ctx->state[5], ctx->state[6], ctx->state[4],
-		       	ctx->state[10] );
+		                  ctx->state[6], ctx->state[0], ctx->state[6],
+		                  ctx->state[7], ctx->state[5], ctx->state[11],
+		                  ctx->state[5], ctx->state[6], ctx->state[4],
+		       	         ctx->state[10] );
         ctx->base++;
         pmsg += 4;
         uBlockCount--;
@@ -265,14 +290,14 @@ void Compress512(hashState_fugue *ctx, const unsigned char *pmsg, unsigned int u
      case 2:
         TIX512( pmsg, ctx->state[11], ctx->state[6], ctx->state[0],
                       ctx->state[ 1], ctx->state[2], ctx->state[4],
-		       ctx->state[ 5], ctx->state[6], _t0, _t1, _t2);
+                       ctx->state[ 5], ctx->state[6], _t0, _t1, _t2);

         SUBROUND512_4( ctx->state[4], ctx->state[5], ctx->state[3],
                        ctx->state[9], ctx->state[3], ctx->state[4],
-		       	ctx->state[2], ctx->state[8], ctx->state[2],
-		       	ctx->state[3], ctx->state[1], ctx->state[7],
-		       	ctx->state[1], ctx->state[2], ctx->state[0],
-		       	ctx->state[6]);
+                        ctx->state[2], ctx->state[8], ctx->state[2],
+                        ctx->state[3], ctx->state[1], ctx->state[7],
+                        ctx->state[1], ctx->state[2], ctx->state[0],
+                        ctx->state[6]);

         ctx->base = 0;
         pmsg += 4;
@@ -280,44 +305,42 @@ void Compress512(hashState_fugue *ctx, const unsigned char *pmsg, unsigned int u
      break;
   }

-
   while( uBlockCount > 0 )
   {
-      TIX512( pmsg, ctx->state[ 7], ctx->state[2], ctx->state[8], ctx->state[9],
-                    ctx->state[10], ctx->state[0], ctx->state[1], ctx->state[2],
-              _t0, _t1, _t2 );
-      SUBROUND512_4( ctx->state[0], ctx->state[1], ctx->state[11],
-                     ctx->state[5], ctx->state[11], ctx->state[0],
-		     ctx->state[10], ctx->state[4], ctx->state[10],
-		     ctx->state[11], ctx->state[9], ctx->state[3],
-		     ctx->state[9], ctx->state[10], ctx->state[8],
-		     ctx->state[2] );
+      TIX512( pmsg, ctx->state[ 7],ctx->state[2],ctx->state[8],ctx->state[9],
+                    ctx->state[10],ctx->state[0],ctx->state[1],ctx->state[2],
+                    _t0, _t1, _t2 );
+      SUBROUND512_4( ctx->state[0], ctx->state[1],ctx->state[11],ctx->state[5],
+                     ctx->state[11],ctx->state[0],ctx->state[10],ctx->state[4],
+                     ctx->state[10],ctx->state[11],ctx->state[9],ctx->state[3],
+		               ctx->state[9],ctx->state[10],ctx->state[8],ctx->state[2] );

      ctx->base++;
      pmsg += 4;
      uBlockCount--;
      if( uBlockCount == 0 ) break;

-      TIX512( pmsg, ctx->state[3], ctx->state[10], ctx->state[4], ctx->state[5],
-                    ctx->state[6], ctx->state[8], ctx->state[9], ctx->state[10],
-              _t0, _t1, _t2 );
+      TIX512( pmsg, ctx->state[3],ctx->state[10],ctx->state[4],ctx->state[5],
+                    ctx->state[6],ctx->state[8], ctx->state[9],ctx->state[10],
+                    _t0, _t1, _t2 );

-      SUBROUND512_4( ctx->state[8], ctx->state[9], ctx->state[7], ctx->state[1],                     ctx->state[7], ctx->state[8], ctx->state[6], ctx->state[0],
-		     ctx->state[6], ctx->state[7], ctx->state[5], ctx->state[11],
-		     ctx->state[5], ctx->state[6, ctx->state[4], ctx->state[10]);
+      SUBROUND512_4( ctx->state[8],ctx->state[9],ctx->state[7],ctx->state[1],
+                     ctx->state[7],ctx->state[8],ctx->state[6],ctx->state[0],
+		               ctx->state[6],ctx->state[7],ctx->state[5],ctx->state[11],
+		               ctx->state[5],ctx->state[6],ctx->state[4],ctx->state[10] );

      ctx->base++;
      pmsg += 4;
      uBlockCount--;
      if( uBlockCount == 0 ) break;

-      TIX512( pmsg, ctx->state[11], ctx->state[6], ctx->state[0], ctx->state[1],
-		    ctx->state[2], ctx->state[4], ctx->state[5], ctx->state[6],
-               _t0, _t1, _t2);
-      SUBROUND512_4( ctx->state[4], ctx->state[5], ctx->state[3], ctx->state[9],
-		     ctx->state[3], ctx->state[4], ctx->state[2], ctx->state[8],
-		     ctx->state[2], ctx->state[3], ctx->state[1], ctx->state[7],
-		     ctx->state[1], ctx->state[2], ctx->state[0], ctx->state[6]);
+      TIX512( pmsg, ctx->state[11],ctx->state[6],ctx->state[0],ctx->state[1],
+                    ctx->state[2], ctx->state[4],ctx->state[5],ctx->state[6],
+                    _t0, _t1, _t2);
+      SUBROUND512_4( ctx->state[4],ctx->state[5],ctx->state[3],ctx->state[9],
+                     ctx->state[3],ctx->state[4],ctx->state[2],ctx->state[8],
+                     ctx->state[2],ctx->state[3],ctx->state[1],ctx->state[7],
+		               ctx->state[1],ctx->state[2],ctx->state[0],ctx->state[6]);

      ctx->base = 0;
      pmsg += 4;
@@ -328,8 +351,8 @@ void Compress512(hashState_fugue *ctx, const unsigned char *pmsg, unsigned int u

 void Final512(hashState_fugue *ctx, BitSequence *hashval)
 {
-        unsigned int block[4] __attribute__ ((aligned (32)));
-        unsigned int col[36] __attribute__ ((aligned (16)));
+   unsigned int block[4] __attribute__ ((aligned (32)));
+   unsigned int col[36] __attribute__ ((aligned (16)));
 	unsigned int i, base;
 	__m128i r0, _t0, _t1, _t2, _t3;

@@ -359,7 +382,7 @@ void Final512(hashState_fugue *ctx, BitSequence *hashval)

 		// SMIX
 		LOADCOLUMN(r0, 36, 0);
-		SUBSTITUTE(r0, _t1, _t2, _t3, _t0);
+		SUBSTITUTE(r0, _t2);
 		SUPERMIX(_t2, _t3, _t0, _t1, r0);
 		STORECOLUMN(r0, 36);
 	}
@@ -377,7 +400,7 @@ void Final512(hashState_fugue *ctx, BitSequence *hashval)

 		// SMIX
 		LOADCOLUMN(r0, 36, 0);
-		SUBSTITUTE(r0, _t1, _t2, _t3, _t0);
+		SUBSTITUTE(r0, _t2);
 		SUPERMIX(_t2, _t3, _t0, _t1, r0);
 		STORECOLUMN(r0, 36);

@@ -392,7 +415,7 @@ void Final512(hashState_fugue *ctx, BitSequence *hashval)

 		// SMIX
 		LOADCOLUMN(r0, 36, 0);
-		SUBSTITUTE(r0, _t1, _t2, _t3, _t0);
+		SUBSTITUTE(r0, _t2);
 		SUPERMIX(_t2, _t3, _t0, _t1, r0);
 		STORECOLUMN(r0, 36);

@@ -407,7 +430,7 @@ void Final512(hashState_fugue *ctx, BitSequence *hashval)

 		// SMIX
 		LOADCOLUMN(r0, 36, 0);
-		SUBSTITUTE(r0, _t1, _t2, _t3, _t0);
+		SUBSTITUTE(r0, _t2);
 		SUPERMIX(_t2, _t3, _t0, _t1, r0);
 		STORECOLUMN(r0, 36);

@@ -422,7 +445,7 @@ void Final512(hashState_fugue *ctx, BitSequence *hashval)

 		// SMIX
 		LOADCOLUMN(r0, 36, 0);
-		SUBSTITUTE(r0, _t1, _t2, _t3, _t0);
+		SUBSTITUTE(r0, _t2);
 		SUPERMIX(_t2, _t3, _t0, _t1, r0);
 		STORECOLUMN(r0, 36);
 	}
@@ -462,7 +485,7 @@ HashReturn fugue512_Init(hashState_fugue *ctx, int nHashSize)
 	ctx->uBlockLength = 4;

 	for(i = 0; i < 6; i++)
-		ctx->state[i] = _mm_setzero_si128();
+		ctx->state[i] = m128_zero;

 	ctx->state[6]  = _mm_load_si128((__m128i*)_IV512 + 0);
 	ctx->state[7]  = _mm_load_si128((__m128i*)_IV512 + 1);
--- a/algo/fugue/fugue-aesni.h
+++ b/algo/fugue/fugue-aesni.h
@@ -14,10 +14,14 @@
 #ifndef FUGUE_HASH_API_H
 #define FUGUE_HASH_API_H

-#if defined(__AES__)
+#if defined(__AES__) 
+
+#if !defined(__SSE4_1__)
+#error "Unsupported configuration, AES needs SSE4.1. Compile without AES."
+#endif

 #include "algo/sha/sha3_common.h"
-#include <x86intrin.h>
+#include "simd-utils.h"


 typedef struct
--- a/algo/groestl/aes_ni/groestl-intr-aes.h
+++ b/algo/groestl/aes_ni/groestl-intr-aes.h
@@ -67,11 +67,9 @@ static const __m128i SUBSH_MASK7 = { 0x06090c0f0205080b, 0x0e0104070a0d0003 };
 * xmm[j] will be lost
 * xmm[k] has to be all 0x1b */
 #define MUL2(i, j, k){\
-  j = _mm_xor_si128(j, j);\
-  j = _mm_cmpgt_epi8(j, i);\
+  j = _mm_cmpgt_epi8( m128_zero, i);\
  i = _mm_add_epi8(i, i);\
-  j = _mm_and_si128(j, k);\
-  i = _mm_xor_si128(i, j);\
+  i = mm128_xorand(i, j, k );\
 } 

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

+#endif
+

 /* one round
 * a0-a7 = input rows
--- a/algo/groestl/aes_ni/groestl256-intr-aes.h
+++ b/algo/groestl/aes_ni/groestl256-intr-aes.h
@@ -58,11 +58,9 @@ static const __m128i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e };
 * xmm[j] will be lost
 * xmm[k] has to be all 0x1b */
 #define MUL2(i, j, k){\
-  j = _mm_xor_si128(j, j);\
-  j = _mm_cmpgt_epi8(j, i);\
+  j = _mm_cmpgt_epi8( m128_zero, i);\
  i = _mm_add_epi8(i, i);\
-  j = _mm_and_si128(j, k);\
-  i = _mm_xor_si128(i, j);\
+  i = mm128_xorand(i, j, k );\
 } 

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

+#endif
+
 /* one round
 * i = round number
 * a0-a7 = input rows
--- a/algo/groestl/aes_ni/hash-groestl.h
+++ b/algo/groestl/aes_ni/hash-groestl.h
@@ -43,7 +43,8 @@
 #define ROUNDS (ROUNDS1024)
 //#endif

-#define ROTL64(a,n) ((((a)<<(n))|((a)>>(64-(n))))&li_64(ffffffffffffffff))
+//#define ROTL64(a,n) ((((a)<<(n))|((a)>>(64-(n))))&li_64(ffffffffffffffff))
+#define ROTL64(a,n) rol64( a, n )

 #if (PLATFORM_BYTE_ORDER == IS_BIG_ENDIAN)
 #define EXT_BYTE(var,n) ((u8)((u64)(var) >> (8*(7-(n)))))
--- a/algo/groestl/aes_ni/hash-groestl256.h
+++ b/algo/groestl/aes_ni/hash-groestl256.h
@@ -63,7 +63,8 @@ typedef crypto_uint64 u64;
 //#define ROUNDS (ROUNDS1024)
 //#endif

-#define ROTL64(a,n) ((((a)<<(n))|((a)>>(64-(n))))&li_64(ffffffffffffffff))
+//#define ROTL64(a,n) ((((a)<<(n))|((a)>>(64-(n))))&li_64(ffffffffffffffff))
+#define ROTL64(a,n) rol64( a, n )

 #if (PLATFORM_BYTE_ORDER == IS_BIG_ENDIAN)
 #define EXT_BYTE(var,n) ((u8)((u64)(var) >> (8*(7-(n)))))
--- a/algo/groestl/groestl256-hash-4way.c
+++ b/algo/groestl/groestl256-hash-4way.c
@@ -15,7 +15,9 @@
 #include "miner.h"
 #include "simd-utils.h"

-#if defined(__VAES__) && defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(__AVX2__) && defined(__VAES__)
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)


 int groestl256_4way_init( groestl256_4way_context* ctx, uint64_t hashlen )
@@ -43,13 +45,13 @@ int groestl256_4way_init( groestl256_4way_context* ctx, uint64_t hashlen )
 }

 int groestl256_4way_full( groestl256_4way_context* ctx, void* output,
-                                const void* input, uint64_t databitlen )
+                                const void* input, uint64_t datalen )
 {
-   const int len = (int)databitlen / 128;
-   const int hashlen_m128i = 32 / 16;   // bytes to __m128i
+   const int len = (int)datalen >> 4;
+   const int hashlen_m128i = 32 >> 4;   // bytes to __m128i
   const int hash_offset = SIZE256 - hashlen_m128i;
   int rem = ctx->rem_ptr;
-   int blocks = len / SIZE256;
+   uint64_t blocks = len / SIZE256;
   __m512i* in = (__m512i*)input;
   int i;

@@ -87,21 +89,21 @@ int groestl256_4way_full( groestl256_4way_context* ctx, void* output,
   if ( i == SIZE256 - 1 )
   {        
       // only 1 vector left in buffer, all padding at once
-      ctx->buffer[i] = m512_const2_64( (uint64_t)blocks << 56, 0x80 ); 
+       ctx->buffer[i] = m512_const2_64( blocks << 56, 0x80 ); 
   }   
   else
   {
       // add first padding
-       ctx->buffer[i] = m512_const4_64( 0, 0x80, 0, 0x80 );
+       ctx->buffer[i] = m512_const2_64( 0, 0x80 );
       // add zero padding
       for ( i += 1; i < SIZE256 - 1; i++ )
           ctx->buffer[i] = m512_zero;

       // add length padding, second last byte is zero unless blocks > 255
-      ctx->buffer[i] = m512_const2_64( (uint64_t)blocks << 56, 0 );
+       ctx->buffer[i] = m512_const2_64( blocks << 56, 0 );
   }

-// digest final padding block and do output transform
+   // digest final padding block and do output transform
   TF512_4way( ctx->chaining, ctx->buffer );

   OF512_4way( ctx->chaining );
@@ -120,7 +122,7 @@ int groestl256_4way_update_close( groestl256_4way_context* ctx, void* output,
   const int hashlen_m128i = ctx->hashlen / 16;   // bytes to __m128i
   const int hash_offset = SIZE256 - hashlen_m128i;
   int rem = ctx->rem_ptr;
-   int blocks = len / SIZE256;
+   uint64_t blocks = len / SIZE256;
   __m512i* in = (__m512i*)input;
   int i;

@@ -144,20 +146,18 @@ int groestl256_4way_update_close( groestl256_4way_context* ctx, void* output,
   if ( i == SIZE256 - 1 )
   {
       // only 1 vector left in buffer, all padding at once
-       ctx->buffer[i] = m512_const1_128( _mm_set_epi8(
-                      blocks, blocks>>8,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0x80 ) );
+       ctx->buffer[i] = m512_const2_64( blocks << 56, 0x80 );
   }
   else
   {
       // add first padding
-       ctx->buffer[i] = m512_const4_64( 0, 0x80, 0, 0x80 );
+       ctx->buffer[i] = m512_const2_64( 0, 0x80 );
       // add zero padding
       for ( i += 1; i < SIZE256 - 1; i++ )
           ctx->buffer[i] = m512_zero;

       // add length padding, second last byte is zero unless blocks > 255
-       ctx->buffer[i] = m512_const1_128( _mm_set_epi8(
-                   blocks, blocks>>8, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0 ) );
+       ctx->buffer[i] = m512_const2_64( blocks << 56, 0 );
   }

 // digest final padding block and do output transform
@@ -172,5 +172,159 @@ int groestl256_4way_update_close( groestl256_4way_context* ctx, void* output,
   return 0;
 }

-#endif   // VAES
+#endif   // AVX512

+// AVX2 + VAES
+
+int groestl256_2way_init( groestl256_2way_context* ctx, uint64_t hashlen )
+{
+  int i;
+
+  ctx->hashlen = hashlen;
+
+  if (ctx->chaining == NULL || ctx->buffer == NULL)
+    return 1;
+
+  for ( i = 0; i < SIZE256; i++ )
+  {
+     ctx->chaining[i] = m256_zero;
+     ctx->buffer[i]   = m256_zero;
+  }
+
+  // The only non-zero in the IV is len. It can be hard coded.
+  ctx->chaining[ 3 ] = m256_const2_64( 0, 0x0100000000000000 );
+
+  ctx->buf_ptr = 0;
+  ctx->rem_ptr = 0;
+
+  return 0;
+}
+
+int groestl256_2way_full( groestl256_2way_context* ctx, void* output,
+                                const void* input, uint64_t datalen )
+{
+   const int len = (int)datalen >> 4;
+   const int hashlen_m128i = 32 >> 4;   // bytes to __m128i
+   const int hash_offset = SIZE256 - hashlen_m128i;
+   int rem = ctx->rem_ptr;
+   uint64_t blocks = len / SIZE256;
+   __m256i* in = (__m256i*)input;
+   int i;
+
+   if (ctx->chaining == NULL || ctx->buffer == NULL)
+     return 1;
+
+   for ( i = 0; i < SIZE256; i++ )
+   {
+     ctx->chaining[i] = m256_zero;
+     ctx->buffer[i]   = m256_zero;
+   }
+
+   // The only non-zero in the IV is len. It can be hard coded.
+   ctx->chaining[ 3 ] = m256_const2_64( 0, 0x0100000000000000 );
+   ctx->buf_ptr = 0;
+   ctx->rem_ptr = 0;
+
+   // --- update ---
+
+   // digest any full blocks, process directly from input 
+   for ( i = 0; i < blocks; i++ )
+      TF512_2way( ctx->chaining, &in[ i * SIZE256 ] );
+   ctx->buf_ptr = blocks * SIZE256;
+
+   // copy any remaining data to buffer, it may already contain data
+   // from a previous update for a midstate precalc
+   for ( i = 0; i < len % SIZE256; i++ )
+       ctx->buffer[ rem + i ] = in[ ctx->buf_ptr + i ];
+   i += rem;    // use i as rem_ptr in final
+
+   //--- final ---
+
+   blocks++;      // adjust for final block
+
+   if ( i == SIZE256 - 1 )
+   {
+       // only 1 vector left in buffer, all padding at once
+      ctx->buffer[i] = m256_const2_64( blocks << 56, 0x80 );
+   }
+   else
+   {
+       // add first padding
+       ctx->buffer[i] = m256_const2_64( 0, 0x80 );
+       // add zero padding
+       for ( i += 1; i < SIZE256 - 1; i++ )
+           ctx->buffer[i] = m256_zero;
+
+       // add length padding, second last byte is zero unless blocks > 255
+       ctx->buffer[i] = m256_const2_64( blocks << 56, 0 );
+   }
+
+   // digest final padding block and do output transform
+   TF512_2way( ctx->chaining, ctx->buffer );
+
+   OF512_2way( ctx->chaining );
+
+   // store hash result in output 
+   for ( i = 0; i < hashlen_m128i; i++ )
+      casti_m256i( output, i ) = ctx->chaining[ hash_offset + i ];
+
+   return 0;
+}
+int groestl256_2way_update_close( groestl256_2way_context* ctx, void* output,
+                                const void* input, uint64_t databitlen )
+{
+   const int len = (int)databitlen / 128;
+   const int hashlen_m128i = ctx->hashlen / 16;   // bytes to __m128i
+   const int hash_offset = SIZE256 - hashlen_m128i;
+   int rem = ctx->rem_ptr;
+   uint64_t blocks = len / SIZE256;
+   __m256i* in = (__m256i*)input;
+   int i;
+
+   // --- update ---
+
+   // digest any full blocks, process directly from input 
+   for ( i = 0; i < blocks; i++ )
+      TF512_2way( ctx->chaining, &in[ i * SIZE256 ] );
+   ctx->buf_ptr = blocks * SIZE256;
+
+   // copy any remaining data to buffer, it may already contain data
+   // from a previous update for a midstate precalc
+   for ( i = 0; i < len % SIZE256; i++ )
+       ctx->buffer[ rem + i ] = in[ ctx->buf_ptr + i ];
+   i += rem;    // use i as rem_ptr in final
+
+   //--- final ---
+
+   blocks++;      // adjust for final block
+
+   if ( i == SIZE256 - 1 )
+   {
+       // only 1 vector left in buffer, all padding at once
+       ctx->buffer[i] = m256_const2_64( blocks << 56, 0x80 );
+   }
+   else
+   {
+       // add first padding
+       ctx->buffer[i] = m256_const2_64( 0, 0x80 );
+       // add zero padding
+       for ( i += 1; i < SIZE256 - 1; i++ )
+           ctx->buffer[i] = m256_zero;
+
+       // add length padding, second last byte is zero unless blocks > 255
+       ctx->buffer[i] = m256_const2_64( blocks << 56, 0 );
+   }
+
+// digest final padding block and do output transform
+   TF512_2way( ctx->chaining, ctx->buffer );
+
+   OF512_2way( ctx->chaining );
+
+   // store hash result in output 
+   for ( i = 0; i < hashlen_m128i; i++ )
+      casti_m256i( output, i ) = ctx->chaining[ hash_offset + i ];
+
+   return 0;
+}
+
+#endif  // VAES
--- a/algo/groestl/groestl256-hash-4way.h
+++ b/algo/groestl/groestl256-hash-4way.h
@@ -18,8 +18,8 @@
 #endif
 #include <stdlib.h>

-#if defined(__VAES__) && defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
-   
+#if defined(__AVX2__) && defined(__VAES__)
+
 #define LENGTH (256)

 //#include "brg_endian.h"
@@ -48,6 +48,8 @@

 #define SIZE256 (SIZE_512/16)

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 typedef struct {
  __attribute__ ((aligned (128))) __m512i chaining[SIZE256];
  __attribute__ ((aligned (64))) __m512i buffer[SIZE256];
@@ -55,7 +57,7 @@ typedef struct {
  int blk_count;     // SIZE_m128i
  int buf_ptr;       // __m128i offset
  int rem_ptr;
-  int databitlen;    // bits
+//  int databitlen;    // bits
 } groestl256_4way_context;


@@ -74,5 +76,25 @@ int groestl256_4way_update_close( groestl256_4way_context*,  void*,
 int groestl256_4way_full( groestl256_4way_context*, void*,
                          const void*, uint64_t );

-#endif
-#endif 
+#endif  // AVX512
+
+typedef struct {
+  __attribute__ ((aligned (128))) __m256i chaining[SIZE256];
+  __attribute__ ((aligned (64))) __m256i buffer[SIZE256];
+  int hashlen;       // byte
+  int blk_count;     // SIZE_m128i
+  int buf_ptr;       // __m128i offset
+  int rem_ptr;
+//  int databitlen;    // bits
+} groestl256_2way_context;
+
+int groestl256_2way_init( groestl256_2way_context*, uint64_t );
+
+int groestl256_2way_update_close( groestl256_2way_context*,  void*,
+                                        const void*, uint64_t );
+
+int groestl256_2way_full( groestl256_2way_context*, void*,
+                          const void*, uint64_t );
+
+#endif  // VAES
+#endif  // GROESTL256_HASH_4WAY_H__
--- a/algo/groestl/groestl256-intr-4way.h
+++ b/algo/groestl/groestl256-intr-4way.h
@@ -7,13 +7,13 @@
 * This code is placed in the public domain
 */

-
 #if !defined(GROESTL256_INTR_4WAY_H__)
 #define GROESTL256_INTR_4WAY_H__ 1
      
 #include "groestl256-hash-4way.h"

-#if defined(__VAES__)
+#if defined(__AVX2__) && defined(__VAES__)
+
 static const __m128i round_const_l0[] __attribute__ ((aligned (64))) =
 {
   { 0x7060504030201000, 0xffffffffffffffff },
@@ -42,6 +42,8 @@ static const __m128i round_const_l7[] __attribute__ ((aligned (64))) =
   { 0x0000000000000000, 0x8696a6b6c6d6e6f6 }
 };

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 static const __m512i TRANSP_MASK = { 0x0d0509010c040800, 0x0f070b030e060a02,
                                     0x1d1519111c141810, 0x1f171b131e161a12,
                                     0x2d2529212c242820, 0x2f272b232e262a22,
@@ -94,11 +96,9 @@ static const __m512i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e,
 * xmm[j] will be lost
 * xmm[k] has to be all 0x1b */
 #define MUL2(i, j, k){\
-  j = _mm512_xor_si512(j, j);\
-  j = _mm512_movm_epi8( _mm512_cmpgt_epi8_mask(j, i) );\
+  j = _mm512_movm_epi8( _mm512_cmpgt_epi8_mask( m512_zero, i) );\
  i = _mm512_add_epi8(i, i);\
-  j = _mm512_and_si512(j, k);\
-  i = _mm512_xor_si512(i, j);\
+  i = mm512_xorand( i, j, k );\
 } 

 /* Yet another implementation of MixBytes.
@@ -118,6 +118,95 @@ static const __m512i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e,
   We almost fit into 16 registers, need only 3 spills to memory.
   This implementation costs 7.7 c/b giving total speed on SNB: 10.7c/b.
   K. Matusiewicz, 2011/05/29 */
+
+#define MixBytes( a0, a1, a2, a3, a4, a5, a6, a7, \
+                  b0, b1, b2, b3, b4, b5, b6, b7) { \
+  /* t_i = a_i + a_{i+1} */\
+  b6 = a0; \
+  b7 = a1; \
+  a0 = _mm512_xor_si512( a0, a1 ); \
+  b0 = a2; \
+  a1 = _mm512_xor_si512( a1, a2 ); \
+  b1 = a3; \
+  TEMP2 = _mm512_xor_si512( a2, a3 ); \
+  b2 = a4; \
+  a3 = _mm512_xor_si512( a3, a4 ); \
+  b3 = a5; \
+  a4 = _mm512_xor_si512( a4, a5 );\
+  b4 = a6; \
+  a5 = _mm512_xor_si512( a5, a6 ); \
+  b5 = a7; \
+  a6 = _mm512_xor_si512( a6, a7 ); \
+  a7 = _mm512_xor_si512( a7, b6 ); \
+  \
+  /* build y4 y5 y6 ... in regs xmm8, xmm9, xmm10 by adding t_i*/\
+  TEMP0 = mm512_xor3( b0, a4, a6 ); \
+  /* spill values y_4, y_5 to memory */\
+  TEMP1 = mm512_xor3( b1, a5, a7 ); \
+  b2 = mm512_xor3( b2, a6, a0 ); \
+  /* save values t0, t1, t2 to xmm8, xmm9 and memory */\
+  b0 = a0; \
+  b3 = mm512_xor3( b3, a7, a1 ); \
+  b1 = a1; \
+  b6 = mm512_xor3( b6, a4, TEMP2 ); \
+  b4 = mm512_xor3( b4, a0, TEMP2 ); \
+  b7 = mm512_xor3( b7, a5, a3 ); \
+  b5 = mm512_xor3( b5, a1, a3 ); \
+  \
+  /* compute x_i = t_i + t_{i+3} */\
+  a0 = _mm512_xor_si512( a0, a3 ); \
+  a1 = _mm512_xor_si512( a1, a4 ); \
+  a2 = _mm512_xor_si512( TEMP2, a5 ); \
+  a3 = _mm512_xor_si512( a3, a6 ); \
+  a4 = _mm512_xor_si512( a4, a7 ); \
+  a5 = _mm512_xor_si512( a5, b0 ); \
+  a6 = _mm512_xor_si512( a6, b1 ); \
+  a7 = _mm512_xor_si512( a7, TEMP2 ); \
+  \
+  /* compute z_i : double x_i using temp xmm8 and 1B xmm9 */\
+  /* compute w_i : add y_{i+4} */\
+  b1 = m512_const1_64( 0x1b1b1b1b1b1b1b1b ); \
+  MUL2( a0, b0, b1 ); \
+  a0 = _mm512_xor_si512( a0, TEMP0 ); \
+  MUL2( a1, b0, b1 ); \
+  a1 = _mm512_xor_si512( a1, TEMP1 ); \
+  MUL2( a2, b0, b1 ); \
+  a2 = _mm512_xor_si512( a2, b2 ); \
+  MUL2( a3, b0, b1 ); \
+  a3 = _mm512_xor_si512( a3, b3 ); \
+  MUL2( a4, b0, b1 ); \
+  a4 = _mm512_xor_si512( a4, b4 ); \
+  MUL2( a5, b0, b1 ); \
+  a5 = _mm512_xor_si512( a5, b5 ); \
+  MUL2( a6, b0, b1 ); \
+  a6 = _mm512_xor_si512( a6, b6 ); \
+  MUL2( a7, b0, b1 ); \
+  a7 = _mm512_xor_si512( a7, b7 ); \
+  \
+  /* compute v_i : double w_i      */\
+  /* add to y_4 y_5 .. v3, v4, ... */\
+  MUL2( a0, b0, b1 ); \
+  b5 = _mm512_xor_si512( b5, a0 ); \
+  MUL2( a1, b0, b1 ); \
+  b6 = _mm512_xor_si512( b6, a1 ); \
+  MUL2( a2, b0, b1 ); \
+  b7 = _mm512_xor_si512( b7, a2 ); \
+  MUL2( a5, b0, b1 ); \
+  b2 = _mm512_xor_si512( b2, a5 ); \
+  MUL2( a6, b0, b1 ); \
+  b3 = _mm512_xor_si512( b3, a6 ); \
+  MUL2( a7, b0, b1 ); \
+  b4 = _mm512_xor_si512( b4, a7 ); \
+  MUL2( a3, b0, b1 ); \
+  MUL2( a4, b0, b1 ); \
+  b0 = TEMP0;\
+  b1 = TEMP1;\
+  b0 = _mm512_xor_si512( b0, a3 ); \
+  b1 = _mm512_xor_si512( b1, a4 ); \
+}/*MixBytes*/
+
+
+#if 0
 #define MixBytes(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
  /* t_i = a_i + a_{i+1} */\
  b6 = a0;\
@@ -213,7 +302,7 @@ static const __m512i SUBSH_MASK7 = { 0x090c000306080b07, 0x02050f0a0d01040e,
  b0 = _mm512_xor_si512(b0, a3);\
  b1 = _mm512_xor_si512(b1, a4);\
 }/*MixBytes*/
-
+#endif

 #define ROUND(i, a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
  /* AddRoundConstant */\
@@ -499,5 +588,398 @@ void OF512_4way( __m512i* chaining )
  chaining[3] = xmm11;
 }

+#endif  // AVX512
+
+static const __m256i TRANSP_MASK_2WAY =
+             { 0x0d0509010c040800, 0x0f070b030e060a02,
+               0x1d1519111c141810, 0x1f171b131e161a12 };
+
+static const __m256i SUBSH_MASK0_2WAY =
+             { 0x0c0f0104070b0e00, 0x03060a0d08020509,
+               0x1c1f1114171b1e10, 0x13161a1d18121519 };
+
+static const __m256i SUBSH_MASK1_2WAY =
+             { 0x0e090205000d0801, 0x04070c0f0a03060b,
+               0x1e191215101d1801, 0x14171c1f1a13161b };
+
+static const __m256i SUBSH_MASK2_2WAY =
+               { 0x080b0306010f0a02, 0x05000e090c04070d,
+                 0x181b1316111f1a12, 0x15101e191c14171d };
+
+static const __m256i SUBSH_MASK3_2WAY =
+               { 0x0a0d040702090c03, 0x0601080b0e05000f,
+                 0x1a1d141712191c13, 0x1611181b1e15101f };
+
+static const __m256i SUBSH_MASK4_2WAY =
+               { 0x0b0e0500030a0d04, 0x0702090c0f060108,
+                 0x1b1e1510131a1d14, 0x1712191c1f161118 };
+
+static const __m256i SUBSH_MASK5_2WAY =
+               { 0x0d080601040c0f05, 0x00030b0e0907020a,
+                 0x1d181611141c1f15, 0x10131b1e1917121a };
+
+static const __m256i SUBSH_MASK6_2WAY =
+               { 0x0f0a0702050e0906, 0x01040d080b00030c,
+                 0x1f1a1712151e1916, 0x11141d181b10131c };
+
+static const __m256i SUBSH_MASK7_2WAY =
+               { 0x090c000306080b07, 0x02050f0a0d01040e,
+                 0x191c101316181b17, 0x12151f1a1d11141e, };
+
+#define tos(a)    #a
+#define tostr(a)  tos(a)
+
+/* xmm[i] will be multiplied by 2
+ * xmm[j] will be lost
+ * xmm[k] has to be all 0x1b */
+#define MUL2_2WAY(i, j, k){\
+  j = _mm256_xor_si256(j, j);\
+  j = _mm256_cmpgt_epi8(j, i );\
+  i = _mm256_add_epi8(i, i);\
+  j = _mm256_and_si256(j, k);\
+  i = _mm256_xor_si256(i, j);\
+}
+
+#define MixBytes_2way(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
+  /* t_i = a_i + a_{i+1} */\
+  b6 = a0;\
+  b7 = a1;\
+  a0 = _mm256_xor_si256(a0, a1);\
+  b0 = a2;\
+  a1 = _mm256_xor_si256(a1, a2);\
+  b1 = a3;\
+  a2 = _mm256_xor_si256(a2, a3);\
+  b2 = a4;\
+  a3 = _mm256_xor_si256(a3, a4);\
+  b3 = a5;\
+  a4 = _mm256_xor_si256(a4, a5);\
+  b4 = a6;\
+  a5 = _mm256_xor_si256(a5, a6);\
+  b5 = a7;\
+  a6 = _mm256_xor_si256(a6, a7);\
+  a7 = _mm256_xor_si256(a7, b6);\
+  \
+  /* build y4 y5 y6 ... in regs xmm8, xmm9, xmm10 by adding t_i*/\
+  b0 = _mm256_xor_si256(b0, a4);\
+  b6 = _mm256_xor_si256(b6, a4);\
+  b1 = _mm256_xor_si256(b1, a5);\
+  b7 = _mm256_xor_si256(b7, a5);\
+  b2 = _mm256_xor_si256(b2, a6);\
+  b0 = _mm256_xor_si256(b0, a6);\
+  /* spill values y_4, y_5 to memory */\
+  TEMP0 = b0;\
+  b3 = _mm256_xor_si256(b3, a7);\
+  b1 = _mm256_xor_si256(b1, a7);\
+  TEMP1 = b1;\
+  b4 = _mm256_xor_si256(b4, a0);\
+  b2 = _mm256_xor_si256(b2, a0);\
+  /* save values t0, t1, t2 to xmm8, xmm9 and memory */\
+  b0 = a0;\
+  b5 = _mm256_xor_si256(b5, a1);\
+  b3 = _mm256_xor_si256(b3, a1);\
+  b1 = a1;\
+  b6 = _mm256_xor_si256(b6, a2);\
+  b4 = _mm256_xor_si256(b4, a2);\
+  TEMP2 = a2;\
+  b7 = _mm256_xor_si256(b7, a3);\
+  b5 = _mm256_xor_si256(b5, a3);\
+  \
+  /* compute x_i = t_i + t_{i+3} */\
+  a0 = _mm256_xor_si256(a0, a3);\
+  a1 = _mm256_xor_si256(a1, a4);\
+  a2 = _mm256_xor_si256(a2, a5);\
+  a3 = _mm256_xor_si256(a3, a6);\
+  a4 = _mm256_xor_si256(a4, a7);\
+  a5 = _mm256_xor_si256(a5, b0);\
+  a6 = _mm256_xor_si256(a6, b1);\
+  a7 = _mm256_xor_si256(a7, TEMP2);\
+  \
+  /* compute z_i : double x_i using temp xmm8 and 1B xmm9 */\
+  /* compute w_i : add y_{i+4} */\
+  b1 = m256_const1_64( 0x1b1b1b1b1b1b1b1b );\
+  MUL2_2WAY(a0, b0, b1);\
+  a0 = _mm256_xor_si256(a0, TEMP0);\
+  MUL2_2WAY(a1, b0, b1);\
+  a1 = _mm256_xor_si256(a1, TEMP1);\
+  MUL2_2WAY(a2, b0, b1);\
+  a2 = _mm256_xor_si256(a2, b2);\
+  MUL2_2WAY(a3, b0, b1);\
+  a3 = _mm256_xor_si256(a3, b3);\
+  MUL2_2WAY(a4, b0, b1);\
+  a4 = _mm256_xor_si256(a4, b4);\
+  MUL2_2WAY(a5, b0, b1);\
+  a5 = _mm256_xor_si256(a5, b5);\
+  MUL2_2WAY(a6, b0, b1);\
+  a6 = _mm256_xor_si256(a6, b6);\
+  MUL2_2WAY(a7, b0, b1);\
+  a7 = _mm256_xor_si256(a7, b7);\
+  \
+  /* compute v_i : double w_i      */\
+  /* add to y_4 y_5 .. v3, v4, ... */\
+  MUL2_2WAY(a0, b0, b1);\
+  b5 = _mm256_xor_si256(b5, a0);\
+  MUL2_2WAY(a1, b0, b1);\
+  b6 = _mm256_xor_si256(b6, a1);\
+  MUL2_2WAY(a2, b0, b1);\
+  b7 = _mm256_xor_si256(b7, a2);\
+  MUL2_2WAY(a5, b0, b1);\
+  b2 = _mm256_xor_si256(b2, a5);\
+  MUL2_2WAY(a6, b0, b1);\
+  b3 = _mm256_xor_si256(b3, a6);\
+  MUL2_2WAY(a7, b0, b1);\
+  b4 = _mm256_xor_si256(b4, a7);\
+  MUL2_2WAY(a3, b0, b1);\
+  MUL2_2WAY(a4, b0, b1);\
+  b0 = TEMP0;\
+  b1 = TEMP1;\
+  b0 = _mm256_xor_si256(b0, a3);\
+  b1 = _mm256_xor_si256(b1, a4);\
+}/*MixBytes*/
+
+#define ROUND_2WAY(i, a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
+  /* AddRoundConstant */\
+  b1 = m256_const2_64( 0xffffffffffffffff, 0 ); \
+  a0 = _mm256_xor_si256( a0, m256_const1_128( round_const_l0[i] ) );\
+  a1 = _mm256_xor_si256( a1, b1 );\
+  a2 = _mm256_xor_si256( a2, b1 );\
+  a3 = _mm256_xor_si256( a3, b1 );\
+  a4 = _mm256_xor_si256( a4, b1 );\
+  a5 = _mm256_xor_si256( a5, b1 );\
+  a6 = _mm256_xor_si256( a6, b1 );\
+  a7 = _mm256_xor_si256( a7, m256_const1_128( round_const_l7[i] ) );\
+  \
+  /* ShiftBytes + SubBytes (interleaved) */\
+  b0 = _mm256_xor_si256( b0, b0 );\
+  a0 = _mm256_shuffle_epi8( a0, SUBSH_MASK0_2WAY );\
+  a0 = _mm256_aesenclast_epi128(a0, b0 );\
+  a1 = _mm256_shuffle_epi8( a1, SUBSH_MASK1_2WAY );\
+  a1 = _mm256_aesenclast_epi128(a1, b0 );\
+  a2 = _mm256_shuffle_epi8( a2, SUBSH_MASK2_2WAY );\
+  a2 = _mm256_aesenclast_epi128(a2, b0 );\
+  a3 = _mm256_shuffle_epi8( a3, SUBSH_MASK3_2WAY );\
+  a3 = _mm256_aesenclast_epi128(a3, b0 );\
+  a4 = _mm256_shuffle_epi8( a4, SUBSH_MASK4_2WAY );\
+  a4 = _mm256_aesenclast_epi128(a4, b0 );\
+  a5 = _mm256_shuffle_epi8( a5, SUBSH_MASK5_2WAY );\
+  a5 = _mm256_aesenclast_epi128(a5, b0 );\
+  a6 = _mm256_shuffle_epi8( a6, SUBSH_MASK6_2WAY );\
+  a6 = _mm256_aesenclast_epi128(a6, b0 );\
+  a7 = _mm256_shuffle_epi8( a7, SUBSH_MASK7_2WAY );\
+  a7 = _mm256_aesenclast_epi128( a7, b0 );\
+  \
+  /* MixBytes */\
+  MixBytes_2way(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7);\
+\
+}
+
+/* 10 rounds, P and Q in parallel */
+#define ROUNDS_P_Q_2WAY(){\
+  ROUND_2WAY(0, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);\
+  ROUND_2WAY(1, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);\
+  ROUND_2WAY(2, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);\
+  ROUND_2WAY(3, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);\
+  ROUND_2WAY(4, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);\
+  ROUND_2WAY(5, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);\
+  ROUND_2WAY(6, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);\
+  ROUND_2WAY(7, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);\
+  ROUND_2WAY(8, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);\
+  ROUND_2WAY(9, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);\
+}
+
+#define Matrix_Transpose_A_2way(i0, i1, i2, i3, o1, o2, o3, t0){\
+  t0 = TRANSP_MASK_2WAY;\
+  \
+  i0 = _mm256_shuffle_epi8( i0, t0 );\
+  i1 = _mm256_shuffle_epi8( i1, t0 );\
+  i2 = _mm256_shuffle_epi8( i2, t0 );\
+  i3 = _mm256_shuffle_epi8( i3, t0 );\
+  \
+  o1 = i0;\
+  t0 = i2;\
+  \
+  i0 = _mm256_unpacklo_epi16( i0, i1 );\
+  o1 = _mm256_unpackhi_epi16( o1, i1 );\
+  i2 = _mm256_unpacklo_epi16( i2, i3 );\
+  t0 = _mm256_unpackhi_epi16( t0, i3 );\
+  \
+  i0 = _mm256_shuffle_epi32( i0, 216 );\
+  o1 = _mm256_shuffle_epi32( o1, 216 );\
+  i2 = _mm256_shuffle_epi32( i2, 216 );\
+  t0 = _mm256_shuffle_epi32( t0, 216 );\
+  \
+  o2 = i0;\
+  o3 = o1;\
+  \
+  i0 = _mm256_unpacklo_epi32( i0, i2 );\
+  o1 = _mm256_unpacklo_epi32( o1, t0 );\
+  o2 = _mm256_unpackhi_epi32( o2, i2 );\
+  o3 = _mm256_unpackhi_epi32( o3, t0 );\
+}/**/
+
+#define Matrix_Transpose_B_2way(i0, i1, i2, i3, i4, i5, i6, i7, o1, o2, o3, o4, o5, o6, o7){\
+  o1 = i0;\
+  o2 = i1;\
+  i0 = _mm256_unpacklo_epi64( i0, i4 );\
+  o1 = _mm256_unpackhi_epi64( o1, i4 );\
+  o3 = i1;\
+  o4 = i2;\
+  o2 = _mm256_unpacklo_epi64( o2, i5 );\
+  o3 = _mm256_unpackhi_epi64( o3, i5 );\
+  o5 = i2;\
+  o6 = i3;\
+  o4 = _mm256_unpacklo_epi64( o4, i6 );\
+  o5 = _mm256_unpackhi_epi64( o5, i6 );\
+  o7 = i3;\
+  o6 = _mm256_unpacklo_epi64( o6, i7 );\
+  o7 = _mm256_unpackhi_epi64( o7, i7 );\
+}/**/
+
+#define Matrix_Transpose_B_INV_2way(i0, i1, i2, i3, i4, i5, i6, i7, o0, o1, o2, o3){\
+  o0 = i0;\
+  i0 = _mm256_unpacklo_epi64( i0, i1 );\
+  o0 = _mm256_unpackhi_epi64( o0, i1 );\
+  o1 = i2;\
+  i2 = _mm256_unpacklo_epi64( i2, i3 );\
+  o1 = _mm256_unpackhi_epi64( o1, i3 );\
+  o2 = i4;\
+  i4 = _mm256_unpacklo_epi64( i4, i5 );\
+  o2 = _mm256_unpackhi_epi64( o2, i5 );\
+  o3 = i6;\
+  i6 = _mm256_unpacklo_epi64( i6, i7 );\
+  o3 = _mm256_unpackhi_epi64( o3, i7 );\
+}/**/
+
+#define Matrix_Transpose_O_B_2way(i0, i1, i2, i3, i4, i5, i6, i7, t0){\
+  t0 = _mm256_xor_si256( t0, t0 );\
+  i1 = i0;\
+  i3 = i2;\
+  i5 = i4;\
+  i7 = i6;\
+  i0 = _mm256_unpacklo_epi64( i0, t0 );\
+  i1 = _mm256_unpackhi_epi64( i1, t0 );\
+  i2 = _mm256_unpacklo_epi64( i2, t0 );\
+  i3 = _mm256_unpackhi_epi64( i3, t0 );\
+  i4 = _mm256_unpacklo_epi64( i4, t0 );\
+  i5 = _mm256_unpackhi_epi64( i5, t0 );\
+  i6 = _mm256_unpacklo_epi64( i6, t0 );\
+  i7 = _mm256_unpackhi_epi64( i7, t0 );\
+}/**/
+
+#define Matrix_Transpose_O_B_INV_2way(i0, i1, i2, i3, i4, i5, i6, i7){\
+  i0 = _mm256_unpacklo_epi64( i0, i1 );\
+  i2 = _mm256_unpacklo_epi64( i2, i3 );\
+  i4 = _mm256_unpacklo_epi64( i4, i5 );\
+  i6 = _mm256_unpacklo_epi64( i6, i7 );\
+}/**/
+
+void TF512_2way( __m256i* chaining, __m256i* message )
+{
+  static __m256i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7;
+  static __m256i xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15;
+  static __m256i TEMP0;
+  static __m256i TEMP1;
+  static __m256i TEMP2;
+
+  /* load message into registers xmm12 - xmm15 */
+  xmm12 = message[0];
+  xmm13 = message[1];
+  xmm14 = message[2];
+  xmm15 = message[3];
+
+  /* transform message M from column ordering into row ordering */
+  /* we first put two rows (64 bit) of the message into one 128-bit xmm register */
+  Matrix_Transpose_A_2way(xmm12, xmm13, xmm14, xmm15, xmm2, xmm6, xmm7, xmm0);
+
+  /* load previous chaining value */
+  /* we first put two rows (64 bit) of the CV into one 128-bit xmm register */
+  xmm8 = chaining[0];
+  xmm0 = chaining[1];
+  xmm4 = chaining[2];
+  xmm5 = chaining[3];
+
+  /* xor message to CV get input of P */
+  /* result: CV+M in xmm8, xmm0, xmm4, xmm5 */
+  xmm8 = _mm256_xor_si256( xmm8, xmm12 );
+  xmm0 = _mm256_xor_si256( xmm0, xmm2 );
+  xmm4 = _mm256_xor_si256( xmm4, xmm6 );
+  xmm5 = _mm256_xor_si256( xmm5, xmm7 );
+
+  /* there are now 2 rows of the Groestl state (P and Q) in each xmm register */
+  /* unpack to get 1 row of P (64 bit) and Q (64 bit) into one xmm register */
+  /* result: the 8 rows of P and Q in xmm8 - xmm12 */
+  Matrix_Transpose_B_2way(xmm8, xmm0, xmm4, xmm5, xmm12, xmm2, xmm6, xmm7, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);
+
+  /* compute the two permutations P and Q in parallel */
+  ROUNDS_P_Q_2WAY();
+
+  /* unpack again to get two rows of P or two rows of Q in one xmm register */
+  Matrix_Transpose_B_INV_2way(xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3);
+
+  /* xor output of P and Q */
+  /* result: P(CV+M)+Q(M) in xmm0...xmm3 */
+  xmm0 = _mm256_xor_si256( xmm0, xmm8 );
+  xmm1 = _mm256_xor_si256( xmm1, xmm10 );
+  xmm2 = _mm256_xor_si256( xmm2, xmm12 );
+  xmm3 = _mm256_xor_si256( xmm3, xmm14 );
+
+  /* xor CV (feed-forward) */
+  /* result: P(CV+M)+Q(M)+CV in xmm0...xmm3 */
+  xmm0 = _mm256_xor_si256( xmm0, (chaining[0]) );
+  xmm1 = _mm256_xor_si256( xmm1, (chaining[1]) );
+  xmm2 = _mm256_xor_si256( xmm2, (chaining[2]) );
+  xmm3 = _mm256_xor_si256( xmm3, (chaining[3]) );
+
+  /* store CV */
+  chaining[0] = xmm0;
+  chaining[1] = xmm1;
+  chaining[2] = xmm2;
+  chaining[3] = xmm3;
+
+  return;
+}
+  
+void OF512_2way( __m256i* chaining )
+{
+  static __m256i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7;
+  static __m256i xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15;
+  static __m256i TEMP0;
+  static __m256i TEMP1;
+  static __m256i TEMP2;
+
+  /* load CV into registers xmm8, xmm10, xmm12, xmm14 */
+  xmm8 = chaining[0];
+  xmm10 = chaining[1];
+  xmm12 = chaining[2];
+  xmm14 = chaining[3];
+
+  /* there are now 2 rows of the CV in one xmm register */
+  /* unpack to get 1 row of P (64 bit) into one half of an xmm register */
+  /* result: the 8 input rows of P in xmm8 - xmm15 */
+  Matrix_Transpose_O_B_2way(xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0);
+
+  /* compute the permutation P */
+  /* result: the output of P(CV) in xmm8 - xmm15 */
+  ROUNDS_P_Q_2WAY();
+
+  /* unpack again to get two rows of P in one xmm register */
+  /* result: P(CV) in xmm8, xmm10, xmm12, xmm14 */
+  Matrix_Transpose_O_B_INV_2way(xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);
+
+  /* xor CV to P output (feed-forward) */
+  /* result: P(CV)+CV in xmm8, xmm10, xmm12, xmm14 */
+  xmm8  = _mm256_xor_si256( xmm8,  (chaining[0]) );
+  xmm10 = _mm256_xor_si256( xmm10, (chaining[1]) );
+  xmm12 = _mm256_xor_si256( xmm12, (chaining[2]) );
+  xmm14 = _mm256_xor_si256( xmm14, (chaining[3]) );
+
+  /* transform state back from row ordering into column ordering */
+  /* result: final hash value in xmm9, xmm11 */
+  Matrix_Transpose_A_2way(xmm8, xmm10, xmm12, xmm14, xmm4, xmm9, xmm11, xmm0);
+
+  /* we only need to return the truncated half of the state */
+  chaining[2] = xmm9;
+  chaining[3] = xmm11;
+}
+
 #endif  // VAES
-#endif  // GROESTL512_INTR_4WAY_H__
+#endif  // GROESTL256_INTR_4WAY_H__
--- a/algo/groestl/groestl512-hash-4way.c
+++ b/algo/groestl/groestl512-hash-4way.c
@@ -15,7 +15,9 @@
 #include "miner.h"
 #include "simd-utils.h"

-#if defined(__VAES__) && defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(__AVX2__) && defined(__VAES__)
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

 int groestl512_4way_init( groestl512_4way_context* ctx, uint64_t hashlen )
 {
@@ -41,7 +43,7 @@ int groestl512_4way_update_close( groestl512_4way_context* ctx, void* output,
   const int hashlen_m128i = 64 / 16;   // bytes to __m128i
   const int hash_offset = SIZE512 - hashlen_m128i;
   int rem = ctx->rem_ptr;
-   int blocks = len / SIZE512;
+   uint64_t blocks = len / SIZE512;
   __m512i* in = (__m512i*)input;
   int i;

@@ -62,16 +64,14 @@ int groestl512_4way_update_close( groestl512_4way_context* ctx, void* output,
   if ( i == SIZE512 - 1 )
   {        
       // only 1 vector left in buffer, all padding at once
-       ctx->buffer[i] = m512_const1_128( _mm_set_epi8(
-                      blocks, blocks>>8,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0x80 ) );
+       ctx->buffer[i] = m512_const2_64( blocks << 56, 0x80 );
   }   
   else
   {
-       ctx->buffer[i] = m512_const4_64( 0, 0x80, 0, 0x80 );
+       ctx->buffer[i] = m512_const2_64( 0, 0x80 );
       for ( i += 1; i < SIZE512 - 1; i++ )
           ctx->buffer[i] = m512_zero;
-       ctx->buffer[i] = m512_const1_128( _mm_set_epi8(
-                   blocks, blocks>>8, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0, 0,0 ) );
+       ctx->buffer[i] = m512_const2_64( blocks << 56, 0 );
   }

   TF1024_4way( ctx->chaining, ctx->buffer );
@@ -122,7 +122,7 @@ int groestl512_4way_full( groestl512_4way_context* ctx, void* output,
   }
   else
   {
-       ctx->buffer[i] = m512_const4_64( 0, 0x80, 0, 0x80 );
+       ctx->buffer[i] = m512_const2_64( 0, 0x80 );
       for ( i += 1; i < SIZE512 - 1; i++ )
           ctx->buffer[i] = m512_zero;
       ctx->buffer[i] = m512_const2_64( blocks << 56, 0 );
@@ -137,5 +137,128 @@ int groestl512_4way_full( groestl512_4way_context* ctx, void* output,
   return 0;
 }

+#endif   // AVX512
+
+
+// AVX2 + VAES
+
+int groestl512_2way_init( groestl512_2way_context* ctx, uint64_t hashlen )
+{
+  if (ctx->chaining == NULL || ctx->buffer == NULL)
+    return 1;
+
+  memset_zero_256( ctx->chaining, SIZE512 );
+  memset_zero_256( ctx->buffer, SIZE512 );
+
+  // The only non-zero in the IV is len. It can be hard coded.
+  ctx->chaining[ 6 ] = m256_const2_64( 0x0200000000000000, 0 );
+
+  ctx->buf_ptr = 0;
+  ctx->rem_ptr = 0;
+
+  return 0;
+}
+
+int groestl512_2way_update_close( groestl512_2way_context* ctx, void* output,
+                                const void* input, uint64_t databitlen )
+{
+   const int len = (int)databitlen / 128;
+   const int hashlen_m128i = 64 / 16;   // bytes to __m128i
+   const int hash_offset = SIZE512 - hashlen_m128i;
+   int rem = ctx->rem_ptr;
+   uint64_t blocks = len / SIZE512;
+   __m256i* in = (__m256i*)input;
+   int i;
+
+   // --- update ---
+
+   for ( i = 0; i < blocks; i++ )
+      TF1024_2way( ctx->chaining, &in[ i * SIZE512 ] );
+   ctx->buf_ptr = blocks * SIZE512;
+
+   for ( i = 0; i < len % SIZE512; i++ )
+       ctx->buffer[ rem + i ] = in[ ctx->buf_ptr + i ];
+   i += rem;
+
+   //--- final ---
+
+   blocks++;      // adjust for final block
+
+   if ( i == SIZE512 - 1 )
+   {
+       // only 1 vector left in buffer, all padding at once
+       ctx->buffer[i] = m256_const2_64( blocks << 56, 0x80 );
+   }
+   else
+   {
+       ctx->buffer[i] = m256_const2_64( 0, 0x80 );
+       for ( i += 1; i < SIZE512 - 1; i++ )
+           ctx->buffer[i] = m256_zero;
+       ctx->buffer[i] = m256_const2_64( blocks << 56, 0 );
+   }
+
+   TF1024_2way( ctx->chaining, ctx->buffer );
+   OF1024_2way( ctx->chaining );
+
+   for ( i = 0; i < hashlen_m128i; i++ )
+      casti_m256i( output, i ) = ctx->chaining[ hash_offset + i ];
+
+   return 0;
+}
+
+int groestl512_2way_full( groestl512_2way_context* ctx, void* output,
+                          const void* input, uint64_t datalen )
+{
+   const int len = (int)datalen >> 4;
+   const int hashlen_m128i = 64 >> 4;   // bytes to __m128i
+   const int hash_offset = SIZE512 - hashlen_m128i;
+   uint64_t blocks = len / SIZE512;
+   __m256i* in = (__m256i*)input;
+   int i;
+
+   // --- init ---
+
+   memset_zero_256( ctx->chaining, SIZE512 );
+   memset_zero_256( ctx->buffer, SIZE512 );
+   ctx->chaining[ 6 ] = m256_const2_64( 0x0200000000000000, 0 );
+   ctx->buf_ptr = 0;
+   ctx->rem_ptr = 0;
+
+   // --- update ---
+
+   for ( i = 0; i < blocks; i++ )
+      TF1024_2way( ctx->chaining, &in[ i * SIZE512 ] );
+   ctx->buf_ptr = blocks * SIZE512;
+
+   for ( i = 0; i < len % SIZE512; i++ )
+       ctx->buffer[ ctx->rem_ptr + i ] = in[ ctx->buf_ptr + i ];
+   i += ctx->rem_ptr;
+
+   // --- close ---
+
+   blocks++;
+
+   if ( i == SIZE512 - 1 )
+   {
+       // only 1 vector left in buffer, all padding at once
+       ctx->buffer[i] = m256_const2_64( blocks << 56, 0x80 );
+   }
+   else
+   {
+       ctx->buffer[i] = m256_const2_64( 0, 0x80 );
+       for ( i += 1; i < SIZE512 - 1; i++ )
+           ctx->buffer[i] = m256_zero;
+       ctx->buffer[i] = m256_const2_64( blocks << 56, 0 );
+   }
+
+   TF1024_2way( ctx->chaining, ctx->buffer );
+   OF1024_2way( ctx->chaining );
+
+   for ( i = 0; i < hashlen_m128i; i++ )
+      casti_m256i( output, i ) = ctx->chaining[ hash_offset + i ];
+
+   return 0;
+}
+   
 #endif   // VAES

--- a/algo/groestl/groestl512-hash-4way.h
+++ b/algo/groestl/groestl512-hash-4way.h
@@ -10,7 +10,7 @@
 #endif
 #include <stdlib.h>

-#if defined(__VAES__) && defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(__AVX2__) && defined(__VAES__)

 #define LENGTH (512)

@@ -36,20 +36,19 @@

 #define SIZE512 (SIZE_1024/16)

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 typedef struct {
  __attribute__ ((aligned (128))) __m512i chaining[SIZE512];
  __attribute__ ((aligned (64))) __m512i buffer[SIZE512];
  int blk_count;     // SIZE_m128i
  int buf_ptr;       // __m128i offset
  int rem_ptr;
-  int databitlen;    // bits
 } groestl512_4way_context;


 int groestl512_4way_init( groestl512_4way_context*, uint64_t );

-//int reinit_groestl( hashState_groestl* );
-
 int groestl512_4way_update( groestl512_4way_context*, const void*,
                              uint64_t );
 int groestl512_4way_close( groestl512_4way_context*, void* );
@@ -58,5 +57,29 @@ int groestl512_4way_update_close( groestl512_4way_context*,  void*,
 int groestl512_4way_full( groestl512_4way_context*,  void*,
                          const void*, uint64_t );

+#endif   // AVX512
+
+// AVX2 + VAES
+
+typedef struct {
+  __attribute__ ((aligned (128))) __m256i chaining[SIZE512];
+  __attribute__ ((aligned (64))) __m256i buffer[SIZE512];
+  int blk_count;     // SIZE_m128i
+  int buf_ptr;       // __m128i offset
+  int rem_ptr;
+} groestl512_2way_context;
+
+
+int groestl512_2way_init( groestl512_2way_context*, uint64_t );
+
+int groestl512_2way_update( groestl512_2way_context*, const void*,
+                              uint64_t );
+int groestl512_2way_close( groestl512_2way_context*, void* );
+int groestl512_2way_update_close( groestl512_2way_context*,  void*,
+                                        const void*, uint64_t );
+int groestl512_2way_full( groestl512_2way_context*,  void*,
+                          const void*, uint64_t );
+
+
 #endif   // VAES
 #endif   // GROESTL512_HASH_4WAY_H__
--- a/algo/groestl/groestl512-intr-4way.h
+++ b/algo/groestl/groestl512-intr-4way.h
@@ -7,13 +7,12 @@
 * This code is placed in the public domain
 */

-
 #if !defined(GROESTL512_INTR_4WAY_H__)
 #define GROESTL512_INTR_4WAY_H__ 1
      
 #include "groestl512-hash-4way.h"

-#if defined(__VAES__)
+#if defined(__AVX2__) && defined(__VAES__)

 static const __m128i round_const_p[] __attribute__ ((aligned (64))) =
 {
@@ -51,6 +50,8 @@ static const __m128i round_const_q[] __attribute__ ((aligned (64))) =
   { 0x8292a2b2c2d2e2f2, 0x0212223242526272 }
 };

+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
 static const __m512i TRANSP_MASK = { 0x0d0509010c040800, 0x0f070b030e060a02,
                                     0x1d1519111c141810, 0x1f171b131e161a12,
                                     0x2d2529212c242820, 0x2f272b232e262a22,
@@ -103,11 +104,9 @@ static const __m512i SUBSH_MASK7 = { 0x06090c0f0205080b, 0x0e0104070a0d0003,
 * xmm[j] will be lost
 * xmm[k] has to be all 0x1b */
 #define MUL2(i, j, k){\
-  j = _mm512_xor_si512(j, j);\
-  j = _mm512_movm_epi8( _mm512_cmpgt_epi8_mask(j, i) );\
+  j = _mm512_movm_epi8( _mm512_cmpgt_epi8_mask( m512_zero, i) );\
  i = _mm512_add_epi8(i, i);\
-  j = _mm512_and_si512(j, k);\
-  i = _mm512_xor_si512(i, j);\
+  i = mm512_xorand( i, j, k );\
 } 

 /**/
@@ -129,100 +128,90 @@ static const __m512i SUBSH_MASK7 = { 0x06090c0f0205080b, 0x0e0104070a0d0003,
   We almost fit into 16 registers, need only 3 spills to memory.
   This implementation costs 7.7 c/b giving total speed on SNB: 10.7c/b.
   K. Matusiewicz, 2011/05/29 */
-#define MixBytes(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
+#define MixBytes( a0, a1, a2, a3, a4, a5, a6, a7, \
+                  b0, b1, b2, b3, b4, b5, b6, b7) { \
  /* t_i = a_i + a_{i+1} */\
-  b6 = a0;\
-  b7 = a1;\
-  a0 = _mm512_xor_si512(a0, a1);\
-  b0 = a2;\
-  a1 = _mm512_xor_si512(a1, a2);\
-  b1 = a3;\
-  a2 = _mm512_xor_si512(a2, a3);\
-  b2 = a4;\
-  a3 = _mm512_xor_si512(a3, a4);\
-  b3 = a5;\
-  a4 = _mm512_xor_si512(a4, a5);\
-  b4 = a6;\
-  a5 = _mm512_xor_si512(a5, a6);\
-  b5 = a7;\
-  a6 = _mm512_xor_si512(a6, a7);\
-  a7 = _mm512_xor_si512(a7, b6);\
+  b6 = a0; \
+  b7 = a1; \
+  a0 = _mm512_xor_si512( a0, a1 ); \
+  b0 = a2; \
+  a1 = _mm512_xor_si512( a1, a2 ); \
+  b1 = a3; \
+  TEMP2 = _mm512_xor_si512( a2, a3 ); \
+  b2 = a4; \
+  a3 = _mm512_xor_si512( a3, a4 ); \
+  b3 = a5; \
+  a4 = _mm512_xor_si512( a4, a5 );\
+  b4 = a6; \
+  a5 = _mm512_xor_si512( a5, a6 ); \
+  b5 = a7; \
+  a6 = _mm512_xor_si512( a6, a7 ); \
+  a7 = _mm512_xor_si512( a7, b6 ); \
  \
  /* build y4 y5 y6 ... in regs xmm8, xmm9, xmm10 by adding t_i*/\
-  b0 = _mm512_xor_si512(b0, a4);\
-  b6 = _mm512_xor_si512(b6, a4);\
-  b1 = _mm512_xor_si512(b1, a5);\
-  b7 = _mm512_xor_si512(b7, a5);\
-  b2 = _mm512_xor_si512(b2, a6);\
-  b0 = _mm512_xor_si512(b0, a6);\
+  TEMP0 = mm512_xor3( b0, a4, a6 ); \
  /* spill values y_4, y_5 to memory */\
-  TEMP0 = b0;\
-  b3 = _mm512_xor_si512(b3, a7);\
-  b1 = _mm512_xor_si512(b1, a7);\
-  TEMP1 = b1;\
-  b4 = _mm512_xor_si512(b4, a0);\
-  b2 = _mm512_xor_si512(b2, a0);\
+  TEMP1 = mm512_xor3( b1, a5, a7 ); \
+  b2 = mm512_xor3( b2, a6, a0 ); \
  /* save values t0, t1, t2 to xmm8, xmm9 and memory */\
-  b0 = a0;\
-  b5 = _mm512_xor_si512(b5, a1);\
-  b3 = _mm512_xor_si512(b3, a1);\
-  b1 = a1;\
-  b6 = _mm512_xor_si512(b6, a2);\
-  b4 = _mm512_xor_si512(b4, a2);\
-  TEMP2 = a2;\
-  b7 = _mm512_xor_si512(b7, a3);\
-  b5 = _mm512_xor_si512(b5, a3);\
+  b0 = a0; \
+  b3 = mm512_xor3( b3, a7, a1 ); \
+  b1 = a1; \
+  b6 = mm512_xor3( b6, a4, TEMP2 ); \
+  b4 = mm512_xor3( b4, a0, TEMP2 ); \
+  b7 = mm512_xor3( b7, a5, a3 ); \
+  b5 = mm512_xor3( b5, a1, a3 ); \
  \
  /* compute x_i = t_i + t_{i+3} */\
-  a0 = _mm512_xor_si512(a0, a3);\
-  a1 = _mm512_xor_si512(a1, a4);\
-  a2 = _mm512_xor_si512(a2, a5);\
-  a3 = _mm512_xor_si512(a3, a6);\
-  a4 = _mm512_xor_si512(a4, a7);\
-  a5 = _mm512_xor_si512(a5, b0);\
-  a6 = _mm512_xor_si512(a6, b1);\
-  a7 = _mm512_xor_si512(a7, TEMP2);\
+  a0 = _mm512_xor_si512( a0, a3 ); \
+  a1 = _mm512_xor_si512( a1, a4 ); \
+  a2 = _mm512_xor_si512( TEMP2, a5 ); \
+  a3 = _mm512_xor_si512( a3, a6 ); \
+  a4 = _mm512_xor_si512( a4, a7 ); \
+  a5 = _mm512_xor_si512( a5, b0 ); \
+  a6 = _mm512_xor_si512( a6, b1 ); \
+  a7 = _mm512_xor_si512( a7, TEMP2 ); \
  \
  /* compute z_i : double x_i using temp xmm8 and 1B xmm9 */\
  /* compute w_i : add y_{i+4} */\
-  b1 = m512_const1_64( 0x1b1b1b1b1b1b1b1b );\
-  MUL2(a0, b0, b1);\
-  a0 = _mm512_xor_si512(a0, TEMP0);\
-  MUL2(a1, b0, b1);\
-  a1 = _mm512_xor_si512(a1, TEMP1);\
-  MUL2(a2, b0, b1);\
-  a2 = _mm512_xor_si512(a2, b2);\
-  MUL2(a3, b0, b1);\
-  a3 = _mm512_xor_si512(a3, b3);\
-  MUL2(a4, b0, b1);\
-  a4 = _mm512_xor_si512(a4, b4);\
-  MUL2(a5, b0, b1);\
-  a5 = _mm512_xor_si512(a5, b5);\
-  MUL2(a6, b0, b1);\
-  a6 = _mm512_xor_si512(a6, b6);\
-  MUL2(a7, b0, b1);\
-  a7 = _mm512_xor_si512(a7, b7);\
+  b1 = m512_const1_64( 0x1b1b1b1b1b1b1b1b ); \
+  MUL2( a0, b0, b1 ); \
+  a0 = _mm512_xor_si512( a0, TEMP0 ); \
+  MUL2( a1, b0, b1 ); \
+  a1 = _mm512_xor_si512( a1, TEMP1 ); \
+  MUL2( a2, b0, b1 ); \
+  a2 = _mm512_xor_si512( a2, b2 ); \
+  MUL2( a3, b0, b1 ); \
+  a3 = _mm512_xor_si512( a3, b3 ); \
+  MUL2( a4, b0, b1 ); \
+  a4 = _mm512_xor_si512( a4, b4 ); \
+  MUL2( a5, b0, b1 ); \
+  a5 = _mm512_xor_si512( a5, b5 ); \
+  MUL2( a6, b0, b1 ); \
+  a6 = _mm512_xor_si512( a6, b6 ); \
+  MUL2( a7, b0, b1 ); \
+  a7 = _mm512_xor_si512( a7, b7 ); \
  \
  /* compute v_i : double w_i      */\
  /* add to y_4 y_5 .. v3, v4, ... */\
-  MUL2(a0, b0, b1);\
-  b5 = _mm512_xor_si512(b5, a0);\
-  MUL2(a1, b0, b1);\
-  b6 = _mm512_xor_si512(b6, a1);\
-  MUL2(a2, b0, b1);\
-  b7 = _mm512_xor_si512(b7, a2);\
-  MUL2(a5, b0, b1);\
-  b2 = _mm512_xor_si512(b2, a5);\
-  MUL2(a6, b0, b1);\
-  b3 = _mm512_xor_si512(b3, a6);\
-  MUL2(a7, b0, b1);\
-  b4 = _mm512_xor_si512(b4, a7);\
-  MUL2(a3, b0, b1);\
-  MUL2(a4, b0, b1);\
+  MUL2( a0, b0, b1 ); \
+  b5 = _mm512_xor_si512( b5, a0 ); \
+  MUL2( a1, b0, b1 ); \
+  b6 = _mm512_xor_si512( b6, a1 ); \
+  MUL2( a2, b0, b1 ); \
+  b7 = _mm512_xor_si512( b7, a2 ); \
+  MUL2( a5, b0, b1 ); \
+  b2 = _mm512_xor_si512( b2, a5 ); \
+  MUL2( a6, b0, b1 ); \
+  b3 = _mm512_xor_si512( b3, a6 ); \
+  MUL2( a7, b0, b1 ); \
+  b4 = _mm512_xor_si512( b4, a7 ); \
+  MUL2( a3, b0, b1 ); \
+  MUL2( a4, b0, b1 ); \
  b0 = TEMP0;\
  b1 = TEMP1;\
-  b0 = _mm512_xor_si512(b0, a3);\
-  b1 = _mm512_xor_si512(b1, a4);\
+  b0 = _mm512_xor_si512( b0, a3 ); \
+  b1 = _mm512_xor_si512( b1, a4 ); \
 }/*MixBytes*/

 /* one round
@@ -661,5 +650,576 @@ void OF1024_4way( __m512i* chaining )
  return;
 }

+#endif  // AVX512
+
+// AVX2 + VAES
+
+static const __m256i TRANSP_MASK_2WAY =
+    { 0x0d0509010c040800, 0x0f070b030e060a02,
+      0x1d1519111c141810, 0x1f171b131e161a12 };
+
+static const __m256i SUBSH_MASK0_2WAY =
+    { 0x0b0e0104070a0d00, 0x0306090c0f020508,
+      0x1b1e1114171a1d10, 0x1316191c1f121518 };
+
+static const __m256i SUBSH_MASK1_2WAY =
+    { 0x0c0f0205080b0e01, 0x04070a0d00030609,
+      0x1c1f1215181b1e11, 0x14171a1d10131619 };
+
+static const __m256i SUBSH_MASK2_2WAY =
+    { 0x0d000306090c0f02, 0x05080b0e0104070a,
+      0x1d101316191c1f12, 0x15181b1e1114171a };
+
+static const __m256i SUBSH_MASK3_2WAY =
+    { 0x0e0104070a0d0003, 0x06090c0f0205080b,
+      0x1e1114171a1d1013, 0x16191c1f1215181b };
+
+static const __m256i SUBSH_MASK4_2WAY = 
+    { 0x0f0205080b0e0104, 0x070a0d000306090c,
+      0x1f1215181b1e1114, 0x171a1d101316191c };
+
+static const __m256i SUBSH_MASK5_2WAY =
+    { 0x000306090c0f0205, 0x080b0e0104070a0d,
+      0x101316191c1f1215, 0x181b1e1114171a1d };
+
+static const __m256i SUBSH_MASK6_2WAY =
+    { 0x0104070a0d000306, 0x090c0f0205080b0e,
+      0x1114171a1d101316, 0x191c1f1215181b1e };
+
+static const __m256i SUBSH_MASK7_2WAY =
+    { 0x06090c0f0205080b, 0x0e0104070a0d0003,
+      0x16191c1f1215181b, 0x1e1114171a1d1013 };
+
+#define tos(a)    #a
+#define tostr(a)  tos(a)
+
+/* xmm[i] will be multiplied by 2
+ * xmm[j] will be lost
+ * xmm[k] has to be all 0x1b */
+#define MUL2_2WAY(i, j, k){\
+  j = _mm256_cmpgt_epi8( m256_zero, i );\
+  i = _mm256_add_epi8(i, i);\
+  i = mm256_xorand( i, j, k );\
+}
+
+#define MixBytes_2way(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
+  /* t_i = a_i + a_{i+1} */\
+  b6 = a0;\
+  b7 = a1;\
+  a0 = _mm256_xor_si256(a0, a1);\
+  b0 = a2;\
+  a1 = _mm256_xor_si256(a1, a2);\
+  b1 = a3;\
+  a2 = _mm256_xor_si256(a2, a3);\
+  b2 = a4;\
+  a3 = _mm256_xor_si256(a3, a4);\
+  b3 = a5;\
+  a4 = _mm256_xor_si256(a4, a5);\
+  b4 = a6;\
+  a5 = _mm256_xor_si256(a5, a6);\
+  b5 = a7;\
+  a6 = _mm256_xor_si256(a6, a7);\
+  a7 = _mm256_xor_si256(a7, b6);\
+  \
+  /* build y4 y5 y6 ... in regs xmm8, xmm9, xmm10 by adding t_i*/\
+  b0 = _mm256_xor_si256(b0, a4);\
+  b6 = _mm256_xor_si256(b6, a4);\
+  b1 = _mm256_xor_si256(b1, a5);\
+  b7 = _mm256_xor_si256(b7, a5);\
+  b2 = _mm256_xor_si256(b2, a6);\
+  b0 = _mm256_xor_si256(b0, a6);\
+  /* spill values y_4, y_5 to memory */\
+  TEMP0 = b0;\
+  b3 = _mm256_xor_si256(b3, a7);\
+  b1 = _mm256_xor_si256(b1, a7);\
+  TEMP1 = b1;\
+  b4 = _mm256_xor_si256(b4, a0);\
+  b2 = _mm256_xor_si256(b2, a0);\
+  /* save values t0, t1, t2 to xmm8, xmm9 and memory */\
+  b0 = a0;\
+  b5 = _mm256_xor_si256(b5, a1);\
+  b3 = _mm256_xor_si256(b3, a1);\
+  b1 = a1;\
+  b6 = _mm256_xor_si256(b6, a2);\
+  b4 = _mm256_xor_si256(b4, a2);\
+  TEMP2 = a2;\
+  b7 = _mm256_xor_si256(b7, a3);\
+  b5 = _mm256_xor_si256(b5, a3);\
+  \
+  /* compute x_i = t_i + t_{i+3} */\
+  a0 = _mm256_xor_si256(a0, a3);\
+  a1 = _mm256_xor_si256(a1, a4);\
+  a2 = _mm256_xor_si256(a2, a5);\
+  a3 = _mm256_xor_si256(a3, a6);\
+  a4 = _mm256_xor_si256(a4, a7);\
+  a5 = _mm256_xor_si256(a5, b0);\
+  a6 = _mm256_xor_si256(a6, b1);\
+  a7 = _mm256_xor_si256(a7, TEMP2);\
+  \
+  /* compute z_i : double x_i using temp xmm8 and 1B xmm9 */\
+  /* compute w_i : add y_{i+4} */\
+  b1 = m256_const1_64( 0x1b1b1b1b1b1b1b1b );\
+  MUL2_2WAY(a0, b0, b1);\
+  a0 = _mm256_xor_si256(a0, TEMP0);\
+  MUL2_2WAY(a1, b0, b1);\
+  a1 = _mm256_xor_si256(a1, TEMP1);\
+  MUL2_2WAY(a2, b0, b1);\
+  a2 = _mm256_xor_si256(a2, b2);\
+  MUL2_2WAY(a3, b0, b1);\
+  a3 = _mm256_xor_si256(a3, b3);\
+  MUL2_2WAY(a4, b0, b1);\
+  a4 = _mm256_xor_si256(a4, b4);\
+  MUL2_2WAY(a5, b0, b1);\
+  a5 = _mm256_xor_si256(a5, b5);\
+  MUL2_2WAY(a6, b0, b1);\
+  a6 = _mm256_xor_si256(a6, b6);\
+  MUL2_2WAY(a7, b0, b1);\
+  a7 = _mm256_xor_si256(a7, b7);\
+  \
+  /* compute v_i : double w_i      */\
+  /* add to y_4 y_5 .. v3, v4, ... */\
+  MUL2_2WAY(a0, b0, b1);\
+  b5 = _mm256_xor_si256(b5, a0);\
+  MUL2_2WAY(a1, b0, b1);\
+  b6 = _mm256_xor_si256(b6, a1);\
+  MUL2_2WAY(a2, b0, b1);\
+  b7 = _mm256_xor_si256(b7, a2);\
+  MUL2_2WAY(a5, b0, b1);\
+  b2 = _mm256_xor_si256(b2, a5);\
+  MUL2_2WAY(a6, b0, b1);\
+  b3 = _mm256_xor_si256(b3, a6);\
+  MUL2_2WAY(a7, b0, b1);\
+  b4 = _mm256_xor_si256(b4, a7);\
+  MUL2_2WAY(a3, b0, b1);\
+  MUL2_2WAY(a4, b0, b1);\
+  b0 = TEMP0;\
+  b1 = TEMP1;\
+  b0 = _mm256_xor_si256(b0, a3);\
+  b1 = _mm256_xor_si256(b1, a4);\
+}/*MixBytes*/
+
+/* one round
+ * a0-a7 = input rows
+ * b0-b7 = output rows
+ */
+#define SUBMIX_2WAY(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7){\
+  /* SubBytes */\
+  b0 = _mm256_xor_si256( b0, b0 );\
+  a0 = _mm256_aesenclast_epi128( a0, b0 );\
+  a1 = _mm256_aesenclast_epi128( a1, b0 );\
+  a2 = _mm256_aesenclast_epi128( a2, b0 );\
+  a3 = _mm256_aesenclast_epi128( a3, b0 );\
+  a4 = _mm256_aesenclast_epi128( a4, b0 );\
+  a5 = _mm256_aesenclast_epi128( a5, b0 );\
+  a6 = _mm256_aesenclast_epi128( a6, b0 );\
+  a7 = _mm256_aesenclast_epi128( a7, b0 );\
+  /* MixBytes */\
+  MixBytes_2way(a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3, b4, b5, b6, b7);\
+}
+
+#define ROUNDS_P_2WAY(){\
+  uint8_t round_counter = 0;\
+  for ( round_counter = 0; round_counter < 14; round_counter += 2 ) \
+  { \
+    /* AddRoundConstant P1024 */\
+    xmm8 = _mm256_xor_si256( xmm8, m256_const1_128( \
+             casti_m128i( round_const_p, round_counter ) ) ); \
+    /* ShiftBytes P1024 + pre-AESENCLAST */\
+    xmm8  = _mm256_shuffle_epi8( xmm8,  SUBSH_MASK0_2WAY ); \
+    xmm9  = _mm256_shuffle_epi8( xmm9,  SUBSH_MASK1_2WAY );\
+    xmm10 = _mm256_shuffle_epi8( xmm10, SUBSH_MASK2_2WAY );\
+    xmm11 = _mm256_shuffle_epi8( xmm11, SUBSH_MASK3_2WAY );\
+    xmm12 = _mm256_shuffle_epi8( xmm12, SUBSH_MASK4_2WAY );\
+    xmm13 = _mm256_shuffle_epi8( xmm13, SUBSH_MASK5_2WAY );\
+    xmm14 = _mm256_shuffle_epi8( xmm14, SUBSH_MASK6_2WAY );\
+    xmm15 = _mm256_shuffle_epi8( xmm15, SUBSH_MASK7_2WAY );\
+    /* SubBytes + MixBytes */\
+    SUBMIX_2WAY(xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);\
+    \
+     /* AddRoundConstant P1024 */\
+    xmm0 = _mm256_xor_si256( xmm0, m256_const1_128( \
+             casti_m128i( round_const_p, round_counter+1 ) ) ); \
+    /* ShiftBytes P1024 + pre-AESENCLAST */\
+    xmm0 = _mm256_shuffle_epi8( xmm0, SUBSH_MASK0_2WAY );\
+    xmm1 = _mm256_shuffle_epi8( xmm1, SUBSH_MASK1_2WAY );\
+    xmm2 = _mm256_shuffle_epi8( xmm2, SUBSH_MASK2_2WAY );\
+    xmm3 = _mm256_shuffle_epi8( xmm3, SUBSH_MASK3_2WAY );\
+    xmm4 = _mm256_shuffle_epi8( xmm4, SUBSH_MASK4_2WAY );\
+    xmm5 = _mm256_shuffle_epi8( xmm5, SUBSH_MASK5_2WAY );\
+    xmm6 = _mm256_shuffle_epi8( xmm6, SUBSH_MASK6_2WAY );\
+    xmm7 = _mm256_shuffle_epi8( xmm7, SUBSH_MASK7_2WAY );\
+    /* SubBytes + MixBytes */\
+     SUBMIX_2WAY(xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);\
+  }\
+}
+
+#define ROUNDS_Q_2WAY(){\
+  uint8_t round_counter = 0;\
+  for ( round_counter = 0; round_counter < 14; round_counter += 2) \
+  { \
+    /* AddRoundConstant Q1024 */\
+    xmm1 = m256_neg1;\
+    xmm8  = _mm256_xor_si256( xmm8,  xmm1 );\
+    xmm9  = _mm256_xor_si256( xmm9,  xmm1 );\
+    xmm10 = _mm256_xor_si256( xmm10, xmm1 );\
+    xmm11 = _mm256_xor_si256( xmm11, xmm1 );\
+    xmm12 = _mm256_xor_si256( xmm12, xmm1 );\
+    xmm13 = _mm256_xor_si256( xmm13, xmm1 );\
+    xmm14 = _mm256_xor_si256( xmm14, xmm1 );\
+    xmm15 = _mm256_xor_si256( xmm15, m256_const1_128( \
+                 casti_m128i( round_const_q, round_counter ) ) ); \
+    /* ShiftBytes Q1024 + pre-AESENCLAST */\
+    xmm8  = _mm256_shuffle_epi8( xmm8,  SUBSH_MASK1_2WAY );\
+    xmm9  = _mm256_shuffle_epi8( xmm9,  SUBSH_MASK3_2WAY );\
+    xmm10 = _mm256_shuffle_epi8( xmm10, SUBSH_MASK5_2WAY );\
+    xmm11 = _mm256_shuffle_epi8( xmm11, SUBSH_MASK7_2WAY );\
+    xmm12 = _mm256_shuffle_epi8( xmm12, SUBSH_MASK0_2WAY );\
+    xmm13 = _mm256_shuffle_epi8( xmm13, SUBSH_MASK2_2WAY );\
+    xmm14 = _mm256_shuffle_epi8( xmm14, SUBSH_MASK4_2WAY );\
+    xmm15 = _mm256_shuffle_epi8( xmm15, SUBSH_MASK6_2WAY );\
+    /* SubBytes + MixBytes */\
+    SUBMIX_2WAY(xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);\
+    \
+    /* AddRoundConstant Q1024 */\
+    xmm9 = m256_neg1;\
+    xmm0 = _mm256_xor_si256( xmm0, xmm9 );\
+    xmm1 = _mm256_xor_si256( xmm1, xmm9 );\
+    xmm2 = _mm256_xor_si256( xmm2, xmm9 );\
+    xmm3 = _mm256_xor_si256( xmm3, xmm9 );\
+    xmm4 = _mm256_xor_si256( xmm4, xmm9 );\
+    xmm5 = _mm256_xor_si256( xmm5, xmm9 );\
+    xmm6 = _mm256_xor_si256( xmm6, xmm9 );\
+    xmm7 = _mm256_xor_si256( xmm7, m256_const1_128( \
+             casti_m128i( round_const_q, round_counter+1 ) ) ); \
+    /* ShiftBytes Q1024 + pre-AESENCLAST */\
+    xmm0 = _mm256_shuffle_epi8( xmm0, SUBSH_MASK1_2WAY );\
+    xmm1 = _mm256_shuffle_epi8( xmm1, SUBSH_MASK3_2WAY );\
+    xmm2 = _mm256_shuffle_epi8( xmm2, SUBSH_MASK5_2WAY );\
+    xmm3 = _mm256_shuffle_epi8( xmm3, SUBSH_MASK7_2WAY );\
+    xmm4 = _mm256_shuffle_epi8( xmm4, SUBSH_MASK0_2WAY );\
+    xmm5 = _mm256_shuffle_epi8( xmm5, SUBSH_MASK2_2WAY );\
+    xmm6 = _mm256_shuffle_epi8( xmm6, SUBSH_MASK4_2WAY );\
+    xmm7 = _mm256_shuffle_epi8( xmm7, SUBSH_MASK6_2WAY );\
+    /* SubBytes + MixBytes */\
+    SUBMIX_2WAY(xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15);\
+  }\
+}
+
+#define Matrix_Transpose_2way(i0, i1, i2, i3, i4, i5, i6, i7, t0, t1, t2, t3, t4, t5, t6, t7){\
+  t0 = TRANSP_MASK_2WAY;\
+\
+  i6 = _mm256_shuffle_epi8(i6, t0);\
+  i0 = _mm256_shuffle_epi8(i0, t0);\
+  i1 = _mm256_shuffle_epi8(i1, t0);\
+  i2 = _mm256_shuffle_epi8(i2, t0);\
+  i3 = _mm256_shuffle_epi8(i3, t0);\
+  t1 = i2;\
+  i4 = _mm256_shuffle_epi8(i4, t0);\
+  i5 = _mm256_shuffle_epi8(i5, t0);\
+  t2 = i4;\
+  t3 = i6;\
+  i7 = _mm256_shuffle_epi8(i7, t0);\
+\
+  /* continue with unpack using 4 temp registers */\
+  t0 = i0;\
+  t2 = _mm256_unpackhi_epi16(t2, i5);\
+  i4 = _mm256_unpacklo_epi16(i4, i5);\
+  t3 = _mm256_unpackhi_epi16(t3, i7);\
+  i6 = _mm256_unpacklo_epi16(i6, i7);\
+  t0 = _mm256_unpackhi_epi16(t0, i1);\
+  t1 = _mm256_unpackhi_epi16(t1, i3);\
+  i2 = _mm256_unpacklo_epi16(i2, i3);\
+  i0 = _mm256_unpacklo_epi16(i0, i1);\
+\
+  /* shuffle with immediate */\
+  t0 = _mm256_shuffle_epi32(t0, 216);\
+  t1 = _mm256_shuffle_epi32(t1, 216);\
+  t2 = _mm256_shuffle_epi32(t2, 216);\
+  t3 = _mm256_shuffle_epi32(t3, 216);\
+  i0 = _mm256_shuffle_epi32(i0, 216);\
+  i2 = _mm256_shuffle_epi32(i2, 216);\
+  i4 = _mm256_shuffle_epi32(i4, 216);\
+  i6 = _mm256_shuffle_epi32(i6, 216);\
+\
+  /* continue with unpack */\
+  t4 = i0;\
+  i0 = _mm256_unpacklo_epi32(i0, i2);\
+  t4 = _mm256_unpackhi_epi32(t4, i2);\
+  t5 = t0;\
+  t0 = _mm256_unpacklo_epi32(t0, t1);\
+  t5 = _mm256_unpackhi_epi32(t5, t1);\
+  t6 = i4;\
+  i4 = _mm256_unpacklo_epi32(i4, i6);\
+  t7 = t2;\
+  t6 = _mm256_unpackhi_epi32(t6, i6);\
+  i2 = t0;\
+  t2 = _mm256_unpacklo_epi32(t2, t3);\
+  i3 = t0;\
+  t7 = _mm256_unpackhi_epi32(t7, t3);\
+\
+  /* there are now 2 rows in each xmm */\
+  /* unpack to get 1 row of CV in each xmm */\
+  i1 = i0;\
+  i1 = _mm256_unpackhi_epi64(i1, i4);\
+  i0 = _mm256_unpacklo_epi64(i0, i4);\
+  i4 = t4;\
+  i3 = _mm256_unpackhi_epi64(i3, t2);\
+  i5 = t4;\
+  i2 = _mm256_unpacklo_epi64(i2, t2);\
+  i6 = t5;\
+  i5 = _mm256_unpackhi_epi64(i5, t6);\
+  i7 = t5;\
+  i4 = _mm256_unpacklo_epi64(i4, t6);\
+  i7 = _mm256_unpackhi_epi64(i7, t7);\
+  i6 = _mm256_unpacklo_epi64(i6, t7);\
+  /* transpose done */\
+}/**/
+
+#define Matrix_Transpose_INV_2way(i0, i1, i2, i3, i4, i5, i6, i7, o0, o1, o2, t0, t1, t2, t3, t4){\
+  /*  transpose matrix to get output format */\
+  o1 = i0;\
+  i0 = _mm256_unpacklo_epi64(i0, i1);\
+  o1 = _mm256_unpackhi_epi64(o1, i1);\
+  t0 = i2;\
+  i2 = _mm256_unpacklo_epi64(i2, i3);\
+  t0 = _mm256_unpackhi_epi64(t0, i3);\
+  t1 = i4;\
+  i4 = _mm256_unpacklo_epi64(i4, i5);\
+  t1 = _mm256_unpackhi_epi64(t1, i5);\
+  t2 = i6;\
+  o0 = TRANSP_MASK_2WAY;\
+  i6 = _mm256_unpacklo_epi64(i6, i7);\
+  t2 = _mm256_unpackhi_epi64(t2, i7);\
+  /* load transpose mask into a register, because it will be used 8 times */\
+  i0 = _mm256_shuffle_epi8(i0, o0);\
+  i2 = _mm256_shuffle_epi8(i2, o0);\
+  i4 = _mm256_shuffle_epi8(i4, o0);\
+  i6 = _mm256_shuffle_epi8(i6, o0);\
+  o1 = _mm256_shuffle_epi8(o1, o0);\
+  t0 = _mm256_shuffle_epi8(t0, o0);\
+  t1 = _mm256_shuffle_epi8(t1, o0);\
+  t2 = _mm256_shuffle_epi8(t2, o0);\
+  /* continue with unpack using 4 temp registers */\
+  t3 = i4;\
+  o2 = o1;\
+  o0 = i0;\
+  t4 = t1;\
+  \
+  t3 = _mm256_unpackhi_epi16(t3, i6);\
+  i4 = _mm256_unpacklo_epi16(i4, i6);\
+  o0 = _mm256_unpackhi_epi16(o0, i2);\
+  i0 = _mm256_unpacklo_epi16(i0, i2);\
+  o2 = _mm256_unpackhi_epi16(o2, t0);\
+  o1 = _mm256_unpacklo_epi16(o1, t0);\
+  t4 = _mm256_unpackhi_epi16(t4, t2);\
+  t1 = _mm256_unpacklo_epi16(t1, t2);\
+  /* shuffle with immediate */\
+  i4 = _mm256_shuffle_epi32(i4, 216);\
+  t3 = _mm256_shuffle_epi32(t3, 216);\
+  o1 = _mm256_shuffle_epi32(o1, 216);\
+  o2 = _mm256_shuffle_epi32(o2, 216);\
+  i0 = _mm256_shuffle_epi32(i0, 216);\
+  o0 = _mm256_shuffle_epi32(o0, 216);\
+  t1 = _mm256_shuffle_epi32(t1, 216);\
+  t4 = _mm256_shuffle_epi32(t4, 216);\
+  /* continue with unpack */\
+  i1 = i0;\
+  i3 = o0;\
+  i5 = o1;\
+  i7 = o2;\
+  i0 = _mm256_unpacklo_epi32(i0, i4);\
+  i1 = _mm256_unpackhi_epi32(i1, i4);\
+  o0 = _mm256_unpacklo_epi32(o0, t3);\
+  i3 = _mm256_unpackhi_epi32(i3, t3);\
+  o1 = _mm256_unpacklo_epi32(o1, t1);\
+  i5 = _mm256_unpackhi_epi32(i5, t1);\
+  o2 = _mm256_unpacklo_epi32(o2, t4);\
+  i7 = _mm256_unpackhi_epi32(i7, t4);\
+  /* transpose done */\
+}/**/
+
+void INIT_2way( __m256i *chaining )
+{
+  static __m256i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7;
+  static __m256i xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15;
+
+  /* load IV into registers xmm8 - xmm15 */
+  xmm8 = chaining[0];
+  xmm9 = chaining[1];
+  xmm10 = chaining[2];
+  xmm11 = chaining[3];
+  xmm12 = chaining[4];
+  xmm13 = chaining[5];
+  xmm14 = chaining[6];
+  xmm15 = chaining[7];
+
+  /* transform chaining value from column ordering into row ordering */
+  Matrix_Transpose_2way(xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);
+
+  /* store transposed IV */
+  chaining[0] = xmm8;
+  chaining[1] = xmm9;
+  chaining[2] = xmm10;
+  chaining[3] = xmm11;
+  chaining[4] = xmm12;
+  chaining[5] = xmm13;
+  chaining[6] = xmm14;
+  chaining[7] = xmm15;
+}
+
+void TF1024_2way( __m256i *chaining, const __m256i *message )
+{
+  static __m256i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7;
+  static __m256i xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15;
+  static __m256i QTEMP[8];
+  static __m256i TEMP0;
+  static __m256i TEMP1;
+  static __m256i TEMP2;
+
+  /* load message into registers xmm8 - xmm15 (Q = message) */
+  xmm8 = message[0];
+  xmm9 = message[1];
+  xmm10 = message[2];
+  xmm11 = message[3];
+  xmm12 = message[4];
+  xmm13 = message[5];
+  xmm14 = message[6];
+  xmm15 = message[7];
+
+  /* transform message M from column ordering into row ordering */
+  Matrix_Transpose_2way(xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7);
+
+  /* store message M (Q input) for later */
+  QTEMP[0] = xmm8;
+  QTEMP[1] = xmm9;
+  QTEMP[2] = xmm10;
+  QTEMP[3] = xmm11;
+  QTEMP[4] = xmm12;
+  QTEMP[5] = xmm13;
+  QTEMP[6] = xmm14;
+  QTEMP[7] = xmm15;
+
+  /* xor CV to message to get P input */
+  /* result: CV+M in xmm8...xmm15 */
+  xmm8 = _mm256_xor_si256( xmm8,  (chaining[0]) );
+  xmm9 = _mm256_xor_si256( xmm9,  (chaining[1]) );
+  xmm10 = _mm256_xor_si256( xmm10, (chaining[2]) );
+  xmm11 = _mm256_xor_si256( xmm11, (chaining[3]) );
+  xmm12 = _mm256_xor_si256( xmm12, (chaining[4]) );
+  xmm13 = _mm256_xor_si256( xmm13, (chaining[5]) );
+  xmm14 = _mm256_xor_si256( xmm14, (chaining[6]) );
+  xmm15 = _mm256_xor_si256( xmm15, (chaining[7]) );
+
+  /* compute permutation P */
+  /* result: P(CV+M) in xmm8...xmm15 */
+  ROUNDS_P_2WAY();
+
+  /* xor CV to P output (feed-forward) */
+  /* result: P(CV+M)+CV in xmm8...xmm15 */
+  xmm8 = _mm256_xor_si256( xmm8,  (chaining[0]) );
+  xmm9 = _mm256_xor_si256( xmm9,  (chaining[1]) );
+  xmm10 = _mm256_xor_si256( xmm10, (chaining[2]) );
+  xmm11 = _mm256_xor_si256( xmm11, (chaining[3]) );
+  xmm12 = _mm256_xor_si256( xmm12, (chaining[4]) );
+  xmm13 = _mm256_xor_si256( xmm13, (chaining[5]) );
+  xmm14 = _mm256_xor_si256( xmm14, (chaining[6]) );
+  xmm15 = _mm256_xor_si256( xmm15, (chaining[7]) );
+
+  /* store P(CV+M)+CV */
+  chaining[0] = xmm8;
+  chaining[1] = xmm9;
+  chaining[2] = xmm10;
+  chaining[3] = xmm11;
+  chaining[4] = xmm12;
+  chaining[5] = xmm13;
+  chaining[6] = xmm14;
+  chaining[7] = xmm15;
+
+  /* load message M (Q input) into xmm8-15 */
+  xmm8 = QTEMP[0];
+  xmm9 = QTEMP[1];
+  xmm10 = QTEMP[2];
+  xmm11 = QTEMP[3];
+  xmm12 = QTEMP[4];
+  xmm13 = QTEMP[5];
+  xmm14 = QTEMP[6];
+  xmm15 = QTEMP[7];
+
+  /* compute permutation Q */
+  /* result: Q(M) in xmm8...xmm15 */
+  ROUNDS_Q_2WAY();
+
+  /* xor Q output */
+  /* result: P(CV+M)+CV+Q(M) in xmm8...xmm15 */
+  xmm8 = _mm256_xor_si256( xmm8,  (chaining[0]) );
+  xmm9 = _mm256_xor_si256( xmm9,  (chaining[1]) );
+  xmm10 = _mm256_xor_si256( xmm10, (chaining[2]) );
+  xmm11 = _mm256_xor_si256( xmm11, (chaining[3]) );
+  xmm12 = _mm256_xor_si256( xmm12, (chaining[4]) );
+  xmm13 = _mm256_xor_si256( xmm13, (chaining[5]) );
+  xmm14 = _mm256_xor_si256( xmm14, (chaining[6]) );
+  xmm15 = _mm256_xor_si256( xmm15, (chaining[7]) );
+
+  /* store CV */
+  chaining[0] = xmm8;
+  chaining[1] = xmm9;
+  chaining[2] = xmm10;
+  chaining[3] = xmm11;
+  chaining[4] = xmm12;
+  chaining[5] = xmm13;
+  chaining[6] = xmm14;
+  chaining[7] = xmm15;
+
+  return;
+}
+
+void OF1024_2way( __m256i* chaining )
+{
+  static __m256i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7;
+  static __m256i xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15;
+  static __m256i TEMP0;
+  static __m256i TEMP1;
+  static __m256i TEMP2;
+
+  /* load CV into registers xmm8 - xmm15 */
+  xmm8 = chaining[0];
+  xmm9 = chaining[1];
+  xmm10 = chaining[2];
+  xmm11 = chaining[3];
+  xmm12 = chaining[4];
+  xmm13 = chaining[5];
+  xmm14 = chaining[6];
+  xmm15 = chaining[7];
+
+  /* compute permutation P */
+  /* result: P(CV) in xmm8...xmm15 */
+  ROUNDS_P_2WAY();
+
+  /* xor CV to P output (feed-forward) */
+  /* result: P(CV)+CV in xmm8...xmm15 */
+  xmm8 = _mm256_xor_si256( xmm8,  (chaining[0]) );
+  xmm9 = _mm256_xor_si256( xmm9,  (chaining[1]) );
+  xmm10 = _mm256_xor_si256( xmm10, (chaining[2]) );
+  xmm11 = _mm256_xor_si256( xmm11, (chaining[3]) );
+  xmm12 = _mm256_xor_si256( xmm12, (chaining[4]) );
+  xmm13 = _mm256_xor_si256( xmm13, (chaining[5]) );
+  xmm14 = _mm256_xor_si256( xmm14, (chaining[6]) );
+  xmm15 = _mm256_xor_si256( xmm15, (chaining[7]) );
+
+  /* transpose CV back from row ordering to column ordering */
+  /* result: final hash value in xmm0, xmm6, xmm13, xmm15 */
+  Matrix_Transpose_INV_2way(xmm8, xmm9, xmm10, xmm11, xmm12, xmm13, xmm14, xmm15, xmm4, xmm0, xmm6, xmm1, xmm2, xmm3, xmm5, xmm7);
+
+  /* we only need to return the truncated half of the state */
+  chaining[4] = xmm0;
+  chaining[5] = xmm6;
+  chaining[6] = xmm13;
+  chaining[7] = xmm15;
+
+  return;
+}
+
+
+
 #endif  // VAES
 #endif  // GROESTL512_INTR_4WAY_H__
--- a/algo/groestl/myr-groestl.c
+++ b/algo/groestl/myr-groestl.c
@@ -11,7 +11,7 @@
 #else
  #include "sph_groestl.h"
 #endif
-#include <openssl/sha.h>
+#include "algo/sha/sha256-hash.h"

 typedef struct {
 #ifdef __AES__
@@ -19,7 +19,6 @@ typedef struct {
 #else
    sph_groestl512_context  groestl;
 #endif
-    SHA256_CTX              sha;
 } myrgr_ctx_holder;

 myrgr_ctx_holder myrgr_ctx;
@@ -31,7 +30,6 @@ void init_myrgr_ctx()
 #else
     sph_groestl512_init( &myrgr_ctx.groestl );
 #endif
-     SHA256_Init( &myrgr_ctx.sha );
 }

 void myriad_hash(void *output, const void *input)
@@ -39,54 +37,54 @@ void myriad_hash(void *output, const void *input)
   myrgr_ctx_holder ctx;
   memcpy( &ctx, &myrgr_ctx, sizeof(myrgr_ctx) );

- 	uint32_t _ALIGN(32) hash[16];
+   uint32_t _ALIGN(32) hash[16];

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

-   SHA256_Update( &ctx.sha, (unsigned char*)hash, 64 );
-   SHA256_Final( (unsigned char*)hash, &ctx.sha );
+   sha256_full( hash, hash, 64 );

-	memcpy(output, hash, 32);
+   memcpy(output, hash, 32);
 }

 int scanhash_myriad( struct work *work, uint32_t max_nonce,
                     uint64_t *hashes_done, struct thr_info *mythr )
 {
-	uint32_t _ALIGN(64) endiandata[20];
+   uint32_t _ALIGN(64) endiandata[20];
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
-	const uint32_t first_nonce = pdata[19];
-	uint32_t nonce = first_nonce;
-   int thr_id = mythr->id;  // thr_id arg is deprecated
+   const uint32_t first_nonce = pdata[19];
+   uint32_t nonce = first_nonce;
+   int thr_id = mythr->id;

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

-        swab32_array( endiandata, pdata, 20 );
+   swab32_array( endiandata, pdata, 20 );

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

-		if (hash[7] <= Htarg && fulltest(hash, ptarget)) {
-			pdata[19] = nonce;
-			*hashes_done = pdata[19] - first_nonce;
-			return 1;
-		}
-		nonce++;
+      if (hash[7] <= Htarg && fulltest(hash, ptarget))
+      {
+         pdata[19] = nonce;
+         *hashes_done = pdata[19] - first_nonce;
+         return 1;
+      }
+      nonce++;

-	} while (nonce < max_nonce && !work_restart[thr_id].restart);
+   } while (nonce < max_nonce && !work_restart[thr_id].restart);

-	pdata[19] = nonce;
-	*hashes_done = pdata[19] - first_nonce + 1;
-	return 0;
+   pdata[19] = nonce;
+   *hashes_done = pdata[19] - first_nonce + 1;
+   return 0;
 }
 #endif
--- a/algo/groestl/myrgr-4way.c
+++ b/algo/groestl/myrgr-4way.c
@@ -44,6 +44,7 @@ void myriad_8way_hash( void *output, const void *input )

     rintrlv_8x64_4x128( vhashA, vhashB, input, 640 );
     groestl512_4way_update_close( &ctx.groestl, vhashA, vhashA, 640 );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(groestl512_4way_context) );
     groestl512_4way_update_close( &ctx.groestl, vhashB, vhashB, 640 );

     uint32_t hash0[20] __attribute__ ((aligned (64)));
@@ -58,8 +59,6 @@ void myriad_8way_hash( void *output, const void *input )
 //     rintrlv_4x128_8x32( vhash, vhashA, vhashB, 512 );
     dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhashA );
     dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhashB );
-     intrlv_8x32_512( vhash, hash0, hash1, hash2, hash3, hash4, hash5,
-                       hash6, hash7 );

 #else

@@ -76,27 +75,27 @@ void myriad_8way_hash( void *output, const void *input )
                   hash4, hash5, hash6, hash7, input, 640 );

     update_and_final_groestl( &ctx.groestl, (char*)hash0, (char*)hash0, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
     update_and_final_groestl( &ctx.groestl, (char*)hash1, (char*)hash1, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
     update_and_final_groestl( &ctx.groestl, (char*)hash2, (char*)hash2, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
     update_and_final_groestl( &ctx.groestl, (char*)hash3, (char*)hash3, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
     update_and_final_groestl( &ctx.groestl, (char*)hash4, (char*)hash4, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
     update_and_final_groestl( &ctx.groestl, (char*)hash5, (char*)hash5, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
     update_and_final_groestl( &ctx.groestl, (char*)hash6, (char*)hash6, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );
     update_and_final_groestl( &ctx.groestl, (char*)hash7, (char*)hash7, 640 );
-     memcpy( &ctx.groestl, &myrgr_4way_ctx.groestl, sizeof(hashState_groestl) );
-
-     intrlv_8x32( vhash, hash0, hash1, hash2, hash3,
-                         hash4, hash5, hash6, hash7, 512 );
+     memcpy( &ctx.groestl, &myrgr_8way_ctx.groestl, sizeof(hashState_groestl) );

 #endif

+     intrlv_8x32_512( vhash, hash0, hash1, hash2, hash3, hash4, hash5,
+                       hash6, hash7 );
+     
     sha256_8way_update( &ctx.sha, vhash, 64 );
     sha256_8way_close( &ctx.sha, output );
 }
--- a/algo/hamsi/hamsi-hash-4way.c
+++ b/algo/hamsi/hamsi-hash-4way.c
@@ -548,7 +548,7 @@ static const sph_u32 T512[64][16] = {

 #if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

-// Hamsi 8 way 
+// Hamsi 8 way AVX512 

 #define INPUT_BIG8 \
 do { \
@@ -560,22 +560,14 @@ do { \
     __m512i dm = _mm512_and_si512( db, m512_one_64 ) ; \
     dm = mm512_negate_32( _mm512_or_si512( dm, \
                                          _mm512_slli_epi64( dm, 32 ) ) ); \
-     m0 = _mm512_xor_si512( m0, _mm512_and_si512( dm, \
-                                          m512_const1_64( tp[0] ) ) ); \
-     m1 = _mm512_xor_si512( m1, _mm512_and_si512( dm, \
-                                          m512_const1_64( tp[1] ) ) ); \
-     m2 = _mm512_xor_si512( m2, _mm512_and_si512( dm, \
-                                          m512_const1_64( tp[2] ) ) ); \
-     m3 = _mm512_xor_si512( m3, _mm512_and_si512( dm, \
-                                          m512_const1_64( tp[3] ) ) ); \
-     m4 = _mm512_xor_si512( m4, _mm512_and_si512( dm, \
-                                          m512_const1_64( tp[4] ) ) ); \
-     m5 = _mm512_xor_si512( m5, _mm512_and_si512( dm, \
-                                          m512_const1_64( tp[5] ) ) ); \
-     m6 = _mm512_xor_si512( m6, _mm512_and_si512( dm, \
-                                          m512_const1_64( tp[6] ) ) ); \
-     m7 = _mm512_xor_si512( m7, _mm512_and_si512( dm, \
-                                          m512_const1_64( tp[7] ) ) ); \
+     m0 = mm512_xorand( m0, dm, m512_const1_64( tp[0] ) ); \
+     m1 = mm512_xorand( m1, dm, m512_const1_64( tp[1] ) ); \
+     m2 = mm512_xorand( m2, dm, m512_const1_64( tp[2] ) ); \
+     m3 = mm512_xorand( m3, dm, m512_const1_64( tp[3] ) ); \
+     m4 = mm512_xorand( m4, dm, m512_const1_64( tp[4] ) ); \
+     m5 = mm512_xorand( m5, dm, m512_const1_64( tp[5] ) ); \
+     m6 = mm512_xorand( m6, dm, m512_const1_64( tp[6] ) ); \
+     m7 = mm512_xorand( m7, dm, m512_const1_64( tp[7] ) ); \
     tp += 8; \
     db = _mm512_srli_epi64( db, 1 ); \
  } \
@@ -585,20 +577,13 @@ do { \
 do { \
  __m512i t; \
  t = a; \
-  a = _mm512_and_si512( a, c ); \
-  a = _mm512_xor_si512( a, d ); \
-  c = _mm512_xor_si512( c, b ); \
-  c = _mm512_xor_si512( c, a ); \
-  d = _mm512_or_si512( d, t ); \
-  d = _mm512_xor_si512( d, b ); \
+  a = mm512_xorand( d, a, c ); \
+  c = mm512_xor3( a, b, c ); \
+  b = mm512_xoror( b, d, t ); \
  t = _mm512_xor_si512( t, c ); \
-  b = d; \
-  d = _mm512_or_si512( d, t ); \
-  d = _mm512_xor_si512( d, a ); \
-  a = _mm512_and_si512( a, b ); \
-  t = _mm512_xor_si512( t, a ); \
-  b = _mm512_xor_si512( b, d ); \
-  b = _mm512_xor_si512( b, t ); \
+  d = mm512_xoror( a, b, t ); \
+  t = mm512_xorand( t, a, b ); \
+  b = mm512_xor3( b, d, t ); \
  a = c; \
  c = b; \
  b = d; \
@@ -609,14 +594,12 @@ do { \
 do { \
   a = mm512_rol_32( a, 13 ); \
   c = mm512_rol_32( c,  3 ); \
-   b = _mm512_xor_si512( b, _mm512_xor_si512( a, c ) ); \
-   d = _mm512_xor_si512( d, _mm512_xor_si512( c, \
-                                              _mm512_slli_epi32( a, 3 ) ) ); \
+   b = mm512_xor3( a, b, c ); \
+   d = mm512_xor3( d, c, _mm512_slli_epi32( a, 3 ) ); \
   b = mm512_rol_32( b, 1 ); \
   d = mm512_rol_32( d, 7 ); \
-   a = _mm512_xor_si512( a, _mm512_xor_si512( b, d ) ); \
-   c = _mm512_xor_si512( c, _mm512_xor_si512( d, \
-                                              _mm512_slli_epi32( b, 7 ) ) ); \
+   a = mm512_xor3( a, b, d ); \
+   c = mm512_xor3( c, d, _mm512_slli_epi32( b, 7 ) ); \
   a = mm512_rol_32( a,  5 ); \
   c = mm512_rol_32( c, 22 ); \
 } while (0)
@@ -649,26 +632,25 @@ do { \
 } while (0)


-#define ROUND_BIG8(rc, alpha) \
+#define ROUND_BIG8( alpha ) \
 do { \
   __m512i t0, t1, t2, t3; \
-   s0 = _mm512_xor_si512( s0, m512_const1_64( \
-                   ( (uint64_t)(rc) << 32 ) ^ ( (uint64_t*)(alpha) )[ 0] ) ); \
-   s1 = _mm512_xor_si512( s1, m512_const1_64( ( (uint64_t*)(alpha) )[ 1] ) ); \
-   s2 = _mm512_xor_si512( s2, m512_const1_64( ( (uint64_t*)(alpha) )[ 2] ) ); \
-   s3 = _mm512_xor_si512( s3, m512_const1_64( ( (uint64_t*)(alpha) )[ 3] ) ); \
-   s4 = _mm512_xor_si512( s4, m512_const1_64( ( (uint64_t*)(alpha) )[ 4] ) ); \
-   s5 = _mm512_xor_si512( s5, m512_const1_64( ( (uint64_t*)(alpha) )[ 5] ) ); \
-   s6 = _mm512_xor_si512( s6, m512_const1_64( ( (uint64_t*)(alpha) )[ 6] ) ); \
-   s7 = _mm512_xor_si512( s7, m512_const1_64( ( (uint64_t*)(alpha) )[ 7] ) ); \
-   s8 = _mm512_xor_si512( s8, m512_const1_64( ( (uint64_t*)(alpha) )[ 8] ) ); \
-   s9 = _mm512_xor_si512( s9, m512_const1_64( ( (uint64_t*)(alpha) )[ 9] ) ); \
-   sA = _mm512_xor_si512( sA, m512_const1_64( ( (uint64_t*)(alpha) )[10] ) ); \
-   sB = _mm512_xor_si512( sB, m512_const1_64( ( (uint64_t*)(alpha) )[11] ) ); \
-   sC = _mm512_xor_si512( sC, m512_const1_64( ( (uint64_t*)(alpha) )[12] ) ); \
-   sD = _mm512_xor_si512( sD, m512_const1_64( ( (uint64_t*)(alpha) )[13] ) ); \
-   sE = _mm512_xor_si512( sE, m512_const1_64( ( (uint64_t*)(alpha) )[14] ) ); \
-   sF = _mm512_xor_si512( sF, m512_const1_64( ( (uint64_t*)(alpha) )[15] ) ); \
+   s0 = _mm512_xor_si512( s0, alpha[ 0] ); \
+   s1 = _mm512_xor_si512( s1, alpha[ 1] ); \
+   s2 = _mm512_xor_si512( s2, alpha[ 2] ); \
+   s3 = _mm512_xor_si512( s3, alpha[ 3] ); \
+   s4 = _mm512_xor_si512( s4, alpha[ 4] ); \
+   s5 = _mm512_xor_si512( s5, alpha[ 5] ); \
+   s6 = _mm512_xor_si512( s6, alpha[ 6] ); \
+   s7 = _mm512_xor_si512( s7, alpha[ 7] ); \
+   s8 = _mm512_xor_si512( s8, alpha[ 8] ); \
+   s9 = _mm512_xor_si512( s9, alpha[ 9] ); \
+   sA = _mm512_xor_si512( sA, alpha[10] ); \
+   sB = _mm512_xor_si512( sB, alpha[11] ); \
+   sC = _mm512_xor_si512( sC, alpha[12] ); \
+   sD = _mm512_xor_si512( sD, alpha[13] ); \
+   sE = _mm512_xor_si512( sE, alpha[14] ); \
+   sF = _mm512_xor_si512( sF, alpha[15] ); \
 \
  SBOX8( s0, s4, s8, sC ); \
  SBOX8( s1, s5, s9, sD ); \
@@ -748,28 +730,66 @@ do { \

 #define P_BIG8 \
 do { \
-   ROUND_BIG8(0, alpha_n); \
-   ROUND_BIG8(1, alpha_n); \
-   ROUND_BIG8(2, alpha_n); \
-   ROUND_BIG8(3, alpha_n); \
-   ROUND_BIG8(4, alpha_n); \
-   ROUND_BIG8(5, alpha_n); \
+   __m512i alpha[16]; \
+   for( int i = 0; i < 16; i++ ) \
+      alpha[i] = m512_const1_64( ( (uint64_t*)alpha_n )[i] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)1 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)2 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)3 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)4 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)5 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG8( alpha ); \
 } while (0)

 #define PF_BIG8 \
 do { \
-   ROUND_BIG8( 0, alpha_f); \
-   ROUND_BIG8( 1, alpha_f); \
-   ROUND_BIG8( 2, alpha_f); \
-   ROUND_BIG8( 3, alpha_f); \
-   ROUND_BIG8( 4, alpha_f); \
-   ROUND_BIG8( 5, alpha_f); \
-   ROUND_BIG8( 6, alpha_f); \
-   ROUND_BIG8( 7, alpha_f); \
-   ROUND_BIG8( 8, alpha_f); \
-   ROUND_BIG8( 9, alpha_f); \
-   ROUND_BIG8(10, alpha_f); \
-   ROUND_BIG8(11, alpha_f); \
+   __m512i alpha[16]; \
+   for( int i = 0; i < 16; i++ ) \
+      alpha[i] = m512_const1_64( ( (uint64_t*)alpha_f )[i] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)1 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)2 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)3 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)4 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)5 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)6 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)7 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)8 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)9 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)10 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
+   alpha[0] = m512_const1_64( ( (uint64_t)11 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG8( alpha ); \
 } while (0)

 #define T_BIG8 \
@@ -849,13 +869,11 @@ void hamsi512_8way_update( hamsi_8way_big_context *sc, const void *data,
 void hamsi512_8way_close( hamsi_8way_big_context *sc, void *dst )
 {
   __m512i pad[1];
-   int ch, cl;
+   uint32_t ch, cl;

   sph_enc32be( &ch, sc->count_high );
   sph_enc32be( &cl, sc->count_low + ( sc->partial_len << 3 ) );
-   pad[0] =  _mm512_set_epi32( cl, ch, cl, ch, cl, ch, cl, ch,
-                               cl, ch, cl, ch, cl, ch, cl, ch );
-//   pad[0] =  m512_const2_32( cl, ch );
+   pad[0] = _mm512_set1_epi64( ((uint64_t)cl << 32 ) | (uint64_t)ch );
   sc->buf[0] = m512_const1_64( 0x80 );
   hamsi_8way_big( sc, sc->buf, 1 );
   hamsi_8way_big_final( sc, pad );
@@ -863,11 +881,9 @@ void hamsi512_8way_close( hamsi_8way_big_context *sc, void *dst )
   mm512_block_bswap_32( (__m512i*)dst, sc->h );
 }

-
 #endif // AVX512

-
-// Hamsi 4 way
+// Hamsi 4 way AVX2

 #define INPUT_BIG \
 do { \
@@ -986,26 +1002,25 @@ do { \
 #define sF   m7
 */

-#define ROUND_BIG(rc, alpha) \
+#define ROUND_BIG( alpha ) \
 do { \
   __m256i t0, t1, t2, t3; \
-   s0 = _mm256_xor_si256( s0, m256_const1_64( \
-                   ( (uint64_t)(rc) << 32 ) ^ ( (uint64_t*)(alpha) )[ 0] ) ); \
-   s1 = _mm256_xor_si256( s1, m256_const1_64( ( (uint64_t*)(alpha) )[ 1] ) ); \
-   s2 = _mm256_xor_si256( s2, m256_const1_64( ( (uint64_t*)(alpha) )[ 2] ) ); \
-   s3 = _mm256_xor_si256( s3, m256_const1_64( ( (uint64_t*)(alpha) )[ 3] ) ); \
-   s4 = _mm256_xor_si256( s4, m256_const1_64( ( (uint64_t*)(alpha) )[ 4] ) ); \
-   s5 = _mm256_xor_si256( s5, m256_const1_64( ( (uint64_t*)(alpha) )[ 5] ) ); \
-   s6 = _mm256_xor_si256( s6, m256_const1_64( ( (uint64_t*)(alpha) )[ 6] ) ); \
-   s7 = _mm256_xor_si256( s7, m256_const1_64( ( (uint64_t*)(alpha) )[ 7] ) ); \
-   s8 = _mm256_xor_si256( s8, m256_const1_64( ( (uint64_t*)(alpha) )[ 8] ) ); \
-   s9 = _mm256_xor_si256( s9, m256_const1_64( ( (uint64_t*)(alpha) )[ 9] ) ); \
-   sA = _mm256_xor_si256( sA, m256_const1_64( ( (uint64_t*)(alpha) )[10] ) ); \
-   sB = _mm256_xor_si256( sB, m256_const1_64( ( (uint64_t*)(alpha) )[11] ) ); \
-   sC = _mm256_xor_si256( sC, m256_const1_64( ( (uint64_t*)(alpha) )[12] ) ); \
-   sD = _mm256_xor_si256( sD, m256_const1_64( ( (uint64_t*)(alpha) )[13] ) ); \
-   sE = _mm256_xor_si256( sE, m256_const1_64( ( (uint64_t*)(alpha) )[14] ) ); \
-   sF = _mm256_xor_si256( sF, m256_const1_64( ( (uint64_t*)(alpha) )[15] ) ); \
+   s0 = _mm256_xor_si256( s0, alpha[ 0] ); \
+   s1 = _mm256_xor_si256( s1, alpha[ 1] ); \
+   s2 = _mm256_xor_si256( s2, alpha[ 2] ); \
+   s3 = _mm256_xor_si256( s3, alpha[ 3] ); \
+   s4 = _mm256_xor_si256( s4, alpha[ 4] ); \
+   s5 = _mm256_xor_si256( s5, alpha[ 5] ); \
+   s6 = _mm256_xor_si256( s6, alpha[ 6] ); \
+   s7 = _mm256_xor_si256( s7, alpha[ 7] ); \
+   s8 = _mm256_xor_si256( s8, alpha[ 8] ); \
+   s9 = _mm256_xor_si256( s9, alpha[ 9] ); \
+   sA = _mm256_xor_si256( sA, alpha[10] ); \
+   sB = _mm256_xor_si256( sB, alpha[11] ); \
+   sC = _mm256_xor_si256( sC, alpha[12] ); \
+   sD = _mm256_xor_si256( sD, alpha[13] ); \
+   sE = _mm256_xor_si256( sE, alpha[14] ); \
+   sF = _mm256_xor_si256( sF, alpha[15] ); \
 \
  SBOX( s0, s4, s8, sC ); \
  SBOX( s1, s5, s9, sD ); \
@@ -1085,28 +1100,66 @@ do { \

 #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); \
+   __m256i alpha[16]; \
+   for( int i = 0; i < 16; i++ ) \
+      alpha[i] = m256_const1_64( ( (uint64_t*)alpha_n )[i] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)1 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)2 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)3 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)4 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)5 << 32 ) \
+                            ^ ( (uint64_t*)alpha_n )[0] ); \
+   ROUND_BIG( alpha ); \
 } 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); \
+   __m256i alpha[16]; \
+   for( int i = 0; i < 16; i++ ) \
+      alpha[i] = m256_const1_64( ( (uint64_t*)alpha_f )[i] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)1 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)2 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)3 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)4 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)5 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)6 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)7 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)8 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)9 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)10 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
+   alpha[0] = m256_const1_64( ( (uint64_t)11 << 32 ) \
+                            ^ ( (uint64_t*)alpha_f )[0] ); \
+   ROUND_BIG( alpha ); \
 } while (0)

 #define T_BIG \
@@ -1186,14 +1239,12 @@ void hamsi512_4way_update( hamsi_4way_big_context *sc, const void *data,
 void hamsi512_4way_close( hamsi_4way_big_context *sc, void *dst )
 {
   __m256i pad[1];
-   int ch, cl;
+   uint32_t 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 );
+   pad[0] = _mm256_set1_epi64x( ((uint64_t)cl << 32 ) | (uint64_t)ch );
   sc->buf[0] = m256_const1_64( 0x80 );
-//      sc->buf[0] = _mm256_set_epi32( 0UL, 0x80UL, 0UL, 0x80UL,
-//                                  0UL, 0x80UL, 0UL, 0x80UL );
   hamsi_big( sc, sc->buf, 1 );
   hamsi_big_final( sc, pad );

--- a/algo/haval/haval-hash-4way.c
+++ b/algo/haval/haval-hash-4way.c
@@ -522,50 +522,53 @@ do { \

 // Haval-256 8 way 32 bit avx2

+#if defined (__AVX512VL__)
+
+// ( ~( a ^ b ) ) & c
+#define mm256_andnotxor( a, b, c ) \
+   _mm256_ternarylogic_epi32( a, b, c, 0x82  )
+
+#else
+
+#define mm256_andnotxor( a, b, c ) \
+   _mm256_andnot_si256( _mm256_xor_si256( a, b ), c )
+
+#endif
+
 #define F1_8W(x6, x5, x4, x3, x2, x1, x0) \
-   _mm256_xor_si256( x0, \
-       _mm256_xor_si256( _mm256_and_si256(_mm256_xor_si256( x0, x4 ), x1 ), \
-                      _mm256_xor_si256( _mm256_and_si256( x2, x5 ), \
-                                     _mm256_and_si256( x3, x6 ) ) ) ) \
+ mm256_xor3( x0, mm256_andxor( x1, x0, x4 ), \
+                 _mm256_xor_si256( _mm256_and_si256( x2, x5 ), \
+                                   _mm256_and_si256( x3, x6 ) ) ) \

 #define F2_8W(x6, x5, x4, x3, x2, x1, x0) \
-   _mm256_xor_si256( \
-      _mm256_and_si256( x2, \
-         _mm256_xor_si256( _mm256_andnot_si256( x3, x1 ), \
-                        _mm256_xor_si256( _mm256_and_si256( x4, x5 ), \
-                                       _mm256_xor_si256( x6, x0 ) ) ) ), \
-         _mm256_xor_si256( \
-             _mm256_and_si256( x4, _mm256_xor_si256( x1, x5 ) ), \
-             _mm256_xor_si256( _mm256_and_si256( x3, x5 ), x0 ) ) ) \
+   mm256_xor3( mm256_andxor( x2, _mm256_andnot_si256( x3, x1 ), \
+                       mm256_xor3( _mm256_and_si256( x4, x5 ), x6, x0 )  ), \
+               mm256_andxor( x4, x1, x5 ), \
+               mm256_xorand( x0, x3, x5 ) ) \

 #define F3_8W(x6, x5, x4, x3, x2, x1, x0) \
-  _mm256_xor_si256( \
-    _mm256_and_si256( x3, \
-      _mm256_xor_si256( _mm256_and_si256( x1, x2 ), \
-                     _mm256_xor_si256( x6, x0 ) ) ), \
-      _mm256_xor_si256( _mm256_xor_si256(_mm256_and_si256( x1, x4 ), \
-                                   _mm256_and_si256( x2, x5 ) ), x0 ) )
+  mm256_xor3( x0, \
+              _mm256_and_si256( x3, \
+                         mm256_xor3( _mm256_and_si256( x1, x2 ), x6, x0 ) ), \
+              _mm256_xor_si256( _mm256_and_si256( x1, x4 ), \
+                                _mm256_and_si256( x2, x5 ) ) )

 #define F4_8W(x6, x5, x4, x3, x2, x1, x0) \
-  _mm256_xor_si256( \
-     _mm256_xor_si256( \
-        _mm256_and_si256( x3, \
-           _mm256_xor_si256( _mm256_xor_si256( _mm256_and_si256( x1, x2 ), \
-                                         _mm256_or_si256( x4, x6 ) ), x5 ) ), \
-        _mm256_and_si256( x4, \
-           _mm256_xor_si256( _mm256_xor_si256( _mm256_and_si256( mm256_not(x2), x5 ), \
-                          _mm256_xor_si256( x1, x6 ) ), x0 ) ) ), \
-     _mm256_xor_si256( _mm256_and_si256( x2, x6 ), x0 ) )
-
+  mm256_xor3( \
+      mm256_andxor( x3, x5, \
+                    _mm256_xor_si256( _mm256_and_si256( x1, x2 ), \
+                                      _mm256_or_si256( x4, x6 ) ) ), \
+      _mm256_and_si256( x4, \
+                        mm256_xor3( x0, _mm256_andnot_si256( x2, x5 ), \
+                                    _mm256_xor_si256( x1, x6 ) ) ), \
+      mm256_xorand( x0, x2, x6 ) )

 #define F5_8W(x6, x5, x4, x3, x2, x1, x0) \
   _mm256_xor_si256( \
-       _mm256_and_si256( x0, \
-            mm256_not( _mm256_xor_si256( \
-                    _mm256_and_si256( _mm256_and_si256( x1, x2 ), x3 ), x5 ) ) ), \
-      _mm256_xor_si256( _mm256_xor_si256( _mm256_and_si256( x1, x4 ), \
-                                    _mm256_and_si256( x2, x5 ) ), \
-                                    _mm256_and_si256( x3, x6 ) ) )
+         mm256_andnotxor( mm256_and3( x1, x2, x3 ), x5, x0 ), \
+         mm256_xor3( _mm256_and_si256( x1, x4 ), \
+                     _mm256_and_si256( x2, x5 ), \
+                     _mm256_and_si256( x3, x6 ) ) )

 #define FP3_1_8W(x6, x5, x4, x3, x2, x1, x0) \
   F1_8W(x1, x0, x3, x5, x6, x2, x4)
--- a/algo/hodl/hodl-gate.c
+++ b/algo/hodl/hodl-gate.c
@@ -99,13 +99,13 @@ void hodl_build_block_header( struct work* g_work, uint32_t version,
 // called only by thread 0, saves a backup of g_work
 void hodl_get_new_work( struct work* work, struct work* g_work)
 {
-   pthread_rwlock_rdlock( &g_work_lock );
+//   pthread_rwlock_rdlock( &g_work_lock );

   work_free( &hodl_work );
   work_copy( &hodl_work, g_work );
   hodl_work.data[ algo_gate.nonce_index ] = ( clock() + rand() ) % 9999;

-   pthread_rwlock_unlock( &g_work_lock );
+//   pthread_rwlock_unlock( &g_work_lock );
 }

 json_t *hodl_longpoll_rpc_call( CURL *curl, int *err, char* lp_url )
@@ -125,7 +125,7 @@ json_t *hodl_longpoll_rpc_call( CURL *curl, int *err, char* lp_url )
 }

 // called by every thread, copies the backup to each thread's work.
-void hodl_resync_threads( struct work* work )
+void hodl_resync_threads( int thr_id, struct work* work )
 {
   int nonce_index = algo_gate.nonce_index;
   pthread_barrier_wait( &hodl_barrier );
@@ -135,6 +135,7 @@ void hodl_resync_threads( struct work* work )
      work_copy( work, &hodl_work );
   }
   work->data[ nonce_index ] = swab32( hodl_work.data[ nonce_index ] );
+   work_restart[thr_id].restart = 0;
 }

 bool hodl_do_this_thread( int thr_id )
--- a/algo/hodl/hodl-wolf.c
+++ b/algo/hodl/hodl-wolf.c
@@ -7,6 +7,7 @@
 #include "hodl-gate.h"
 #include "hodl-wolf.h"
 #include "miner.h"
+#include "algo/sha/sha256d.h"

 #if defined(__AES__)               

--- a/algo/jh/jh-hash-4way.c
+++ b/algo/jh/jh-hash-4way.c
@@ -51,15 +51,15 @@ extern "C"{
 do { \
   __m512i cc = _mm512_set1_epi64( c ); \
    x3 = mm512_not( x3 ); \
-    x0 = _mm512_xor_si512( x0, _mm512_andnot_si512( x2, cc ) ); \
-    tmp = _mm512_xor_si512( cc, _mm512_and_si512( x0, x1 ) ); \
-    x0 = _mm512_xor_si512( x0, _mm512_and_si512( x2, x3 ) ); \
-    x3 = _mm512_xor_si512( x3, _mm512_andnot_si512( x1, x2 ) ); \
-    x1 = _mm512_xor_si512( x1, _mm512_and_si512( x0, x2 ) ); \
-    x2 = _mm512_xor_si512( x2, _mm512_andnot_si512( x3, x0 ) ); \
-    x0 = _mm512_xor_si512( x0, _mm512_or_si512( x1, x3 ) ); \
-    x3 = _mm512_xor_si512( x3, _mm512_and_si512( x1, x2 ) ); \
-    x1 = _mm512_xor_si512( x1, _mm512_and_si512( tmp, x0 ) ); \
+    x0 = mm512_xorandnot( x0, x2, cc ); \
+    tmp = mm512_xorand( cc, x0, x1 ); \
+    x0 = mm512_xorand( x0, x2, x3 ); \
+    x3 = mm512_xorandnot( x3, x1, x2 ); \
+    x1 = mm512_xorand( x1, x0, x2 ); \
+    x2 = mm512_xorandnot( x2, x3, x0 ); \
+    x0 = mm512_xoror( x0, x1, x3 ); \
+    x3 = mm512_xorand( x3, x1, x2 ); \
+    x1 = mm512_xorand( x1, tmp, x0 ); \
    x2 = _mm512_xor_si512( x2, tmp ); \
 } while (0)

@@ -67,11 +67,11 @@ do { \
 do { \
    x4 = _mm512_xor_si512( x4, x1 ); \
    x5 = _mm512_xor_si512( x5, x2 ); \
-    x6 = _mm512_xor_si512( x6, _mm512_xor_si512( x3, x0 ) ); \
+    x6 = mm512_xor3( x6, x3, x0 ); \
    x7 = _mm512_xor_si512( x7, x0 ); \
    x0 = _mm512_xor_si512( x0, x5 ); \
    x1 = _mm512_xor_si512( x1, x6 ); \
-    x2 = _mm512_xor_si512( x2, _mm512_xor_si512( x7, x4 ) ); \
+    x2 = mm512_xor3( x2, x7, x4 ); \
    x3 = _mm512_xor_si512( x3, x4 ); \
 } while (0)

@@ -318,12 +318,12 @@ static const sph_u64 C[] = {
 #define Wz_8W(x, c, n) \
 do { \
   __m512i t = _mm512_slli_epi64( _mm512_and_si512(x ## h, (c)), (n) ); \
-   x ## h = _mm512_or_si512( _mm512_and_si512( \
-                                _mm512_srli_epi64(x ## h, (n)), (c)), t ); \
+   x ## h = mm512_orand( t, _mm512_srli_epi64( x ## h, (n) ), (c) ); \
   t = _mm512_slli_epi64( _mm512_and_si512(x ## l, (c)), (n) ); \
-   x ## l = _mm512_or_si512( _mm512_and_si512((x ## l >> (n)), (c)), t ); \
+   x ## l = mm512_orand( t, (x ## l >> (n)), (c) ); \
 } while (0)

+
 #define W80(x)   Wz_8W(x, m512_const1_64( 0x5555555555555555 ),  1 )
 #define W81(x)   Wz_8W(x, m512_const1_64( 0x3333333333333333 ),  2 )
 #define W82(x)   Wz_8W(x, m512_const1_64( 0x0F0F0F0F0F0F0F0F ),  4 )
--- a/algo/keccak/keccak-gate.c
+++ b/algo/keccak/keccak-gate.c
@@ -1,5 +1,6 @@
 #include "keccak-gate.h"
 #include "sph_keccak.h"
+#include "algo/sha/sha256d.h"

 int hard_coded_eb = 1;

--- a/algo/keccak/keccak-hash-4way.c
+++ b/algo/keccak/keccak-hash-4way.c
@@ -70,12 +70,15 @@ static const uint64_t RC[] = {

 // Targetted macros, keccak-macros.h is included for each target.

-#define DECL64(x)        __m512i x
-#define XOR64(d, a, b)   (d = _mm512_xor_si512(a,b))
-#define AND64(d, a, b)   (d = _mm512_and_si512(a,b))
-#define OR64(d, a, b)    (d = _mm512_or_si512(a,b))
-#define NOT64(d, s)      (d = _mm512_xor_si512(s,m512_neg1))
-#define ROL64(d, v, n)   (d = mm512_rol_64(v, n))
+#define DECL64(x)          __m512i x
+#define XOR64(d, a, b)     (d = _mm512_xor_si512(a,b))
+#define AND64(d, a, b)     (d = _mm512_and_si512(a,b))
+#define OR64(d, a, b)      (d = _mm512_or_si512(a,b))
+#define NOT64(d, s)        (d = _mm512_xor_si512(s,m512_neg1))
+#define ROL64(d, v, n)     (d = mm512_rol_64(v, n))
+#define XOROR(d, a, b, c)  (d = mm512_xoror(a, b, c))
+#define XORAND(d, a, b, c) (d = mm512_xorand(a, b, c))
+

 #include "keccak-macros.c"

@@ -238,6 +241,8 @@ keccak512_8way_close(void *cc, void *dst)
 #undef NOT64
 #undef ROL64
 #undef KECCAK_F_1600
+#undef XOROR
+#undef XORAND

 #endif  // AVX512

@@ -255,6 +260,8 @@ keccak512_8way_close(void *cc, void *dst)
 #define OR64(d, a, b)    (d = _mm256_or_si256(a,b))
 #define NOT64(d, s)      (d = _mm256_xor_si256(s,m256_neg1))
 #define ROL64(d, v, n)   (d = mm256_rol_64(v, n))
+#define XOROR(d, a, b, c) (d = _mm256_xor_si256(a, _mm256_or_si256(b, c)))
+#define XORAND(d, a, b, c) (d = _mm256_xor_si256(a, _mm256_and_si256(b, c)))

 #include "keccak-macros.c"

@@ -419,5 +426,7 @@ keccak512_4way_close(void *cc, void *dst)
 #undef NOT64
 #undef ROL64
 #undef KECCAK_F_1600
+#undef XOROR
+#undef XORAND

 #endif  // AVX2
--- a/algo/keccak/keccak-macros.c
+++ b/algo/keccak/keccak-macros.c
@@ -110,20 +110,34 @@
 #ifdef KHI_XO
 #undef KHI_XO
 #endif
+
+#define KHI_XO(d, a, b, c)   do { \
+      XOROR(d, a, b, c); \
+   } while (0)
+
+/*
 #define KHI_XO(d, a, b, c)   do { \
                DECL64(kt); \
                OR64(kt, b, c); \
                XOR64(d, a, kt); \
        } while (0)
+*/

 #ifdef KHI_XA
 #undef KHI_XA
 #endif
+
+#define KHI_XA(d, a, b, c)   do { \
+      XORAND(d, a, b, c); \
+   } while (0)
+
+/*
 #define KHI_XA(d, a, b, c)   do { \
                DECL64(kt); \
                AND64(kt, b, c); \
                XOR64(d, a, kt); \
        } while (0)
+*/

 #ifdef KHI
 #undef KHI
@@ -134,65 +148,47 @@
        do { \
                DECL64(c0); \
                DECL64(c1); \
-                DECL64(c2); \
-                DECL64(c3); \
-                DECL64(c4); \
                DECL64(bnn); \
                NOT64(bnn, b20); \
                KHI_XO(c0, b00, b10, b20); \
                KHI_XO(c1, b10, bnn, b30); \
-                KHI_XA(c2, b20, b30, b40); \
-                KHI_XO(c3, b30, b40, b00); \
-                KHI_XA(c4, b40, b00, b10); \
+                KHI_XA(b20, b20, b30, b40); \
+                KHI_XO(b30, b30, b40, b00); \
+                KHI_XA(b40, b40, b00, b10); \
                MOV64(b00, c0); \
                MOV64(b10, c1); \
-                MOV64(b20, c2); \
-                MOV64(b30, c3); \
-                MOV64(b40, c4); \
                NOT64(bnn, b41); \
                KHI_XO(c0, b01, b11, b21); \
                KHI_XA(c1, b11, b21, b31); \
-                KHI_XO(c2, b21, b31, bnn); \
-                KHI_XO(c3, b31, b41, b01); \
-                KHI_XA(c4, b41, b01, b11); \
+                KHI_XO(b21, b21, b31, bnn); \
+                KHI_XO(b31, b31, b41, b01); \
+                KHI_XA(b41, b41, b01, b11); \
                MOV64(b01, c0); \
                MOV64(b11, c1); \
-                MOV64(b21, c2); \
-                MOV64(b31, c3); \
-                MOV64(b41, c4); \
                NOT64(bnn, b32); \
                KHI_XO(c0, b02, b12, b22); \
                KHI_XA(c1, b12, b22, b32); \
-                KHI_XA(c2, b22, bnn, b42); \
-                KHI_XO(c3, bnn, b42, b02); \
-                KHI_XA(c4, b42, b02, b12); \
+                KHI_XA(b22, b22, bnn, b42); \
+                KHI_XO(b32, bnn, b42, b02); \
+                KHI_XA(b42, b42, b02, b12); \
                MOV64(b02, c0); \
                MOV64(b12, c1); \
-                MOV64(b22, c2); \
-                MOV64(b32, c3); \
-                MOV64(b42, c4); \
                NOT64(bnn, b33); \
                KHI_XA(c0, b03, b13, b23); \
                KHI_XO(c1, b13, b23, b33); \
-                KHI_XO(c2, b23, bnn, b43); \
-                KHI_XA(c3, bnn, b43, b03); \
-                KHI_XO(c4, b43, b03, b13); \
+                KHI_XO(b23, b23, bnn, b43); \
+                KHI_XA(b33, bnn, b43, b03); \
+                KHI_XO(b43, b43, b03, b13); \
                MOV64(b03, c0); \
                MOV64(b13, c1); \
-                MOV64(b23, c2); \
-                MOV64(b33, c3); \
-                MOV64(b43, c4); \
                NOT64(bnn, b14); \
                KHI_XA(c0, b04, bnn, b24); \
                KHI_XO(c1, bnn, b24, b34); \
-                KHI_XA(c2, b24, b34, b44); \
-                KHI_XO(c3, b34, b44, b04); \
-                KHI_XA(c4, b44, b04, b14); \
+                KHI_XA(b24, b24, b34, b44); \
+                KHI_XO(b34, b34, b44, b04); \
+                KHI_XA(b44, b44, b04, b14); \
                MOV64(b04, c0); \
                MOV64(b14, c1); \
-                MOV64(b24, c2); \
-                MOV64(b34, c3); \
-                MOV64(b44, c4); \
        } while (0)

 #ifdef IOTA
@@ -201,6 +197,7 @@
 #define IOTA(r)   XOR64_IOTA(a00, a00, r)

 #ifdef P0
+#undef P0
 #undef P1
 #undef P2
 #undef P3
--- a/algo/luffa/luffa-hash-2way.c
+++ b/algo/luffa/luffa-hash-2way.c
@@ -66,6 +66,17 @@ static const uint32 CNS_INIT[128] __attribute((aligned(64))) = {
    a = _mm512_xor_si512(a,c0);\
    b = _mm512_xor_si512(b,c1);

+#define MULT24W( a0, a1 ) \
+do { \
+  __m512i b = _mm512_xor_si512( a0, \
+                     _mm512_maskz_shuffle_epi32( 0xbbbb, a1, 16 ) ); \
+  a0 = _mm512_or_si512( _mm512_bsrli_epi128(  b, 4 ), \
+                        _mm512_bslli_epi128( a1,12 ) ); \
+  a1 = _mm512_or_si512( _mm512_bsrli_epi128( a1, 4 ), \
+                        _mm512_bslli_epi128(  b,12 ) ); \
+} while(0)
+
+/*
 #define MULT24W( a0, a1, mask ) \
 do { \
  __m512i b = _mm512_xor_si512( a0, \
@@ -73,6 +84,7 @@ do { \
  a0 = _mm512_or_si512( _mm512_bsrli_epi128(b,4), _mm512_bslli_epi128(a1,12) );\
  a1 = _mm512_or_si512( _mm512_bsrli_epi128(a1,4), _mm512_bslli_epi128(b,12) );\
 } while(0)
+*/

 // confirm pointer arithmetic
 // ok but use array indexes
@@ -85,6 +97,21 @@ do { \
    MIXWORD4W(*(x+3),*(x+7),*t,*(t+1));\
    ADD_CONSTANT4W(*x, *(x+4), c0, c1);

+#define SUBCRUMB4W(a0,a1,a2,a3,t)\
+    t  = a0;\
+    a0 = mm512_xoror( a3, a0, a1 ); \
+    a2 = _mm512_xor_si512(a2,a3);\
+    a1 = _mm512_ternarylogic_epi64( a1, a3, t, 0x87 ); /* a1 xnor (a3 & t) */ \
+    a3 = mm512_xorand( a2, a3, t ); \
+    a2 = mm512_xorand( a1, a2, a0);\
+    a1 = _mm512_or_si512(a1,a3);\
+    a3 = _mm512_xor_si512(a3,a2);\
+    t  = _mm512_xor_si512(t,a1);\
+    a2 = _mm512_and_si512(a2,a1);\
+    a1 = mm512_xnor(a1,a0);\
+    a0 = t;
+
+/*
 #define SUBCRUMB4W(a0,a1,a2,a3,t)\
    t  = _mm512_load_si512(&a0);\
    a0 = _mm512_or_si512(a0,a1);\
@@ -103,7 +130,25 @@ do { \
    a2 = _mm512_and_si512(a2,a1);\
    a1 = _mm512_xor_si512(a1,a0);\
    a0 = _mm512_load_si512(&t);
+*/

+#define MIXWORD4W(a,b,t1,t2)\
+    b  = _mm512_xor_si512(a,b);\
+    t1 = _mm512_slli_epi32(a,2);\
+    t2 = _mm512_srli_epi32(a,30);\
+    a  = mm512_xoror( b, t1, t2 ); \
+    t1 = _mm512_slli_epi32(b,14);\
+    t2 = _mm512_srli_epi32(b,18);\
+    b  = _mm512_or_si512(t1,t2);\
+    b  = mm512_xoror( a, t1, t2 ); \
+    t1 = _mm512_slli_epi32(a,10);\
+    t2 = _mm512_srli_epi32(a,22);\
+    a  = mm512_xoror( b, t1, t2 ); \
+    t1 = _mm512_slli_epi32(b,1);\
+    t2 = _mm512_srli_epi32(b,31);\
+    b  = _mm512_or_si512(t1,t2);
+
+/*
 #define MIXWORD4W(a,b,t1,t2)\
    b  = _mm512_xor_si512(a,b);\
    t1 = _mm512_slli_epi32(a,2);\
@@ -121,6 +166,7 @@ do { \
    t1 = _mm512_slli_epi32(b,1);\
    t2 = _mm512_srli_epi32(b,31);\
    b  = _mm512_or_si512(t1,t2);
+*/

 #define STEP_PART24W(a0,a1,t0,t1,c0,c1,tmp0,tmp1)\
    a1 = _mm512_shuffle_epi32(a1,147);\
@@ -235,21 +281,13 @@ void rnd512_4way( luffa_4way_context *state, __m512i *msg )
    __m512i msg0, msg1;
    __m512i tmp[2];
    __m512i x[8];
-    const __m512i MASK = m512_const2_64( 0, 0x00000000ffffffff );

-    t0 = chainv[0];
-    t1 = chainv[1];
+    t0 = mm512_xor3( chainv[0], chainv[2], chainv[4] );
+    t1 = mm512_xor3( chainv[1], chainv[3], chainv[5] );
+    t0 = mm512_xor3( t0, chainv[6], chainv[8] );
+    t1 = mm512_xor3( t1, chainv[7], chainv[9] );

-    t0 = _mm512_xor_si512( t0, chainv[2] );
-    t1 = _mm512_xor_si512( t1, chainv[3] );
-    t0 = _mm512_xor_si512( t0, chainv[4] );
-    t1 = _mm512_xor_si512( t1, chainv[5] );
-    t0 = _mm512_xor_si512( t0, chainv[6] );
-    t1 = _mm512_xor_si512( t1, chainv[7] );
-    t0 = _mm512_xor_si512( t0, chainv[8] );
-    t1 = _mm512_xor_si512( t1, chainv[9] );
-
-    MULT24W( t0, t1, MASK );
+    MULT24W( t0, t1 );

    msg0 = _mm512_shuffle_epi32( msg[0], 27 );
    msg1 = _mm512_shuffle_epi32( msg[1], 27 );
@@ -268,68 +306,67 @@ void rnd512_4way( luffa_4way_context *state, __m512i *msg )
    t0 = chainv[0];
    t1 = chainv[1];

-    MULT24W( chainv[0], chainv[1], MASK );
+    MULT24W( chainv[0], chainv[1] );
    chainv[0] = _mm512_xor_si512( chainv[0], chainv[2] );
    chainv[1] = _mm512_xor_si512( chainv[1], chainv[3] );

-    MULT24W( chainv[2], chainv[3], MASK );
+    MULT24W( chainv[2], chainv[3] );
    chainv[2] = _mm512_xor_si512(chainv[2], chainv[4]);
    chainv[3] = _mm512_xor_si512(chainv[3], chainv[5]);

-    MULT24W( chainv[4], chainv[5], MASK );
+    MULT24W( chainv[4], chainv[5] );
    chainv[4] = _mm512_xor_si512(chainv[4], chainv[6]);
    chainv[5] = _mm512_xor_si512(chainv[5], chainv[7]);

-    MULT24W( chainv[6], chainv[7], MASK );
+    MULT24W( chainv[6], chainv[7] );
    chainv[6] = _mm512_xor_si512(chainv[6], chainv[8]);
    chainv[7] = _mm512_xor_si512(chainv[7], chainv[9]);

-    MULT24W( chainv[8], chainv[9], MASK );
+    MULT24W( chainv[8], chainv[9] );
    chainv[8] = _mm512_xor_si512( chainv[8], t0 );
    chainv[9] = _mm512_xor_si512( chainv[9], t1 );

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

-    MULT24W( chainv[8], chainv[9], MASK );
+    MULT24W( chainv[8], chainv[9] );
    chainv[8] = _mm512_xor_si512( chainv[8], chainv[6] );
    chainv[9] = _mm512_xor_si512( chainv[9], chainv[7] );

-    MULT24W( chainv[6], chainv[7], MASK );
+    MULT24W( chainv[6], chainv[7] );
    chainv[6] = _mm512_xor_si512( chainv[6], chainv[4] );
    chainv[7] = _mm512_xor_si512( chainv[7], chainv[5] );

-    MULT24W( chainv[4], chainv[5], MASK );
+    MULT24W( chainv[4], chainv[5] );
    chainv[4] = _mm512_xor_si512( chainv[4], chainv[2] );
    chainv[5] = _mm512_xor_si512( chainv[5], chainv[3] );

-    MULT24W( chainv[2], chainv[3], MASK );
+    MULT24W( chainv[2], chainv[3] );
    chainv[2] = _mm512_xor_si512( chainv[2], chainv[0] );
    chainv[3] = _mm512_xor_si512( chainv[3], chainv[1] );

-    MULT24W( chainv[0], chainv[1], MASK );
-    chainv[0] = _mm512_xor_si512( _mm512_xor_si512( chainv[0], t0 ), msg0 );
-    chainv[1] = _mm512_xor_si512( _mm512_xor_si512( chainv[1], t1 ), msg1 );
+    MULT24W( chainv[0], chainv[1] );
+    chainv[0] = mm512_xor3( chainv[0], t0, msg0 );
+    chainv[1] = mm512_xor3( chainv[1], t1, msg1 );

-    MULT24W( msg0, msg1, MASK );
+    MULT24W( msg0, msg1 );
    chainv[2] = _mm512_xor_si512( chainv[2], msg0 );
    chainv[3] = _mm512_xor_si512( chainv[3], msg1 );

-    MULT24W( msg0, msg1, MASK );
+    MULT24W( msg0, msg1 );
    chainv[4] = _mm512_xor_si512( chainv[4], msg0 );
    chainv[5] = _mm512_xor_si512( chainv[5], msg1 );

-    MULT24W( msg0, msg1, MASK );
+    MULT24W( msg0, msg1 );
    chainv[6] = _mm512_xor_si512( chainv[6], msg0 );
    chainv[7] = _mm512_xor_si512( chainv[7], msg1 );

-    MULT24W( msg0, msg1, MASK );
+    MULT24W( msg0, msg1);
    chainv[8] = _mm512_xor_si512( chainv[8], msg0 );
    chainv[9] = _mm512_xor_si512( chainv[9], msg1 );

-    MULT24W( msg0, msg1, MASK );
+    MULT24W( msg0, msg1 );

-    // replace with ror
    chainv[3] = _mm512_rol_epi32( chainv[3], 1 );
    chainv[5] = _mm512_rol_epi32( chainv[5], 2 );
    chainv[7] = _mm512_rol_epi32( chainv[7], 3 );
@@ -388,19 +425,11 @@ void finalization512_4way( luffa_4way_context *state, uint32 *b )

    /*---- blank round with m=0 ----*/
    rnd512_4way( state, zero );
-
-    t[0] = chainv[0];
-    t[1] = chainv[1];
-
-    t[0] = _mm512_xor_si512( t[0], chainv[2] );
-    t[1] = _mm512_xor_si512( t[1], chainv[3] );
-    t[0] = _mm512_xor_si512( t[0], chainv[4] );
-    t[1] = _mm512_xor_si512( t[1], chainv[5] );
-    t[0] = _mm512_xor_si512( t[0], chainv[6] );
-    t[1] = _mm512_xor_si512( t[1], chainv[7] );
-    t[0] = _mm512_xor_si512( t[0], chainv[8] );
-    t[1] = _mm512_xor_si512( t[1], chainv[9] );
-
+    
+    t[0] = mm512_xor3( chainv[0], chainv[2], chainv[4] );
+    t[1] = mm512_xor3( chainv[1], chainv[3], chainv[5] );
+    t[0] = mm512_xor3( t[0], chainv[6], chainv[8] );
+    t[1] = mm512_xor3( t[1], chainv[7], chainv[9] );
    t[0] = _mm512_shuffle_epi32( t[0], 27 );
    t[1] = _mm512_shuffle_epi32( t[1], 27 );

@@ -496,7 +525,7 @@ int luffa_4way_update( luffa_4way_context *state, const void *data,
    {
      // remaining data bytes
      buffer[0] = _mm512_shuffle_epi8( vdata[0], shuff_bswap32 );
-      buffer[1] = m512_const2_64( 0, 0x0000000080000000 );
+      buffer[1] = m512_const1_i128(  0x0000000080000000 );
    }
    return 0;
 }
@@ -520,7 +549,7 @@ int luffa_4way_close( luffa_4way_context *state, void *hashval )
      rnd512_4way( state, buffer );
    else
    {     // empty pad block, constant data
-      msg[0] = m512_const2_64( 0, 0x0000000080000000 );
+      msg[0] = m512_const1_i128(  0x0000000080000000 );
      msg[1] = m512_zero;
      rnd512_4way( state, msg );
    }
@@ -583,13 +612,13 @@ int luffa512_4way_full( luffa_4way_context *state, void *output,
    {
       // padding of partial block
       msg[0] = _mm512_shuffle_epi8( vdata[ 0 ], shuff_bswap32 );
-       msg[1] = m512_const2_64( 0, 0x0000000080000000 );
+       msg[1] = m512_const1_i128(  0x0000000080000000 );
       rnd512_4way( state, msg );
    }
    else
    {
       // empty pad block
-       msg[0] = m512_const2_64( 0, 0x0000000080000000 );
+       msg[0] = m512_const1_i128( 0x0000000080000000 );
       msg[1] = m512_zero;
       rnd512_4way( state, msg );
    }
@@ -631,13 +660,13 @@ int luffa_4way_update_close( luffa_4way_context *state,
    {
       // padding of partial block
       msg[0] = _mm512_shuffle_epi8( vdata[ 0 ], shuff_bswap32 );
-       msg[1] = m512_const2_64( 0, 0x0000000080000000 );
+       msg[1] = m512_const1_i128( 0x0000000080000000 );
       rnd512_4way( state, msg );
    }
    else
    {
       // empty pad block
-       msg[0] = m512_const2_64( 0, 0x0000000080000000 );
+       msg[0] = m512_const1_i128( 0x0000000080000000 );
       msg[1] = m512_zero;
       rnd512_4way( state, msg );
    }
@@ -666,8 +695,6 @@ do { \
  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,c0,c1,t)\
    SUBCRUMB(*x,*(x+1),*(x+2),*(x+3),*t);\
    SUBCRUMB(*(x+5),*(x+6),*(x+7),*(x+4),*t);\
@@ -678,23 +705,23 @@ do { \
    ADD_CONSTANT(*x, *(x+4), c0, c1);

 #define SUBCRUMB(a0,a1,a2,a3,t)\
-    t  = _mm256_load_si256(&a0);\
+    t  = a0;\
    a0 = _mm256_or_si256(a0,a1);\
    a2 = _mm256_xor_si256(a2,a3);\
-    a1 = _mm256_andnot_si256(a1, m256_neg1 );\
+    a1 = mm256_not( a1 );\
    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 );\
+    a0 = mm256_not( a0 );\
    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);\
+    a0 = t;\

 #define MIXWORD(a,b,t1,t2)\
    b  = _mm256_xor_si256(a,b);\
@@ -832,7 +859,7 @@ void rnd512_2way( luffa_2way_context *state, __m256i *msg )
    __m256i msg0, msg1;
    __m256i tmp[2];
    __m256i x[8];
-    const __m256i MASK = m256_const2_64( 0, 0x00000000ffffffff );
+    const __m256i MASK = m256_const1_i128( 0x00000000ffffffff );

    t0 = chainv[0];
    t1 = chainv[1];
@@ -1088,7 +1115,7 @@ int luffa_2way_update( luffa_2way_context *state, const void *data,
    {
      // remaining data bytes
      buffer[0] = _mm256_shuffle_epi8( vdata[0], shuff_bswap32 );
-      buffer[1] = m256_const2_64( 0, 0x0000000080000000 );
+      buffer[1] = m256_const1_i128( 0x0000000080000000 );
    }
    return 0;
 }
@@ -1104,7 +1131,7 @@ int luffa_2way_close( luffa_2way_context *state, void *hashval )
      rnd512_2way( state, buffer );
    else
    {     // empty pad block, constant data
-      msg[0] = m256_const2_64( 0, 0x0000000080000000 );
+      msg[0] = m256_const1_i128( 0x0000000080000000 );
      msg[1] = m256_zero;
      rnd512_2way( state, msg );
    }
@@ -1159,13 +1186,13 @@ int luffa512_2way_full( luffa_2way_context *state, void *output,
    {
       // padding of partial block
       msg[0] = _mm256_shuffle_epi8( vdata[ 0 ], shuff_bswap32 );
-       msg[1] = m256_const2_64( 0, 0x0000000080000000 );
+       msg[1] = m256_const1_i128( 0x0000000080000000 );
       rnd512_2way( state, msg );
    }
    else
    {
       // empty pad block
-       msg[0] = m256_const2_64( 0, 0x0000000080000000 );
+       msg[0] = m256_const1_i128( 0x0000000080000000 );
       msg[1] = m256_zero;
       rnd512_2way( state, msg );
    }
@@ -1206,13 +1233,13 @@ int luffa_2way_update_close( luffa_2way_context *state,
    {
       // padding of partial block
       msg[0] = _mm256_shuffle_epi8( vdata[ 0 ], shuff_bswap32 );
-       msg[1] = m256_const2_64( 0, 0x0000000080000000 );
+       msg[1] = m256_const1_i128( 0x0000000080000000 );
       rnd512_2way( state, msg );
    }
    else
    {
       // empty pad block
-       msg[0] = m256_const2_64( 0, 0x0000000080000000 );
+       msg[0] = m256_const1_i128( 0x0000000080000000 );
       msg[1] = m256_zero;
       rnd512_2way( state, msg );
    }
--- a/algo/luffa/luffa_for_sse2.c
+++ b/algo/luffa/luffa_for_sse2.c
@@ -23,7 +23,7 @@
 #include "simd-utils.h"
 #include "luffa_for_sse2.h"

-#define MULT2(a0,a1) do \
+#define MULT2( a0, a1 ) do \
 { \
  __m128i b =  _mm_xor_si128( a0, _mm_shuffle_epi32( _mm_and_si128(a1,MASK), 16 ) ); \
  a0 = _mm_or_si128( _mm_srli_si128(b,4), _mm_slli_si128(a1,12) ); \
@@ -345,11 +345,11 @@ HashReturn update_and_final_luffa( hashState_luffa *state, BitSequence* output,
    // 16 byte partial block exists for 80 byte len
    if ( state->rembytes  )
       // padding of partial block
-       rnd512( state, m128_const_64( 0, 0x80000000 ),
+       rnd512( state, m128_const_i128(  0x80000000 ),
                      mm128_bswap_32( cast_m128i( data ) ) );
    else
       // empty pad block
-       rnd512( state, m128_zero, m128_const_64( 0, 0x80000000 ) );
+       rnd512( state, m128_zero, m128_const_i128( 0x80000000 ) );

    finalization512( state, (uint32*) output );
    if ( state->hashbitlen > 512 )
@@ -394,11 +394,11 @@ int luffa_full( hashState_luffa *state, BitSequence* output, int hashbitlen,
    // 16 byte partial block exists for 80 byte len
    if ( state->rembytes  )
       // padding of partial block
-       rnd512( state, m128_const_64( 0, 0x80000000 ),
+       rnd512( state, m128_const_i128( 0x80000000 ),
                      mm128_bswap_32( cast_m128i( data ) ) );
    else
       // empty pad block
-       rnd512( state, m128_zero, m128_const_64( 0, 0x80000000 ) );
+       rnd512( state, m128_zero, m128_const_i128( 0x80000000 ) );

    finalization512( state, (uint32*) output );
    if ( state->hashbitlen > 512 )
@@ -606,7 +606,6 @@ static void finalization512( hashState_luffa *state, uint32 *b )

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

    rnd512( state, zero, zero );

@@ -621,7 +620,6 @@ static void finalization512( hashState_luffa *state, uint32 *b )

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

 #else
--- a/algo/lyra2/allium-4way.c
+++ b/algo/lyra2/allium-4way.c
@@ -16,7 +16,7 @@
 typedef struct {
   blake256_16way_context     blake;
   keccak256_8way_context    keccak;
-   cube_4way_context          cube;
+   cube_4way_2buf_context    cube;
   skein256_8way_context     skein;
 #if defined(__VAES__)
   groestl256_4way_context groestl;
@@ -30,13 +30,7 @@ static __thread allium_16way_ctx_holder allium_16way_ctx;
 bool init_allium_16way_ctx()
 {
   keccak256_8way_init( &allium_16way_ctx.keccak );
-   cube_4way_init( &allium_16way_ctx.cube, 256, 16, 32 );
   skein256_8way_init( &allium_16way_ctx.skein );
-#if defined(__VAES__)
-   groestl256_4way_init( &allium_16way_ctx.groestl, 32 );
-#else
-   init_groestl256( &allium_16way_ctx.groestl, 32 );
-#endif
   return true;
 }

@@ -111,12 +105,11 @@ void allium_16way_hash( void *state, const void *input )
   intrlv_2x256( vhash, hash14, hash15, 256 );
   LYRA2RE_2WAY( vhash, 32, vhash, 32, 1, 8, 8 );
   dintrlv_2x256( hash14, hash15, vhash, 256 );
-  
+
   intrlv_4x128( vhashA, hash0, hash1, hash2, hash3, 256 );
   intrlv_4x128( vhashB, hash4, hash5, hash6, hash7, 256 );

-   cube_4way_full( &ctx.cube, vhashA, 256, vhashA, 32 );
-   cube_4way_full( &ctx.cube, vhashB, 256, vhashB, 32 );
+   cube_4way_2buf_full( &ctx.cube, vhashA, vhashB, 256, vhashA, vhashB, 32 );

   dintrlv_4x128( hash0, hash1, hash2, hash3, vhashA, 256 );
   dintrlv_4x128( hash4, hash5, hash6, hash7, vhashB, 256 );
@@ -124,8 +117,7 @@ void allium_16way_hash( void *state, const void *input )
   intrlv_4x128( vhashA, hash8, hash9, hash10, hash11, 256 );
   intrlv_4x128( vhashB, hash12, hash13, hash14, hash15, 256 );

-   cube_4way_full( &ctx.cube, vhashA, 256, vhashA, 32 );
-   cube_4way_full( &ctx.cube, vhashB, 256, vhashB, 32 );
+   cube_4way_2buf_full( &ctx.cube, vhashA, vhashB, 256, vhashA, vhashB, 32 );

   dintrlv_4x128( hash8, hash9, hash10, hash11, vhashA, 256 );
   dintrlv_4x128( hash12, hash13, hash14, hash15, vhashB, 256 );
@@ -174,24 +166,19 @@ void allium_16way_hash( void *state, const void *input )
 #if defined(__VAES__)

   intrlv_4x128( vhash, hash0, hash1, hash2, hash3, 256 );
-
-   groestl256_4way_full( &ctx.groestl, vhash, vhash, 256 );
-
+   groestl256_4way_full( &ctx.groestl, vhash, vhash, 32 );
   dintrlv_4x128( state, state+32, state+64, state+96, vhash, 256 );
+
   intrlv_4x128( vhash, hash4, hash5, hash6, hash7, 256 );
-
-   groestl256_4way_full( &ctx.groestl, vhash, vhash, 256 );
-   
+   groestl256_4way_full( &ctx.groestl, vhash, vhash, 32 );
   dintrlv_4x128( state+128, state+160, state+192, state+224, vhash, 256 );
+
   intrlv_4x128( vhash, hash8, hash9, hash10, hash11, 256 );
-
-   groestl256_4way_full( &ctx.groestl, vhash, vhash, 256 );
-
+   groestl256_4way_full( &ctx.groestl, vhash, vhash, 32 );
   dintrlv_4x128( state+256, state+288, state+320, state+352, vhash, 256 );
-   intrlv_4x128( vhash, hash12, hash13, hash14, hash15, 256 );

-   groestl256_4way_full( &ctx.groestl, vhash, vhash, 256 );
- 
+   intrlv_4x128( vhash, hash12, hash13, hash14, hash15, 256 );
+   groestl256_4way_full( &ctx.groestl, vhash, vhash, 32 );
   dintrlv_4x128( state+384, state+416, state+448, state+480, vhash, 256 );
   
 #else
@@ -260,10 +247,13 @@ int scanhash_allium_16way( struct work *work, uint32_t max_nonce,
 typedef struct {
   blake256_8way_context     blake;
   keccak256_4way_context    keccak;
-   cubehashParam             cube;
+   cube_2way_context         cube;
   skein256_4way_context     skein;
+#if defined(__VAES__)
+   groestl256_2way_context   groestl;
+#else
   hashState_groestl256      groestl;
-
+#endif
 } allium_8way_ctx_holder;

 static __thread allium_8way_ctx_holder allium_8way_ctx;
@@ -271,9 +261,7 @@ static __thread allium_8way_ctx_holder allium_8way_ctx;
 bool init_allium_8way_ctx()
 {
   keccak256_4way_init( &allium_8way_ctx.keccak );
-   cubehashInit( &allium_8way_ctx.cube, 256, 16, 32 );
   skein256_4way_init( &allium_8way_ctx.skein );
-   init_groestl256( &allium_8way_ctx.groestl, 32 );
   return true;
 }

@@ -318,21 +306,20 @@ void allium_8way_hash( void *hash, const void *input )
   LYRA2RE( hash6, 32, hash6, 32, hash6, 32, 1, 8, 8 );
   LYRA2RE( hash7, 32, hash7, 32, hash7, 32, 1, 8, 8 );

-   cubehashUpdateDigest( &ctx.cube, (byte*)hash0, (const byte*)hash0, 32 );
-   cubehashInit( &ctx.cube, 256, 16, 32 );
-   cubehashUpdateDigest( &ctx.cube, (byte*)hash1, (const byte*)hash1, 32 );
-   cubehashInit( &ctx.cube, 256, 16, 32 );
-   cubehashUpdateDigest( &ctx.cube, (byte*)hash2, (const byte*)hash2, 32 );
-   cubehashInit( &ctx.cube, 256, 16, 32 );
-   cubehashUpdateDigest( &ctx.cube, (byte*)hash3, (const byte*)hash3, 32 );
-   cubehashInit( &ctx.cube, 256, 16, 32 );
-   cubehashUpdateDigest( &ctx.cube, (byte*)hash4, (const byte*)hash4, 32 );
-   cubehashInit( &ctx.cube, 256, 16, 32 );
-   cubehashUpdateDigest( &ctx.cube, (byte*)hash5, (const byte*)hash5, 32 );
-   cubehashInit( &ctx.cube, 256, 16, 32 );
-   cubehashUpdateDigest( &ctx.cube, (byte*)hash6, (const byte*)hash6, 32 );
-   cubehashInit( &ctx.cube, 256, 16, 32 );
-   cubehashUpdateDigest( &ctx.cube, (byte*)hash7, (const byte*)hash7, 32 );
+
+   intrlv_2x128( vhashA, hash0, hash1, 256 );
+   intrlv_2x128( vhashB, hash2, hash3, 256 );
+   cube_2way_full( &ctx.cube, vhashA, 256, vhashA, 32 );
+   cube_2way_full( &ctx.cube, vhashB, 256, vhashB, 32 );
+   dintrlv_2x128( hash0, hash1, vhashA, 256 );
+   dintrlv_2x128( hash2, hash3, vhashB, 256 );
+
+   intrlv_2x128( vhashA, hash4, hash5, 256 );
+   intrlv_2x128( vhashB, hash6, hash7, 256 );
+   cube_2way_full( &ctx.cube, vhashA, 256, vhashA, 32 );
+   cube_2way_full( &ctx.cube, vhashB, 256, vhashB, 32 );
+   dintrlv_2x128( hash4, hash5, vhashA, 256 );
+   dintrlv_2x128( hash6, hash7, vhashB, 256 );

   LYRA2RE( hash0, 32, hash0, 32, hash0, 32, 1, 8, 8 );
   LYRA2RE( hash1, 32, hash1, 32, hash1, 32, 1, 8, 8 );
@@ -352,9 +339,28 @@ void allium_8way_hash( void *hash, const void *input )
   skein256_4way_update( &ctx.skein, vhashB, 32 );
   skein256_4way_close( &ctx.skein, vhashB );

+#if defined(__VAES__)
+
+   uint64_t vhashC[4*2] __attribute__ ((aligned (64)));
+   uint64_t vhashD[4*2] __attribute__ ((aligned (64)));
+   
+   rintrlv_4x64_2x128( vhashC, vhashD, vhashA, 256 );
+   groestl256_2way_full( &ctx.groestl, vhashC, vhashC, 32 );
+   groestl256_2way_full( &ctx.groestl, vhashD, vhashD, 32 );
+   dintrlv_2x128( hash0, hash1, vhashC, 256 );
+   dintrlv_2x128( hash2, hash3, vhashD, 256 );
+
+   rintrlv_4x64_2x128( vhashC, vhashD, vhashB, 256 );
+   groestl256_2way_full( &ctx.groestl, vhashC, vhashC, 32 );
+   groestl256_2way_full( &ctx.groestl, vhashD, vhashD, 32 );
+   dintrlv_2x128( hash4, hash5, vhashC, 256 );
+   dintrlv_2x128( hash6, hash7, vhashD, 256 );
+
+#else
+
   dintrlv_4x64( hash0, hash1, hash2, hash3, vhashA, 256 );
   dintrlv_4x64( hash4, hash5, hash6, hash7, vhashB, 256 );
-
+   
   groestl256_full( &ctx.groestl, hash0, hash0, 256 );
   groestl256_full( &ctx.groestl, hash1, hash1, 256 );
   groestl256_full( &ctx.groestl, hash2, hash2, 256 );
@@ -363,6 +369,8 @@ void allium_8way_hash( void *hash, const void *input )
   groestl256_full( &ctx.groestl, hash5, hash5, 256 );
   groestl256_full( &ctx.groestl, hash6, hash6, 256 );
   groestl256_full( &ctx.groestl, hash7, hash7, 256 );
+
+#endif
 }

 int scanhash_allium_8way( struct work *work, uint32_t max_nonce,
--- a/algo/lyra2/lyra2-gate.c
+++ b/algo/lyra2/lyra2-gate.c
@@ -187,7 +187,8 @@ bool register_allium_algo( algo_gate_t* gate )
  gate->scanhash  = (void*)&scanhash_allium;
  gate->hash      = (void*)&allium_hash;
 #endif
-  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT | VAES_OPT;
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT
+	              | VAES_OPT | VAES256_OPT;
  opt_target_factor = 256.0;
  return true;
 };
--- a/algo/lyra2/phi2-4way.c
+++ b/algo/lyra2/phi2-4way.c
@@ -4,7 +4,7 @@
 #include "algo/gost/sph_gost.h"
 #include "algo/cubehash/cubehash_sse2.h"
 #include "lyra2.h"
-#if defined(__VAES__)
+#if defined(__VAES__) && defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
  #include "algo/echo/echo-hash-4way.h"
 #elif defined(__AES__)
  #include "algo/echo/aes_ni/hash_api.h"
--- a/algo/lyra2/sponge.h
+++ b/algo/lyra2/sponge.h
@@ -66,13 +66,13 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){

 #define LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
   G2W_4X64( s0, s1, s2, s3 ); \
-   s1 = mm512_ror256_64( s1); \
+   s3 = mm512_shufll256_64( s3 ); \
+   s1 = mm512_shuflr256_64( s1); \
   s2 = mm512_swap256_128( s2 ); \
-   s3 = mm512_rol256_64( s3 ); \
   G2W_4X64( s0, s1, s2, s3 ); \
-   s1 = mm512_rol256_64( s1 ); \
-   s2 = mm512_swap256_128( s2 ); \
-   s3 = mm512_ror256_64( s3 );
+   s3 = mm512_shuflr256_64( s3 ); \
+   s1 = mm512_shufll256_64( s1 ); \
+   s2 = mm512_swap256_128( s2 ); 

 #define LYRA_12_ROUNDS_2WAY_AVX512( s0, s1, s2, s3 ) \
   LYRA_ROUND_2WAY_AVX512( s0, s1, s2, s3 ) \
@@ -107,13 +107,13 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){

 #define LYRA_ROUND_AVX2( s0, s1, s2, s3 ) \
   G_4X64( s0, s1, s2, s3 ); \
-   s1 = mm256_ror_1x64( s1); \
+   s3 = mm256_shufll_64( s3 ); \
+   s1 = mm256_shuflr_64( s1); \
   s2 = mm256_swap_128( s2 ); \
-   s3 = mm256_rol_1x64( s3 ); \
   G_4X64( s0, s1, s2, s3 ); \
-   s1 = mm256_rol_1x64( s1 ); \
-   s2 = mm256_swap_128( s2 ); \
-   s3 = mm256_ror_1x64( s3 );
+   s3 = mm256_shuflr_64( s3 ); \
+   s1 = mm256_shufll_64( s1 ); \
+   s2 = mm256_swap_128( s2 );

 #define LYRA_12_ROUNDS_AVX2( s0, s1, s2, s3 ) \
   LYRA_ROUND_AVX2( s0, s1, s2, s3 ) \
@@ -148,14 +148,14 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
 #define LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
   G_2X64( s0, s2, s4, s6 ); \
   G_2X64( s1, s3, s5, s7 ); \
-   mm128_ror256_64( s2, s3 ); \
+   mm128_vrol256_64( s6, s7 ); \
+   mm128_vror256_64( s2, s3 ); \
   mm128_swap256_128( s4, s5 ); \
-   mm128_rol256_64( s6, s7 ); \
   G_2X64( s0, s2, s4, s6 ); \
   G_2X64( s1, s3, s5, s7 ); \
-   mm128_rol256_64( s2, s3 ); \
-   mm128_swap256_128( s4, s5 ); \
-   mm128_ror256_64( s6, s7 );
+   mm128_vror256_64( s6, s7 ); \
+   mm128_vrol256_64( s2, s3 ); \
+   mm128_swap256_128( s4, s5 );

 #define LYRA_12_ROUNDS_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
   LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
--- a/algo/m7m/m7m.c
+++ b/algo/m7m/m7m.c
@@ -12,8 +12,8 @@
 #include "algo/tiger/sph_tiger.h"
 #include "algo/whirlpool/sph_whirlpool.h"
 #include "algo/ripemd/sph_ripemd.h"
-#include <openssl/sha.h>
-
+#include "algo/sha/sph_sha2.h"
+#include "algo/sha/sha256-hash.h"

 #define EPSa DBL_EPSILON
 #define EPS1 DBL_EPSILON
@@ -105,8 +105,8 @@ uint32_t sw2_( int nnounce )
 }

 typedef struct {
-    SHA256_CTX               sha256;
-    SHA512_CTX               sha512;
+    sha256_context           sha256;
+    sph_sha512_context       sha512;
    sph_keccak512_context    keccak;
    sph_whirlpool_context    whirlpool;
    sph_haval256_5_context   haval;
@@ -118,8 +118,8 @@ m7m_ctx_holder m7m_ctx;

 void init_m7m_ctx()
 {
-    SHA256_Init( &m7m_ctx.sha256 );
-    SHA512_Init( &m7m_ctx.sha512 );
+    sha256_ctx_init( &m7m_ctx.sha256 );
+    sph_sha512_init( &m7m_ctx.sha512 );
    sph_keccak512_init( &m7m_ctx.keccak );
    sph_whirlpool_init( &m7m_ctx.whirlpool );
    sph_haval256_5_init( &m7m_ctx.haval );
@@ -143,11 +143,10 @@ int scanhash_m7m_hash( struct work* work, uint64_t max_nonce,
    uint32_t hash[8] __attribute__((aligned(64)));
    uint8_t bhash[7][64] __attribute__((aligned(64)));
    uint32_t n = pdata[19] - 1;
-    int thr_id = mythr->id;  // thr_id arg is deprecated
+    int thr_id = mythr->id;
    uint32_t usw_, mpzscale;
    const uint32_t first_nonce = pdata[19];
    char data_str[161], hash_str[65], target_str[65];
-    //uint8_t *bdata = 0;
    uint8_t bdata[8192] __attribute__ ((aligned (64)));
    int i, digits;
    int bytes;
@@ -155,12 +154,11 @@ int scanhash_m7m_hash( struct work* work, uint64_t max_nonce,

    m7m_ctx_holder ctx1, ctx2 __attribute__ ((aligned (64)));
    memcpy( &ctx1, &m7m_ctx, sizeof(m7m_ctx) );
-    SHA256_CTX         ctxf_sha256;

    memcpy(data, pdata, 80);

-    SHA256_Update(  &ctx1.sha256,    data, M7_MIDSTATE_LEN );
-    SHA512_Update(  &ctx1.sha512,    data, M7_MIDSTATE_LEN );
+    sha256_update(  &ctx1.sha256,    data, M7_MIDSTATE_LEN );
+    sph_sha512(     &ctx1.sha512,    data, M7_MIDSTATE_LEN );
    sph_keccak512(  &ctx1.keccak,    data, M7_MIDSTATE_LEN );
    sph_whirlpool(  &ctx1.whirlpool, data, M7_MIDSTATE_LEN );
    sph_haval256_5( &ctx1.haval,     data, M7_MIDSTATE_LEN );
@@ -191,11 +189,11 @@ int scanhash_m7m_hash( struct work* work, uint64_t max_nonce,

        memcpy( &ctx2, &ctx1, sizeof(m7m_ctx) );

-        SHA256_Update(  &ctx2.sha256, data_p64, 80 - M7_MIDSTATE_LEN );
-        SHA256_Final( (unsigned char*) (bhash[0]), &ctx2.sha256 );
+        sha256_update( &ctx2.sha256, data_p64, 80 - M7_MIDSTATE_LEN );
+        sha256_final( &ctx2.sha256, bhash[0] );

-        SHA512_Update(  &ctx2.sha512, data_p64, 80 - M7_MIDSTATE_LEN );
-        SHA512_Final( (unsigned char*) (bhash[1]), &ctx2.sha512 );
+        sph_sha512(  &ctx2.sha512, data_p64, 80 - M7_MIDSTATE_LEN );
+        sph_sha512_close( &ctx2.sha512, bhash[1] );

        sph_keccak512( &ctx2.keccak, data_p64, 80 - M7_MIDSTATE_LEN );
        sph_keccak512_close( &ctx2.keccak, (void*)(bhash[2]) );
@@ -227,9 +225,7 @@ int scanhash_m7m_hash( struct work* work, uint64_t max_nonce,
        bytes = mpz_sizeinbase(product, 256);
        mpz_export((void *)bdata, NULL, -1, 1, 0, 0, product);

-        SHA256_Init( &ctxf_sha256 );
-        SHA256_Update(  &ctxf_sha256, bdata, bytes );
-        SHA256_Final( (unsigned char*) hash, &ctxf_sha256 );
+        sha256_full( hash, bdata, bytes );

        digits=(int)((sqrt((double)(n/2))*(1.+EPS))/9000+75);
        mp_bitcnt_t prec = (long int)(digits*BITS_PER_DIGIT+16);
@@ -262,18 +258,11 @@ int scanhash_m7m_hash( struct work* work, uint64_t max_nonce,
            mpzscale=bytes;
            mpz_export(bdata, NULL, -1, 1, 0, 0, product);

-            SHA256_Init( &ctxf_sha256 );
-            SHA256_Update(  &ctxf_sha256, bdata, bytes );
-            SHA256_Final( (unsigned char*) hash, &ctxf_sha256 );
-        }
-
+            sha256_full( hash, bdata, bytes );
+	     }

        if ( unlikely( valid_hash( (uint64_t*)hash, (uint64_t*)ptarget ) 
             && !opt_benchmark ) )
-
-
-//        if ( unlikely( hash[7] <= ptarget[7] ) )
-//        if ( likely( fulltest( hash, ptarget ) && !opt_benchmark ) )        
        {
           if ( opt_debug )
           {
--- a/algo/nist5/zr5.c
+++ b/algo/nist5/zr5.c
@@ -156,7 +156,7 @@ int scanhash_zr5( struct work *work, uint32_t max_nonce,
 void zr5_get_new_work( struct work* work, struct work* g_work, int thr_id,
                       uint32_t* end_nonce_ptr )
 {
-   pthread_rwlock_rdlock( &g_work_lock );
+//   pthread_rwlock_rdlock( &g_work_lock );

   // ignore POK in first word
   const int wkcmp_sz = 72;  // (19-1) * sizeof(uint32_t)
@@ -174,7 +174,7 @@ void zr5_get_new_work( struct work* work, struct work* g_work, int thr_id,
   else
       ++(*nonceptr);

-   pthread_rwlock_unlock( &g_work_lock );
+//   pthread_rwlock_unlock( &g_work_lock );
 }

 void zr5_display_pok( struct work* work )
--- a/algo/panama/panama-hash-4way.c
+++ b/algo/panama/panama-hash-4way.c
@@ -312,10 +312,26 @@ do { \
      BUPDATE1_8W( 7, 1 ); \
 } while (0)

+#if defined(__AVX512VL__)
+
+#define GAMMA_8W(n0, n1, n2, n4)   \
+   ( g ## n0 = _mm256_ternarylogic_epi32( a ## n0, a ## n2, a ## n1, 0x4b ) )  
+
+#define THETA_8W(n0, n1, n2, n4)   \
+   ( g ## n0 = mm256_xor3( a ## n0, a ## n1, a ## n4 ) )   
+
+#else
+
 #define GAMMA_8W(n0, n1, n2, n4)   \
   (g ## n0 = _mm256_xor_si256( a ## n0, \
                         _mm256_or_si256( a ## n1, mm256_not( a ## n2 ) ) ) )

+#define THETA_8W(n0, n1, n2, n4)   \
+   ( g ## n0 = _mm256_xor_si256( a ## n0, _mm256_xor_si256( a ## n1, \
+                                                            a ## n4 ) ) )
+
+#endif
+
 #define PI_ALL_8W   do { \
      a0  = g0; \
      a1  = mm256_rol_32( g7,   1 ); \
@@ -336,9 +352,6 @@ do { \
      a16 = mm256_rol_32( g10,  8 ); \
   } while (0)

-#define THETA_8W(n0, n1, n2, n4)   \
-   ( g ## n0 = _mm256_xor_si256( a ## n0, _mm256_xor_si256( a ## n1, \
-                                                            a ## n4 ) ) )

 #define SIGMA_ALL_8W   do { \
      a0  = _mm256_xor_si256( g0, m256_one_32 ); \
--- a/algo/quark/hmq1725-4way.c
+++ b/algo/quark/hmq1725-4way.c
@@ -912,7 +912,7 @@ extern void hmq1725_4way_hash(void *state, const void *input)
   sph_whirlpool512_full( &ctx.whirlpool, hash2, hash2, 64 );
   sph_whirlpool512_full( &ctx.whirlpool, hash3, hash3, 64 );

-// A = fugue serial, B = sha512 prarallel
+// A = fugue serial, B = sha512 parallel
   
   intrlv_4x64( vhash, hash0, hash1, hash2, hash3, 512 );

--- a/algo/quark/hmq1725.c
+++ b/algo/quark/hmq1725.c
@@ -17,7 +17,7 @@
 #include "algo/shabal/sph_shabal.h"
 #include "algo/whirlpool/sph_whirlpool.h"
 #include "algo/haval/sph-haval.h"
-#include <openssl/sha.h>
+#include "algo/sha/sph_sha2.h"
 #if defined(__AES__)
  #include "algo/groestl/aes_ni/hash-groestl.h"
  #include "algo/echo/aes_ni/hash_api.h"
@@ -44,7 +44,7 @@ typedef struct {
  sph_hamsi512_context    hamsi1;
  sph_shabal512_context   shabal1;
  sph_whirlpool_context   whirlpool1, whirlpool2, whirlpool3, whirlpool4;
-  SHA512_CTX              sha1, sha2;
+  sph_sha512_context      sha1, sha2;
  sph_haval256_5_context  haval1, haval2;
 #if defined(__AES__)
  hashState_echo          echo1, echo2;
@@ -106,8 +106,8 @@ void init_hmq1725_ctx()
    sph_whirlpool_init(&hmq1725_ctx.whirlpool3);
    sph_whirlpool_init(&hmq1725_ctx.whirlpool4);

-    SHA512_Init( &hmq1725_ctx.sha1 );
-    SHA512_Init( &hmq1725_ctx.sha2 );
+    sph_sha512_init( &hmq1725_ctx.sha1 );
+    sph_sha512_init( &hmq1725_ctx.sha2 );

    sph_haval256_5_init(&hmq1725_ctx.haval1);
    sph_haval256_5_init(&hmq1725_ctx.haval2);
@@ -285,8 +285,8 @@ extern void hmq1725hash(void *state, const void *input)
    }
    else
    {
-        SHA512_Update( &h_ctx.sha1, hashB, 64 );
-        SHA512_Final( (unsigned char*) hashA, &h_ctx.sha1 );
+        sph_sha512( &h_ctx.sha1, hashB, 64 );
+        sph_sha512_close( &h_ctx.sha1, hashA );
    }

 #if defined(__AES__)
@@ -297,8 +297,8 @@ extern void hmq1725hash(void *state, const void *input)
    sph_groestl512_close(&h_ctx.groestl2, hashB); //4
 #endif

-    SHA512_Update( &h_ctx.sha2, hashB, 64 );
-    SHA512_Final( (unsigned char*) hashA, &h_ctx.sha2 );
+    sph_sha512( &h_ctx.sha2, hashB, 64 );
+    sph_sha512_close( &h_ctx.sha2, hashA );

    if ( hashA[0] & mask ) //4
    {
--- a/algo/quark/quark-4way.c
+++ b/algo/quark/quark-4way.c
@@ -127,10 +127,8 @@ void quark_8way_hash( void *state, const void *input )

     rintrlv_8x64_4x128( vhashA, vhashB, vhash, 512 );

-     if ( ( vh_mask & 0x0f ) != 0x0f )
-       groestl512_4way_full( &ctx.groestl, vhashA, vhashA, 64 );
-     if ( ( vh_mask & 0xf0 ) != 0xf0 )
-       groestl512_4way_full( &ctx.groestl, vhashB, vhashB, 64 );
+     groestl512_4way_full( &ctx.groestl, vhashA, vhashA, 64 );
+     groestl512_4way_full( &ctx.groestl, vhashB, vhashB, 64 );

     rintrlv_4x128_8x64( vhash, vhashA, vhashB, 512 );

@@ -139,22 +137,14 @@ void quark_8way_hash( void *state, const void *input )
    dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
                  vhash, 512 );

-    if ( hash0[0] & 8 )
-       groestl512_full( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
-    if ( hash1[0] & 8 )
-       groestl512_full( &ctx.groestl, (char*)hash1, (char*)hash1, 512 );
-    if ( hash2[0] & 8)
-       groestl512_full( &ctx.groestl, (char*)hash2, (char*)hash2, 512 );
-    if ( hash3[0] & 8 )
-       groestl512_full( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
-    if ( hash4[0] & 8 )
-       groestl512_full( &ctx.groestl, (char*)hash4, (char*)hash4, 512 );
-    if ( hash5[0] & 8 )
-       groestl512_full( &ctx.groestl, (char*)hash5, (char*)hash5, 512 );
-    if ( hash6[0] & 8 )
-       groestl512_full( &ctx.groestl, (char*)hash6, (char*)hash6, 512 );
-    if ( hash7[0] & 8 )
-       groestl512_full( &ctx.groestl, (char*)hash7, (char*)hash7, 512 );
+    groestl512_full( &ctx.groestl, (char*)hash0, (char*)hash0, 512 );
+    groestl512_full( &ctx.groestl, (char*)hash1, (char*)hash1, 512 );
+    groestl512_full( &ctx.groestl, (char*)hash2, (char*)hash2, 512 );
+    groestl512_full( &ctx.groestl, (char*)hash3, (char*)hash3, 512 );
+    groestl512_full( &ctx.groestl, (char*)hash4, (char*)hash4, 512 );
+    groestl512_full( &ctx.groestl, (char*)hash5, (char*)hash5, 512 );
+    groestl512_full( &ctx.groestl, (char*)hash6, (char*)hash6, 512 );
+    groestl512_full( &ctx.groestl, (char*)hash7, (char*)hash7, 512 );

    intrlv_8x64( vhash, hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
                 512 );
--- a/algo/ripemd/lbry.c
+++ b/algo/ripemd/lbry.c
@@ -7,28 +7,23 @@
 #include <string.h>
 #include <stdio.h>
 #include "sph_ripemd.h"
-#include <openssl/sha.h>
+#include "algo/sha/sha256-hash.h"

 void lbry_hash(void* output, const void* input)
 {
-   SHA256_CTX              ctx_sha256 __attribute__ ((aligned (64)));
-   SHA512_CTX              ctx_sha512 __attribute__ ((aligned (64)));
-   sph_ripemd160_context   ctx_ripemd __attribute__ ((aligned (64)));
+   sha256_context        ctx_sha256 __attribute__ ((aligned (64)));
+   sph_sha512_context    ctx_sha512 __attribute__ ((aligned (64)));
+   sph_ripemd160_context ctx_ripemd __attribute__ ((aligned (64)));
   uint32_t _ALIGN(64) hashA[16];
   uint32_t _ALIGN(64) hashB[16];
   uint32_t _ALIGN(64) hashC[16];

-   SHA256_Init( &ctx_sha256 );
-   SHA256_Update( &ctx_sha256, input, 112 );
-   SHA256_Final( (unsigned char*) hashA, &ctx_sha256 );
+   sha256_full( hashA, input, 112 );
+   sha256_full( hashA, hashA, 32 );

-   SHA256_Init( &ctx_sha256 );
-   SHA256_Update( &ctx_sha256, hashA, 32 );
-   SHA256_Final( (unsigned char*) hashA, &ctx_sha256 );
-
-   SHA512_Init( &ctx_sha512 );
-   SHA512_Update( &ctx_sha512, hashA, 32 );
-   SHA512_Final( (unsigned char*) hashA, &ctx_sha512 );
+   sph_sha512_init( &ctx_sha512 );
+   sph_sha512( &ctx_sha512, hashA, 32 );
+   sph_sha512_close( &ctx_sha512, hashA );

   sph_ripemd160_init( &ctx_ripemd );
   sph_ripemd160 ( &ctx_ripemd, hashA, 32 );
@@ -38,15 +33,13 @@ void lbry_hash(void* output, const void* input)
   sph_ripemd160 ( &ctx_ripemd, hashA+8, 32 );
   sph_ripemd160_close( &ctx_ripemd, hashC );

-   SHA256_Init( &ctx_sha256 );
-   SHA256_Update( &ctx_sha256, hashB, 20 );
-   SHA256_Update( &ctx_sha256, hashC, 20 );
-   SHA256_Final( (unsigned char*) hashA, &ctx_sha256 );
-
-   SHA256_Init( &ctx_sha256 );
-   SHA256_Update( &ctx_sha256, hashA, 32 );
-   SHA256_Final( (unsigned char*) hashA, &ctx_sha256 );
+   sha256_ctx_init( &ctx_sha256 );
+   sha256_update( &ctx_sha256, hashB, 20 );
+   sha256_update( &ctx_sha256, hashC, 20 );
+   sha256_final( &ctx_sha256, hashA );

+   sha256_full( hashA, hashA, 32 );
+   
   memcpy( output, hashA, 32 );
 }

--- a/algo/scrypt/neoscrypt.c
+++ b/algo/scrypt/neoscrypt.c
@@ -69,8 +69,12 @@ typedef unsigned int  uint;
 #define SCRYPT_HASH_BLOCK_SIZE 64U
 #define SCRYPT_HASH_DIGEST_SIZE 32U

-#define ROTL32(a,b) (((a) << (b)) | ((a) >> (32 - b)))
-#define ROTR32(a,b) (((a) >> (b)) | ((a) << (32 - b)))
+//#define ROTL32(a,b) (((a) << (b)) | ((a) >> (32 - b)))
+//#define ROTR32(a,b) (((a) >> (b)) | ((a) << (32 - b)))
+
+#define ROTL32(a,b) rol32(a,b)
+#define ROTR32(a,b) ror32(a,b)
+

 #define U8TO32_BE(p) \
    (((uint32_t)((p)[0]) << 24) | ((uint32_t)((p)[1]) << 16) | \
--- a/algo/scrypt/scrypt-core-4way.c
+++ b/algo/scrypt/scrypt-core-4way.c
--- a/algo/scrypt/scrypt-core-4way.h
+++ b/algo/scrypt/scrypt-core-4way.h
@@ -0,0 +1,70 @@
+#ifndef SCRYPT_CORE_4WAY_H__
+#define SCRYPT_CORE_4WAY_H__
+
+#include "simd-utils.h"
+#include <stdlib.h>
+#include <stdint.h>
+
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+
+void scrypt_core_16way( __m512i *X, __m512i *V, const uint32_t N );
+
+// Serial SIMD over 4 way parallel
+void scrypt_core_simd128_4way( __m128i *X, __m128i *V, const uint32_t N );
+
+// 4 way parallel over serial SIMD
+void scrypt_core_4way_simd128( __m512i *X, __m512i *V, const uint32_t N );
+
+#endif
+
+#if defined(__AVX2__)
+
+void scrypt_core_8way( __m256i *X, __m256i *V, uint32_t N );
+
+// 2 way parallel over SIMD128
+void scrypt_core_2way_simd128( __m256i *X, __m256i *V, const uint32_t N );
+
+// Double buffered 2 way parallel over SIMD128
+void scrypt_core_2way_simd128_2buf( __m256i *X, __m256i *V, const uint32_t N );
+
+// Triplee buffered 2 way parallel over SIMD128
+void scrypt_core_2way_simd128_3buf( __m256i *X, __m256i *V, const uint32_t N );
+
+// Serial SIMD128 over 2 way parallel
+void scrypt_core_simd128_2way( uint64_t *X, uint64_t *V, const uint32_t N );
+
+// Double buffered simd over parallel
+void scrypt_core_simd128_2way_2buf( uint64_t *X, uint64_t *V, const uint32_t N );
+
+// Triple buffered 2 way
+void scrypt_core_simd128_2way_3buf( uint64_t *X, uint64_t *V, const uint32_t N );
+
+// Quadruple buffered
+void scrypt_core_simd128_2way_4buf( uint64_t *X, uint64_t *V, const uint32_t N );
+
+#endif
+
+#if defined(__SSE2__)
+
+// Parallel 4 way, 4x memory
+void scrypt_core_4way( __m128i *X, __m128i *V, const uint32_t N );
+
+// Linear SIMD 1 way, 1x memory, lowest
+void scrypt_core_simd128( uint32_t *X, uint32_t *V, const uint32_t N );
+
+// Double buffered, 2x memory
+void scrypt_core_simd128_2buf( uint32_t *X, uint32_t *V, const uint32_t N );
+
+// Triple buffered
+void scrypt_core_simd128_3buf( uint32_t *X, uint32_t *V, const uint32_t N );
+
+// Quadruple buffered, 4x memory
+void scrypt_core_simd128_4buf( uint32_t *X, uint32_t *V, const uint32_t N );
+
+#endif
+
+// For reference only
+void scrypt_core_1way( uint32_t *X, uint32_t *V, const uint32_t N );
+
+#endif   
+
--- a/algo/scrypt/scrypt-core-ref.c
+++ b/algo/scrypt/scrypt-core-ref.c
@@ -0,0 +1,206 @@
+#include "scrypt-core-ref.h"
+
+#define ROTL(a, b) (((a) << (b)) | ((a) >> (32 - (b))))
+
+static void xor_salsa8(uint32_t * const B, const uint32_t * const C)
+{
+   uint32_t x0 = (B[ 0] ^= C[ 0]),
+            x1 = (B[ 1] ^= C[ 1]),
+            x2 = (B[ 2] ^= C[ 2]),
+            x3 = (B[ 3] ^= C[ 3]);
+   uint32_t x4 = (B[ 4] ^= C[ 4]),
+            x5 = (B[ 5] ^= C[ 5]),
+            x6 = (B[ 6] ^= C[ 6]),
+            x7 = (B[ 7] ^= C[ 7]);
+   uint32_t x8 = (B[ 8] ^= C[ 8]),
+            x9 = (B[ 9] ^= C[ 9]),
+            xa = (B[10] ^= C[10]),
+            xb = (B[11] ^= C[11]);
+   uint32_t xc = (B[12] ^= C[12]),
+            xd = (B[13] ^= C[13]),
+            xe = (B[14] ^= C[14]),
+            xf = (B[15] ^= C[15]);
+
+   /* Operate on columns. */
+   x4 ^= ROTL(x0 + xc,  7);
+   x9 ^= ROTL(x5 + x1,  7);
+   xe ^= ROTL(xa + x6,  7);
+   x3 ^= ROTL(xf + xb,  7);
+   x8 ^= ROTL(x4 + x0,  9);
+   xd ^= ROTL(x9 + x5,  9);
+   x2 ^= ROTL(xe + xa,  9);
+   x7 ^= ROTL(x3 + xf,  9);
+   xc ^= ROTL(x8 + x4, 13);
+   x1 ^= ROTL(xd + x9, 13);
+   x6 ^= ROTL(x2 + xe, 13);
+   xb ^= ROTL(x7 + x3, 13);
+   x0 ^= ROTL(xc + x8, 18);
+   x5 ^= ROTL(x1 + xd, 18);
+   xa ^= ROTL(x6 + x2, 18);
+   xf ^= ROTL(xb + x7, 18);
+
+   /* Operate on rows. */
+   x1 ^= ROTL(x0 + x3,  7);
+   x6 ^= ROTL(x5 + x4,  7);
+   xb ^= ROTL(xa + x9,  7);
+   xc ^= ROTL(xf + xe,  7);
+   x2 ^= ROTL(x1 + x0,  9);
+   x7 ^= ROTL(x6 + x5,  9);
+   x8 ^= ROTL(xb + xa,  9);
+   xd ^= ROTL(xc + xf,  9);
+   x3 ^= ROTL(x2 + x1, 13);
+   x4 ^= ROTL(x7 + x6, 13);
+   x9 ^= ROTL(x8 + xb, 13);
+   xe ^= ROTL(xd + xc, 13);
+   x0 ^= ROTL(x3 + x2, 18);
+   x5 ^= ROTL(x4 + x7, 18);
+   xa ^= ROTL(x9 + x8, 18);
+   xf ^= ROTL(xe + xd, 18);
+
+   /* Operate on columns. */
+   x4 ^= ROTL(x0 + xc,  7);
+   x9 ^= ROTL(x5 + x1,  7);
+   xe ^= ROTL(xa + x6,  7);
+   x3 ^= ROTL(xf + xb,  7);
+   x8 ^= ROTL(x4 + x0,  9);
+   xd ^= ROTL(x9 + x5,  9);
+   x2 ^= ROTL(xe + xa,  9);
+   x7 ^= ROTL(x3 + xf,  9);
+   xc ^= ROTL(x8 + x4, 13);
+   x1 ^= ROTL(xd + x9, 13);
+   x6 ^= ROTL(x2 + xe, 13);
+   xb ^= ROTL(x7 + x3, 13);
+   x0 ^= ROTL(xc + x8, 18);
+   x5 ^= ROTL(x1 + xd, 18);
+   xa ^= ROTL(x6 + x2, 18);
+   xf ^= ROTL(xb + x7, 18);
+
+   /* Operate on rows. */
+   x1 ^= ROTL(x0 + x3,  7);
+   x6 ^= ROTL(x5 + x4,  7);
+   xb ^= ROTL(xa + x9,  7);
+   xc ^= ROTL(xf + xe,  7);
+   x2 ^= ROTL(x1 + x0,  9);
+   x7 ^= ROTL(x6 + x5,  9);
+   x8 ^= ROTL(xb + xa,  9);
+   xd ^= ROTL(xc + xf,  9);
+   x3 ^= ROTL(x2 + x1, 13);
+   x4 ^= ROTL(x7 + x6, 13);
+   x9 ^= ROTL(x8 + xb, 13);
+   xe ^= ROTL(xd + xc, 13);
+   x0 ^= ROTL(x3 + x2, 18);
+   x5 ^= ROTL(x4 + x7, 18);
+   xa ^= ROTL(x9 + x8, 18);
+   xf ^= ROTL(xe + xd, 18);
+
+   /* Operate on columns. */
+   x4 ^= ROTL(x0 + xc,  7);
+   x9 ^= ROTL(x5 + x1,  7);
+   xe ^= ROTL(xa + x6,  7);
+   x3 ^= ROTL(xf + xb,  7);
+   x8 ^= ROTL(x4 + x0,  9);
+   xd ^= ROTL(x9 + x5,  9);
+   x2 ^= ROTL(xe + xa,  9);
+   x7 ^= ROTL(x3 + xf,  9);
+   xc ^= ROTL(x8 + x4, 13);
+   x1 ^= ROTL(xd + x9, 13);
+   x6 ^= ROTL(x2 + xe, 13);
+   xb ^= ROTL(x7 + x3, 13);
+   x0 ^= ROTL(xc + x8, 18);
+   x5 ^= ROTL(x1 + xd, 18);
+   xa ^= ROTL(x6 + x2, 18);
+   xf ^= ROTL(xb + x7, 18);
+
+   /* Operate on rows. */
+   x1 ^= ROTL(x0 + x3,  7);
+   x6 ^= ROTL(x5 + x4,  7);
+   xb ^= ROTL(xa + x9,  7);
+   xc ^= ROTL(xf + xe,  7);
+   x2 ^= ROTL(x1 + x0,  9);
+   x7 ^= ROTL(x6 + x5,  9);
+   x8 ^= ROTL(xb + xa,  9);
+   xd ^= ROTL(xc + xf,  9);
+   x3 ^= ROTL(x2 + x1, 13);
+   x4 ^= ROTL(x7 + x6, 13);
+   x9 ^= ROTL(x8 + xb, 13);
+   xe ^= ROTL(xd + xc, 13);
+   x0 ^= ROTL(x3 + x2, 18);
+   x5 ^= ROTL(x4 + x7, 18);
+   xa ^= ROTL(x9 + x8, 18);
+   xf ^= ROTL(xe + xd, 18);
+
+   /* Operate on columns. */
+   x4 ^= ROTL(x0 + xc,  7);
+   x9 ^= ROTL(x5 + x1,  7);
+   xe ^= ROTL(xa + x6,  7);
+   x3 ^= ROTL(xf + xb,  7);
+   x8 ^= ROTL(x4 + x0,  9);
+   xd ^= ROTL(x9 + x5,  9);
+   x2 ^= ROTL(xe + xa,  9);
+   x7 ^= ROTL(x3 + xf,  9);
+   xc ^= ROTL(x8 + x4, 13);
+   x1 ^= ROTL(xd + x9, 13);
+   x6 ^= ROTL(x2 + xe, 13);
+   xb ^= ROTL(x7 + x3, 13);
+   x0 ^= ROTL(xc + x8, 18);
+   x5 ^= ROTL(x1 + xd, 18);
+   xa ^= ROTL(x6 + x2, 18);
+   xf ^= ROTL(xb + x7, 18);
+
+   /* Operate on rows. */
+   x1 ^= ROTL(x0 + x3,  7);
+   x6 ^= ROTL(x5 + x4,  7);
+   xb ^= ROTL(xa + x9,  7);
+   xc ^= ROTL(xf + xe,  7);
+   x2 ^= ROTL(x1 + x0,  9);
+   x7 ^= ROTL(x6 + x5,  9);
+   x8 ^= ROTL(xb + xa,  9);
+   xd ^= ROTL(xc + xf,  9);
+   x3 ^= ROTL(x2 + x1, 13);
+   x4 ^= ROTL(x7 + x6, 13);
+   x9 ^= ROTL(x8 + xb, 13);
+   xe ^= ROTL(xd + xc, 13);
+   x0 ^= ROTL(x3 + x2, 18);
+   x5 ^= ROTL(x4 + x7, 18);
+   xa ^= ROTL(x9 + x8, 18);
+   xf ^= ROTL(xe + xd, 18);
+
+   B[ 0] += x0;
+   B[ 1] += x1;
+   B[ 2] += x2;
+   B[ 3] += x3;
+   B[ 4] += x4;
+   B[ 5] += x5;
+   B[ 6] += x6;
+   B[ 7] += x7;
+   B[ 8] += x8;
+   B[ 9] += x9;
+   B[10] += xa;
+   B[11] += xb;
+   B[12] += xc;
+   B[13] += xd;
+   B[14] += xe;
+   B[15] += xf;
+}
+
+/**
+ * @param X input/ouput
+ * @param V scratch buffer
+ * @param N factor (def. 1024)
+ */
+void scrypt_core_ref(uint32_t *X, uint32_t *V, uint32_t N)
+{
+   for (uint32_t i = 0; i < N; i++) {
+      memcpy(&V[i * 32], X, 128);
+      xor_salsa8(&X[0], &X[16]);
+      xor_salsa8(&X[16], &X[0]);
+   }
+   for (uint32_t i = 0; i < N; i++) {
+      uint32_t j = 32 * (X[16] & (N - 1));
+      for (uint8_t k = 0; k < 32; k++)
+         X[k] ^= V[j + k];
+      xor_salsa8(&X[0], &X[16]);
+      xor_salsa8(&X[16], &X[0]);
+   }
+}
+
--- a/algo/scrypt/scrypt.c
+++ b/algo/scrypt/scrypt.c
--- a/algo/sha/hmac-sha256-hash.c
+++ b/algo/sha/hmac-sha256-hash.c
@@ -39,10 +39,10 @@
 void
 SHA256_Buf( const void * in, size_t len, uint8_t digest[32] )
 {
-	SHA256_CTX ctx;
-   SHA256_Init( &ctx );
-   SHA256_Update( &ctx, in, len );
-   SHA256_Final( digest, &ctx );
+   sha256_context ctx;
+   sha256_ctx_init( &ctx );
+   sha256_update( &ctx, in, len );
+   sha256_final( &ctx, digest );
 }

 /**
@@ -64,35 +64,36 @@ HMAC_SHA256_Buf( const void *K, size_t Klen, const void *in, size_t len,
 void
 HMAC_SHA256_Init( HMAC_SHA256_CTX *ctx, const void *_K, size_t Klen )
 {
-	unsigned char pad[64];
-	unsigned char khash[32];
-	const unsigned char * K = _K;
-	size_t i;
+   unsigned char pad[64] __attribute__ ((aligned (64)));
+   unsigned char khash[32];
+   const unsigned char * K = _K;
+   size_t i;

-	/* If Klen > 64, the key is really SHA256(K). */
-	if ( Klen > 64 )
+   /* If Klen > 64, the key is really SHA256(K). */
+   if ( Klen > 64 )
   {
-		SHA256_Init( &ctx->ictx );
-		SHA256_Update( &ctx->ictx, K, Klen );
-		SHA256_Final( khash, &ctx->ictx );
-		K = khash;
-		Klen = 32;
-	}
-
-	/* Inner SHA256 operation is SHA256(K xor [block of 0x36] || data). */
-   SHA256_Init( &ctx->ictx );
+      sha256_ctx_init( &ctx->ictx );
+      sha256_update( &ctx->ictx, K, Klen );
+      sha256_final( &ctx->ictx, khash );
+      K = khash;
+      Klen = 32;
+   }

+   /* Inner SHA256 operation is SHA256(K xor [block of 0x36] || data). */
+   sha256_ctx_init( &ctx->ictx );

   for ( i = 0; i < Klen; i++ )  pad[i] = K[i] ^ 0x36;
-   memset( pad + Klen, 0x36, 64 - Klen );
-	SHA256_Update( &ctx->ictx, pad, 64 );

-	/* Outer SHA256 operation is SHA256(K xor [block of 0x5c] || hash). */
-	SHA256_Init( &ctx->octx );
+   memset( pad + Klen, 0x36, 64 - Klen );
+   sha256_update( &ctx->ictx, pad, 64 );
+
+   /* Outer SHA256 operation is SHA256(K xor [block of 0x5c] || hash). */
+   sha256_ctx_init( &ctx->octx );

   for ( i = 0; i < Klen; i++ )  pad[i] = K[i] ^ 0x5c;
+
   memset( pad + Klen, 0x5c, 64 - Klen );
-	SHA256_Update( &ctx->octx, pad, 64 );
+   sha256_update( &ctx->octx, pad, 64 );
 }

 /* Add bytes to the HMAC-SHA256 operation. */
@@ -100,23 +101,17 @@ void
 HMAC_SHA256_Update( HMAC_SHA256_CTX *ctx, const void *in, size_t len )
 {
 	/* Feed data to the inner SHA256 operation. */
-	SHA256_Update( &ctx->ictx, in, len );
+   sha256_update( &ctx->ictx, in, len );
 }

 /* Finish an HMAC-SHA256 operation. */
 void
-HMAC_SHA256_Final( unsigned char digest[32], HMAC_SHA256_CTX *ctx )
+HMAC_SHA256_Final( void *digest, HMAC_SHA256_CTX *ctx )
 {
-	unsigned char ihash[32];
-
-	/* Finish the inner SHA256 operation. */
-	SHA256_Final( ihash, &ctx->ictx );
-
-	/* Feed the inner hash to the outer SHA256 operation. */
-	SHA256_Update( &ctx->octx, ihash, 32 );
-
-	/* Finish the outer SHA256 operation. */
-	SHA256_Final( digest, &ctx->octx );
+   uint32_t ihash[8] __attribute__ ((aligned (32)));
+   sha256_final( &ctx->ictx, ihash );
+   sha256_update( &ctx->octx, ihash, 32 );
+   sha256_final( &ctx->octx, digest );
 }

 /**
@@ -129,8 +124,10 @@ PBKDF2_SHA256( const uint8_t *passwd, size_t passwdlen, const uint8_t *salt,
               size_t saltlen, uint64_t c, uint8_t *buf, size_t dkLen )
 {
 	HMAC_SHA256_CTX PShctx, hctx;
-	uint8_t _ALIGN(128) T[32];
-	uint8_t _ALIGN(128) U[32];
+   uint64_t _ALIGN(128) T[4];
+   uint64_t _ALIGN(128) U[4];
+//   uint8_t _ALIGN(128) T[32];
+//	uint8_t _ALIGN(128) U[32];
   uint32_t ivec;
 	size_t i, clen;
 	uint64_t j;
@@ -166,10 +163,10 @@ PBKDF2_SHA256( const uint8_t *passwd, size_t passwdlen, const uint8_t *salt,
 //         _mm_xor_si128( ((__m128i*)T)[0], ((__m128i*)U)[0] );
 //         _mm_xor_si128( ((__m128i*)T)[1], ((__m128i*)U)[1] );

-//         for ( k = 0; k < 4; k++ )  T[k] ^= U[k];
+         for ( k = 0; k < 4; k++ )  T[k] ^= U[k];
         
-         for ( k = 0; k < 32; k++ )
-				T[k] ^= U[k];
+//         for ( k = 0; k < 32; k++ )
+//				T[k] ^= U[k];
 		}

 		/* Copy as many bytes as necessary into buf. */
--- a/algo/sha/hmac-sha256-hash.h
+++ b/algo/sha/hmac-sha256-hash.h
@@ -31,18 +31,18 @@

 #include <sys/types.h>
 #include <stdint.h>
-#include <openssl/sha.h>
+#include "sha256-hash.h"

 typedef struct HMAC_SHA256Context
 {
-     SHA256_CTX ictx;
-     SHA256_CTX octx;
+   sha256_context ictx;
+   sha256_context octx;
 } HMAC_SHA256_CTX;

 void SHA256_Buf( const void *, size_t len, uint8_t digest[32] );
 void HMAC_SHA256_Init( HMAC_SHA256_CTX *, const void *, size_t );
 void HMAC_SHA256_Update( HMAC_SHA256_CTX *, const void *, size_t );
-void HMAC_SHA256_Final( unsigned char [32], HMAC_SHA256_CTX * );
+void HMAC_SHA256_Final( void*, HMAC_SHA256_CTX * );
 void HMAC_SHA256_Buf( const void *, size_t Klen, const void *,
                      size_t len, uint8_t digest[32] );

--- a/algo/sha/sha-hash-4way.h
+++ b/algo/sha/sha-hash-4way.h
@@ -51,7 +51,6 @@ typedef struct {
   __m128i buf[64>>2];
   __m128i val[8];
   uint32_t count_high, count_low;
-   bool initialized;
 } sha256_4way_context __attribute__ ((aligned (64)));

 void sha256_4way_init( sha256_4way_context *sc );
@@ -59,6 +58,10 @@ void sha256_4way_update( sha256_4way_context *sc, const void *data,
                         size_t len );
 void sha256_4way_close( sha256_4way_context *sc, void *dst );
 void sha256_4way_full( void *dst, const void *data, size_t len );
+void sha256_4way_transform_le( __m128i *state_out,  const __m128i *data,
+                            const __m128i *state_in );
+void sha256_4way_transform_be( __m128i *state_out,  const __m128i *data,
+                            const __m128i *state_in );

 #endif  // SSE2

@@ -70,13 +73,16 @@ typedef struct {
   __m256i buf[64>>2];
   __m256i val[8];
   uint32_t count_high, count_low;
-   bool initialized;
 } sha256_8way_context __attribute__ ((aligned (128)));

 void sha256_8way_init( sha256_8way_context *sc );
 void sha256_8way_update( sha256_8way_context *sc, const void *data, size_t len );
 void sha256_8way_close( sha256_8way_context *sc, void *dst );
 void sha256_8way_full( void *dst, const void *data, size_t len );
+void sha256_8way_transform_le( __m256i *state_out, const __m256i *data,
+                               const __m256i *state_in );
+void sha256_8way_transform_be( __m256i *state_out, const __m256i *data,
+                               const __m256i *state_in );

 #endif  // AVX2

@@ -88,13 +94,20 @@ typedef struct {
   __m512i buf[64>>2];
   __m512i val[8];
   uint32_t count_high, count_low;
-   bool initialized;
 } sha256_16way_context __attribute__ ((aligned (128)));

 void sha256_16way_init( sha256_16way_context *sc );
 void sha256_16way_update( sha256_16way_context *sc, const void *data, size_t len );
 void sha256_16way_close( sha256_16way_context *sc, void *dst );
 void sha256_16way_full( void *dst, const void *data, size_t len );
+void sha256_16way_transform_le( __m512i *state_out, const __m512i *data,
+                             const __m512i *state_in );
+void sha256_16way_transform_be( __m512i *state_out, const __m512i *data,
+                             const __m512i *state_in );
+void sha256_16way_prehash_3rounds( __m512i *state_mid, const __m512i *W,
+                             const __m512i *state_in );
+void sha256_16way_final_rounds( __m512i *state_out, const __m512i *data,
+                          const __m512i *state_in, const __m512i *state_mid );

 #endif // AVX512

--- a/algo/sha/sha2.c
+++ b/algo/sha/sha2.c
@@ -12,7 +12,6 @@

 #include <string.h>
 #include <inttypes.h>
-#include <openssl/sha.h>

 #if defined(USE_ASM) && defined(__arm__) && defined(__APCS_32__)
 #define EXTERN_SHA256
@@ -181,6 +180,7 @@ static const uint32_t sha256d_hash1[16] = {
 	0x00000000, 0x00000000, 0x00000000, 0x00000100
 };

+// this performs the entire hash all over again, why?
 static void sha256d_80_swap(uint32_t *hash, const uint32_t *data)
 {
 	uint32_t S[16];
@@ -196,18 +196,29 @@ static void sha256d_80_swap(uint32_t *hash, const uint32_t *data)
 		hash[i] = swab32(hash[i]);
 }

-extern void sha256d(unsigned char *hash, const unsigned char *data, int len)
+/*
+#if defined (__SHA__)
+
+#include "algo/sha/sph_sha2.h"
+
+void sha256d(unsigned char *hash, const unsigned char *data, int len)
 {
-#if defined(__SHA__)
-   SHA256_CTX ctx;
-   SHA256_Init( &ctx );
-   SHA256_Update( &ctx, data, len );
-   SHA256_Final( (unsigned char*)hash, &ctx );
-   SHA256_Init( &ctx );
-   SHA256_Update( &ctx, hash, 32 );
-   SHA256_Final( (unsigned char*)hash, &ctx );
+   sph_sha256_context ctx __attribute__ ((aligned (64)));
+
+   sph_sha256_init( &ctx );
+   sph_sha256( &ctx, data, len );
+   sph_sha256_close( &ctx, hash );
+
+   sph_sha256_init( &ctx );
+   sph_sha256( &ctx, hash, 32 );
+   sph_sha256_close( &ctx, hash );
+}
+
 #else

+void sha256d(unsigned char *hash, const unsigned char *data, int len)
+{
+
   uint32_t S[16], T[16];
 	int i, r;

@@ -229,9 +240,11 @@ extern void sha256d(unsigned char *hash, const unsigned char *data, int len)
 	sha256_transform(T, S, 0);
 	for (i = 0; i < 8; i++)
 		be32enc((uint32_t *)hash + i, T[i]);
-#endif
 }

+#endif
+*/
+
 static inline void sha256d_preextend(uint32_t *W)
 {
 	W[16] = s1(W[14]) + W[ 9] + s0(W[ 1]) + W[ 0];
@@ -643,6 +656,7 @@ int scanhash_sha256d( struct work *work,
 	return 0;
 }

+/*
 int scanhash_SHA256d( struct work *work, const uint32_t max_nonce,
                      uint64_t *hashes_done, struct thr_info *mythr )
 {
@@ -672,18 +686,13 @@ int scanhash_SHA256d( struct work *work, const uint32_t max_nonce,
   pdata[19] = n;
   return 0;
 }
-
+*/

 bool register_sha256d_algo( algo_gate_t* gate )
 {
-#if defined(__SHA__)
-   gate->optimizations = SHA_OPT;
-   gate->scanhash = (void*)&scanhash_SHA256d;
-#else
   gate->optimizations = SSE2_OPT | AVX2_OPT;
   gate->scanhash = (void*)&scanhash_sha256d;
-#endif
-    gate->hash     = (void*)&sha256d;
-    return true;
+//   gate->hash     = (void*)&sha256d;
+   return true;
 };

--- a/algo/sha/sha256-hash-2way-ni.c
+++ b/algo/sha/sha256-hash-2way-ni.c
@@ -0,0 +1,689 @@
+/*   Intel SHA extensions using C intrinsics               */
+/*   Written and place in public domain by Jeffrey Walton  */
+/*   Based on code from Intel, and by Sean Gulley for      */
+/*   the miTLS project.                                    */
+
+// A stripped down version with byte swapping removed. 
+
+#if defined(__SHA__)
+
+#include "sha256-hash.h"
+
+void sha256_ni2way_transform_le( uint32_t *out_X, uint32_t*out_Y,
+                              const void *msg_X, const void *msg_Y,
+                              const uint32_t *in_X, const uint32_t *in_Y )
+{
+    __m128i STATE0_X, STATE1_X, STATE0_Y, STATE1_Y;
+    __m128i MSG_X, MSG_Y, TMP_X, TMP_Y;
+    __m128i TMSG0_X, TMSG1_X, TMSG2_X, TMSG3_X;
+    __m128i TMSG0_Y, TMSG1_Y, TMSG2_Y, TMSG3_Y;
+    __m128i ABEF_SAVE_X, CDGH_SAVE_X,ABEF_SAVE_Y, CDGH_SAVE_Y;
+
+    // Load initial values
+    TMP_X = _mm_load_si128((__m128i*) &in_X[0]);
+    STATE1_X = _mm_load_si128((__m128i*) &in_X[4]);
+    TMP_Y = _mm_load_si128((__m128i*) &in_Y[0]);
+    STATE1_Y = _mm_load_si128((__m128i*) &in_Y[4]);
+
+    TMP_X = _mm_shuffle_epi32(TMP_X, 0xB1); // CDAB
+    TMP_Y = _mm_shuffle_epi32(TMP_Y, 0xB1); // CDAB
+    STATE1_X = _mm_shuffle_epi32(STATE1_X, 0x1B); // EFGH
+    STATE1_Y = _mm_shuffle_epi32(STATE1_Y, 0x1B); // EFGH
+    STATE0_X = _mm_alignr_epi8(TMP_X, STATE1_X, 8); // ABEF
+    STATE0_Y = _mm_alignr_epi8(TMP_Y, STATE1_Y, 8); // ABEF
+    STATE1_X = _mm_blend_epi16(STATE1_X, TMP_X, 0xF0); // CDGH
+    STATE1_Y = _mm_blend_epi16(STATE1_Y, TMP_Y, 0xF0); // CDGH
+
+    // Save current hash
+    ABEF_SAVE_X = STATE0_X;
+    ABEF_SAVE_Y = STATE0_Y;
+    CDGH_SAVE_X = STATE1_X;
+    CDGH_SAVE_Y = STATE1_Y;
+
+    // Rounds 0-3
+    TMSG0_X = _mm_load_si128((const __m128i*) (msg_X));
+    TMSG0_Y = _mm_load_si128((const __m128i*) (msg_Y));
+    TMP_X = _mm_set_epi64x(0xE9B5DBA5B5C0FBCFULL, 0x71374491428A2F98ULL);
+    MSG_X = _mm_add_epi32( TMSG0_X, TMP_X );
+    MSG_Y = _mm_add_epi32( TMSG0_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+
+    // Rounds 4-7
+    TMSG1_X = _mm_load_si128((const __m128i*) (msg_X+16));
+    TMSG1_Y = _mm_load_si128((const __m128i*) (msg_Y+16));
+    TMP_X = _mm_set_epi64x(0xAB1C5ED5923F82A4ULL, 0x59F111F13956C25BULL);
+    MSG_X = _mm_add_epi32(TMSG1_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG1_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG0_X = _mm_sha256msg1_epu32(TMSG0_X, TMSG1_X);
+    TMSG0_Y = _mm_sha256msg1_epu32(TMSG0_Y, TMSG1_Y);
+
+    // Rounds 8-11
+    TMSG2_X = _mm_load_si128((const __m128i*) (msg_X+32));
+    TMSG2_Y = _mm_load_si128((const __m128i*) (msg_Y+32));
+    TMP_X = _mm_set_epi64x(0x550C7DC3243185BEULL, 0x12835B01D807AA98ULL);
+    MSG_X = _mm_add_epi32(TMSG2_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG2_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG1_X = _mm_sha256msg1_epu32(TMSG1_X, TMSG2_X);
+    TMSG1_Y = _mm_sha256msg1_epu32(TMSG1_Y, TMSG2_Y);
+
+    // Rounds 12-15
+    TMSG3_X = _mm_load_si128((const __m128i*) (msg_X+48));
+    TMSG3_Y = _mm_load_si128((const __m128i*) (msg_Y+48));
+    TMP_X = _mm_set_epi64x(0xC19BF1749BDC06A7ULL, 0x80DEB1FE72BE5D74ULL);
+    MSG_X = _mm_add_epi32(TMSG3_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG3_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG3_X, TMSG2_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG3_Y, TMSG2_Y, 4);
+    TMSG0_X = _mm_add_epi32(TMSG0_X, TMP_X);
+    TMSG0_Y = _mm_add_epi32(TMSG0_Y, TMP_Y);
+    TMSG0_X = _mm_sha256msg2_epu32(TMSG0_X, TMSG3_X);
+    TMSG0_Y = _mm_sha256msg2_epu32(TMSG0_Y, TMSG3_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG2_X = _mm_sha256msg1_epu32(TMSG2_X, TMSG3_X);
+    TMSG2_Y = _mm_sha256msg1_epu32(TMSG2_Y, TMSG3_Y);
+
+    // Rounds 16-19
+    TMP_X = _mm_set_epi64x(0x240CA1CC0FC19DC6ULL, 0xEFBE4786E49B69C1ULL);
+    MSG_X = _mm_add_epi32(TMSG0_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG0_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG0_X, TMSG3_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG0_Y, TMSG3_Y, 4);
+    TMSG1_X = _mm_add_epi32(TMSG1_X, TMP_X);
+    TMSG1_Y = _mm_add_epi32(TMSG1_Y, TMP_Y);
+    TMSG1_X = _mm_sha256msg2_epu32(TMSG1_X, TMSG0_X);
+    TMSG1_Y = _mm_sha256msg2_epu32(TMSG1_Y, TMSG0_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG3_X = _mm_sha256msg1_epu32(TMSG3_X, TMSG0_X);
+    TMSG3_Y = _mm_sha256msg1_epu32(TMSG3_Y, TMSG0_Y);
+
+    // Rounds 20-23
+    TMP_X = _mm_set_epi64x(0x76F988DA5CB0A9DCULL, 0x4A7484AA2DE92C6FULL);
+    MSG_X = _mm_add_epi32(TMSG1_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG1_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG1_X, TMSG0_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG1_Y, TMSG0_Y, 4);
+    TMSG2_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    TMSG2_Y = _mm_add_epi32(TMSG2_Y, TMP_Y);
+    TMSG2_X = _mm_sha256msg2_epu32(TMSG2_X, TMSG1_X);
+    TMSG2_Y = _mm_sha256msg2_epu32(TMSG2_Y, TMSG1_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG0_X = _mm_sha256msg1_epu32(TMSG0_X, TMSG1_X);
+    TMSG0_Y = _mm_sha256msg1_epu32(TMSG0_Y, TMSG1_Y);
+
+    // Rounds 24-27
+    TMP_X = _mm_set_epi64x(0xBF597FC7B00327C8ULL, 0xA831C66D983E5152ULL);
+    MSG_X = _mm_add_epi32(TMSG2_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG2_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG2_X, TMSG1_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG2_Y, TMSG1_Y, 4);
+    TMSG3_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    TMSG3_Y = _mm_add_epi32(TMSG3_Y, TMP_Y);
+    TMSG3_X = _mm_sha256msg2_epu32(TMSG3_X, TMSG2_X);
+    TMSG3_Y = _mm_sha256msg2_epu32(TMSG3_Y, TMSG2_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG1_X = _mm_sha256msg1_epu32(TMSG1_X, TMSG2_X);
+    TMSG1_Y = _mm_sha256msg1_epu32(TMSG1_Y, TMSG2_Y);
+
+    // Rounds 28-31
+    TMP_X = _mm_set_epi64x(0x1429296706CA6351ULL,  0xD5A79147C6E00BF3ULL);
+    MSG_X = _mm_add_epi32(TMSG3_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG3_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG3_X, TMSG2_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG3_Y, TMSG2_Y, 4);
+    TMSG0_X = _mm_add_epi32(TMSG0_X, TMP_X);
+    TMSG0_Y = _mm_add_epi32(TMSG0_Y, TMP_Y);
+    TMSG0_X = _mm_sha256msg2_epu32(TMSG0_X, TMSG3_X);
+    TMSG0_Y = _mm_sha256msg2_epu32(TMSG0_Y, TMSG3_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG2_X = _mm_sha256msg1_epu32(TMSG2_X, TMSG3_X);
+    TMSG2_Y = _mm_sha256msg1_epu32(TMSG2_Y, TMSG3_Y);
+
+    // Rounds 32-35
+    TMP_X = _mm_set_epi64x(0x53380D134D2C6DFCULL, 0x2E1B213827B70A85ULL);
+    MSG_X = _mm_add_epi32(TMSG0_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG0_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG0_X, TMSG3_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG0_Y, TMSG3_Y, 4);
+    TMSG1_X = _mm_add_epi32(TMSG1_X, TMP_X);
+    TMSG1_Y = _mm_add_epi32(TMSG1_Y, TMP_Y);
+    TMSG1_X = _mm_sha256msg2_epu32(TMSG1_X, TMSG0_X);
+    TMSG1_Y = _mm_sha256msg2_epu32(TMSG1_Y, TMSG0_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG3_X = _mm_sha256msg1_epu32(TMSG3_X, TMSG0_X);
+    TMSG3_Y = _mm_sha256msg1_epu32(TMSG3_Y, TMSG0_Y);
+
+    // Rounds 36-39
+    TMP_X = _mm_set_epi64x(0x92722C8581C2C92EULL, 0x766A0ABB650A7354ULL);
+    MSG_X = _mm_add_epi32(TMSG1_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG1_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG1_X, TMSG0_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG1_Y, TMSG0_Y, 4);
+    TMSG2_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    TMSG2_Y = _mm_add_epi32(TMSG2_Y, TMP_Y);
+    TMSG2_X = _mm_sha256msg2_epu32(TMSG2_X, TMSG1_X);
+    TMSG2_Y = _mm_sha256msg2_epu32(TMSG2_Y, TMSG1_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG0_X = _mm_sha256msg1_epu32(TMSG0_X, TMSG1_X);
+    TMSG0_Y = _mm_sha256msg1_epu32(TMSG0_Y, TMSG1_Y);
+
+    // Rounds 40-43
+    TMP_X = _mm_set_epi64x(0xC76C51A3C24B8B70ULL, 0xA81A664BA2BFE8A1ULL);
+    MSG_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG2_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG2_X, TMSG1_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG2_Y, TMSG1_Y, 4);
+    TMSG3_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    TMSG3_Y = _mm_add_epi32(TMSG3_Y, TMP_Y);
+    TMSG3_X = _mm_sha256msg2_epu32(TMSG3_X, TMSG2_X);
+    TMSG3_Y = _mm_sha256msg2_epu32(TMSG3_Y, TMSG2_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG1_X = _mm_sha256msg1_epu32(TMSG1_X, TMSG2_X);
+    TMSG1_Y = _mm_sha256msg1_epu32(TMSG1_Y, TMSG2_Y);
+
+    // Rounds 44-47
+    TMP_X = _mm_set_epi64x(0x106AA070F40E3585ULL, 0xD6990624D192E819ULL);
+    MSG_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG3_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG3_X, TMSG2_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG3_Y, TMSG2_Y, 4);
+    TMSG0_X = _mm_add_epi32(TMSG0_X, TMP_X);
+    TMSG0_Y = _mm_add_epi32(TMSG0_Y, TMP_Y);
+    TMSG0_X = _mm_sha256msg2_epu32(TMSG0_X, TMSG3_X);
+    TMSG0_Y = _mm_sha256msg2_epu32(TMSG0_Y, TMSG3_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG2_X = _mm_sha256msg1_epu32(TMSG2_X, TMSG3_X);
+    TMSG2_Y = _mm_sha256msg1_epu32(TMSG2_Y, TMSG3_Y);
+
+    // Rounds 48-51
+    TMP_X = _mm_set_epi64x(0x34B0BCB52748774CULL, 0x1E376C0819A4C116ULL);
+    MSG_X = _mm_add_epi32(TMSG0_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG0_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG0_X, TMSG3_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG0_Y, TMSG3_Y, 4);
+    TMSG1_X = _mm_add_epi32(TMSG1_X, TMP_X);
+    TMSG1_Y = _mm_add_epi32(TMSG1_Y, TMP_Y);
+    TMSG1_X = _mm_sha256msg2_epu32(TMSG1_X, TMSG0_X);
+    TMSG1_Y = _mm_sha256msg2_epu32(TMSG1_Y, TMSG0_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG3_X = _mm_sha256msg1_epu32(TMSG3_X, TMSG0_X);
+    TMSG3_Y = _mm_sha256msg1_epu32(TMSG3_Y, TMSG0_Y);
+
+    // Rounds 52-55
+    TMP_X = _mm_set_epi64x(0x682E6FF35B9CCA4FULL, 0x4ED8AA4A391C0CB3ULL);
+    MSG_X = _mm_add_epi32(TMSG1_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG1_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG1_X, TMSG0_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG1_Y, TMSG0_Y, 4);
+    TMSG2_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    TMSG2_Y = _mm_add_epi32(TMSG2_Y, TMP_Y);
+    TMSG2_X = _mm_sha256msg2_epu32(TMSG2_X, TMSG1_X);
+    TMSG2_Y = _mm_sha256msg2_epu32(TMSG2_Y, TMSG1_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+
+    // Rounds 56-59
+    TMP_X = _mm_set_epi64x(0x8CC7020884C87814ULL, 0x78A5636F748F82EEULL);
+    MSG_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG2_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG2_X, TMSG1_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG2_Y, TMSG1_Y, 4);
+    TMSG3_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    TMSG3_Y = _mm_add_epi32(TMSG3_Y, TMP_Y);
+    TMSG3_X = _mm_sha256msg2_epu32(TMSG3_X, TMSG2_X);
+    TMSG3_Y = _mm_sha256msg2_epu32(TMSG3_Y, TMSG2_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+
+    // Rounds 60-63
+    TMP_X = _mm_set_epi64x(0xC67178F2BEF9A3F7ULL, 0xA4506CEB90BEFFFAULL);
+    MSG_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG3_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+
+    // Add values back to state
+    STATE0_X = _mm_add_epi32(STATE0_X, ABEF_SAVE_X);
+    STATE1_X = _mm_add_epi32(STATE1_X, CDGH_SAVE_X);
+    STATE0_Y = _mm_add_epi32(STATE0_Y, ABEF_SAVE_Y);
+    STATE1_Y = _mm_add_epi32(STATE1_Y, CDGH_SAVE_Y);
+
+    TMP_X = _mm_shuffle_epi32(STATE0_X, 0x1B); // FEBA
+    TMP_Y = _mm_shuffle_epi32(STATE0_Y, 0x1B); // FEBA
+    STATE1_X = _mm_shuffle_epi32(STATE1_X, 0xB1); // DCHG
+    STATE1_Y = _mm_shuffle_epi32(STATE1_Y, 0xB1); // DCHG
+    STATE0_X = _mm_blend_epi16(TMP_X, STATE1_X, 0xF0); // DCBA
+    STATE0_Y = _mm_blend_epi16(TMP_Y, STATE1_Y, 0xF0); // DCBA
+    STATE1_X = _mm_alignr_epi8(STATE1_X, TMP_X, 8); // ABEF
+    STATE1_Y = _mm_alignr_epi8(STATE1_Y, TMP_Y, 8); // ABEF
+
+    // Save state
+    _mm_store_si128((__m128i*) &out_X[0], STATE0_X);
+    _mm_store_si128((__m128i*) &out_X[4], STATE1_X);
+    _mm_store_si128((__m128i*) &out_Y[0], STATE0_Y);
+    _mm_store_si128((__m128i*) &out_Y[4], STATE1_Y);
+}
+
+void sha256_ni2way_transform_be( uint32_t *out_X, uint32_t*out_Y,
+                              const void *msg_X, const void *msg_Y,
+                              const uint32_t *in_X, const uint32_t *in_Y )
+{
+    __m128i STATE0_X, STATE1_X, STATE0_Y, STATE1_Y;
+    __m128i MSG_X, MSG_Y, TMP_X, TMP_Y, MASK;
+    __m128i TMSG0_X, TMSG1_X, TMSG2_X, TMSG3_X;
+    __m128i TMSG0_Y, TMSG1_Y, TMSG2_Y, TMSG3_Y;
+    __m128i ABEF_SAVE_X, CDGH_SAVE_X, ABEF_SAVE_Y, CDGH_SAVE_Y;
+
+    // Load initial values
+    TMP_X = _mm_load_si128((__m128i*) &in_X[0]);
+    STATE1_X = _mm_load_si128((__m128i*) &in_X[4]);
+    TMP_Y = _mm_load_si128((__m128i*) &in_Y[0]);
+    STATE1_Y = _mm_load_si128((__m128i*) &in_Y[4]);
+    MASK = _mm_set_epi64x(0x0c0d0e0f08090a0bULL, 0x0405060700010203ULL);
+
+    TMP_X = _mm_shuffle_epi32(TMP_X, 0xB1); // CDAB
+    TMP_Y = _mm_shuffle_epi32(TMP_Y, 0xB1); // CDAB
+    STATE1_X = _mm_shuffle_epi32(STATE1_X, 0x1B); // EFGH
+    STATE1_Y = _mm_shuffle_epi32(STATE1_Y, 0x1B); // EFGH
+    STATE0_X = _mm_alignr_epi8(TMP_X, STATE1_X, 8); // ABEF
+    STATE0_Y = _mm_alignr_epi8(TMP_Y, STATE1_Y, 8); // ABEF
+    STATE1_X = _mm_blend_epi16(STATE1_X, TMP_X, 0xF0); // CDGH
+    STATE1_Y = _mm_blend_epi16(STATE1_Y, TMP_Y, 0xF0); // CDGH
+
+    // Save current hash
+    ABEF_SAVE_X = STATE0_X;
+    ABEF_SAVE_Y = STATE0_Y;
+    CDGH_SAVE_X = STATE1_X;
+    CDGH_SAVE_Y = STATE1_Y;
+
+    // Rounds 0-3
+    TMSG0_X = _mm_load_si128((const __m128i*) (msg_X));
+    TMSG0_Y = _mm_load_si128((const __m128i*) (msg_Y));
+    TMP_X = _mm_set_epi64x(0xE9B5DBA5B5C0FBCFULL, 0x71374491428A2F98ULL);
+    TMSG0_X = _mm_shuffle_epi8( TMSG0_X, MASK );
+    TMSG0_Y = _mm_shuffle_epi8( TMSG0_Y, MASK );
+    MSG_X = _mm_add_epi32( TMSG0_X, TMP_X );
+    MSG_Y = _mm_add_epi32( TMSG0_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+
+    // Rounds 4-7
+    TMSG1_X = _mm_load_si128((const __m128i*) (msg_X+16));
+    TMSG1_Y = _mm_load_si128((const __m128i*) (msg_Y+16));
+    TMP_X = _mm_set_epi64x(0xAB1C5ED5923F82A4ULL, 0x59F111F13956C25BULL);
+    TMSG1_X = _mm_shuffle_epi8( TMSG1_X, MASK );
+    TMSG1_Y = _mm_shuffle_epi8( TMSG1_Y, MASK );
+    MSG_X = _mm_add_epi32(TMSG1_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG1_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG0_X = _mm_sha256msg1_epu32(TMSG0_X, TMSG1_X);
+    TMSG0_Y = _mm_sha256msg1_epu32(TMSG0_Y, TMSG1_Y);
+
+    // Rounds 8-11
+    TMSG2_X = _mm_load_si128((const __m128i*) (msg_X+32));
+    TMSG2_Y = _mm_load_si128((const __m128i*) (msg_Y+32));
+    TMP_X = _mm_set_epi64x(0x550C7DC3243185BEULL, 0x12835B01D807AA98ULL);
+    TMSG2_X = _mm_shuffle_epi8( TMSG2_X, MASK );
+    TMSG2_Y = _mm_shuffle_epi8( TMSG2_Y, MASK );
+    MSG_X = _mm_add_epi32(TMSG2_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG2_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG1_X = _mm_sha256msg1_epu32(TMSG1_X, TMSG2_X);
+    TMSG1_Y = _mm_sha256msg1_epu32(TMSG1_Y, TMSG2_Y);
+
+    // Rounds 12-15
+    TMSG3_X = _mm_load_si128((const __m128i*) (msg_X+48));
+    TMSG3_Y = _mm_load_si128((const __m128i*) (msg_Y+48));
+    TMP_X = _mm_set_epi64x(0xC19BF1749BDC06A7ULL, 0x80DEB1FE72BE5D74ULL);
+    TMSG3_X = _mm_shuffle_epi8( TMSG3_X, MASK );
+    TMSG3_Y = _mm_shuffle_epi8( TMSG3_Y, MASK );
+    MSG_X = _mm_add_epi32(TMSG3_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG3_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG3_X, TMSG2_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG3_Y, TMSG2_Y, 4);
+    TMSG0_X = _mm_add_epi32(TMSG0_X, TMP_X);
+    TMSG0_Y = _mm_add_epi32(TMSG0_Y, TMP_Y);
+    TMSG0_X = _mm_sha256msg2_epu32(TMSG0_X, TMSG3_X);
+    TMSG0_Y = _mm_sha256msg2_epu32(TMSG0_Y, TMSG3_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG2_X = _mm_sha256msg1_epu32(TMSG2_X, TMSG3_X);
+    TMSG2_Y = _mm_sha256msg1_epu32(TMSG2_Y, TMSG3_Y);
+
+    // Rounds 16-19
+    TMP_X = _mm_set_epi64x(0x240CA1CC0FC19DC6ULL, 0xEFBE4786E49B69C1ULL);
+    MSG_X = _mm_add_epi32(TMSG0_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG0_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG0_X, TMSG3_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG0_Y, TMSG3_Y, 4);
+    TMSG1_X = _mm_add_epi32(TMSG1_X, TMP_X);
+    TMSG1_Y = _mm_add_epi32(TMSG1_Y, TMP_Y);
+    TMSG1_X = _mm_sha256msg2_epu32(TMSG1_X, TMSG0_X);
+    TMSG1_Y = _mm_sha256msg2_epu32(TMSG1_Y, TMSG0_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG3_X = _mm_sha256msg1_epu32(TMSG3_X, TMSG0_X);
+    TMSG3_Y = _mm_sha256msg1_epu32(TMSG3_Y, TMSG0_Y);
+
+    // Rounds 20-23
+    TMP_X = _mm_set_epi64x(0x76F988DA5CB0A9DCULL, 0x4A7484AA2DE92C6FULL);
+    MSG_X = _mm_add_epi32(TMSG1_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG1_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG1_X, TMSG0_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG1_Y, TMSG0_Y, 4);
+    TMSG2_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    TMSG2_Y = _mm_add_epi32(TMSG2_Y, TMP_Y);
+    TMSG2_X = _mm_sha256msg2_epu32(TMSG2_X, TMSG1_X);
+    TMSG2_Y = _mm_sha256msg2_epu32(TMSG2_Y, TMSG1_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG0_X = _mm_sha256msg1_epu32(TMSG0_X, TMSG1_X);
+    TMSG0_Y = _mm_sha256msg1_epu32(TMSG0_Y, TMSG1_Y);
+
+    // Rounds 24-27
+    TMP_X = _mm_set_epi64x(0xBF597FC7B00327C8ULL, 0xA831C66D983E5152ULL);
+    MSG_X = _mm_add_epi32(TMSG2_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG2_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG2_X, TMSG1_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG2_Y, TMSG1_Y, 4);
+    TMSG3_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    TMSG3_Y = _mm_add_epi32(TMSG3_Y, TMP_Y);
+    TMSG3_X = _mm_sha256msg2_epu32(TMSG3_X, TMSG2_X);
+    TMSG3_Y = _mm_sha256msg2_epu32(TMSG3_Y, TMSG2_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG1_X = _mm_sha256msg1_epu32(TMSG1_X, TMSG2_X);
+    TMSG1_Y = _mm_sha256msg1_epu32(TMSG1_Y, TMSG2_Y);
+
+    // Rounds 28-31
+    TMP_X = _mm_set_epi64x(0x1429296706CA6351ULL,  0xD5A79147C6E00BF3ULL);
+    MSG_X = _mm_add_epi32(TMSG3_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG3_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG3_X, TMSG2_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG3_Y, TMSG2_Y, 4);
+    TMSG0_X = _mm_add_epi32(TMSG0_X, TMP_X);
+    TMSG0_Y = _mm_add_epi32(TMSG0_Y, TMP_Y);
+    TMSG0_X = _mm_sha256msg2_epu32(TMSG0_X, TMSG3_X);
+    TMSG0_Y = _mm_sha256msg2_epu32(TMSG0_Y, TMSG3_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG2_X = _mm_sha256msg1_epu32(TMSG2_X, TMSG3_X);
+    TMSG2_Y = _mm_sha256msg1_epu32(TMSG2_Y, TMSG3_Y);
+
+    // Rounds 32-35
+    TMP_X = _mm_set_epi64x(0x53380D134D2C6DFCULL, 0x2E1B213827B70A85ULL);
+    MSG_X = _mm_add_epi32(TMSG0_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG0_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG0_X, TMSG3_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG0_Y, TMSG3_Y, 4);
+    TMSG1_X = _mm_add_epi32(TMSG1_X, TMP_X);
+    TMSG1_Y = _mm_add_epi32(TMSG1_Y, TMP_Y);
+    TMSG1_X = _mm_sha256msg2_epu32(TMSG1_X, TMSG0_X);
+    TMSG1_Y = _mm_sha256msg2_epu32(TMSG1_Y, TMSG0_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG3_X = _mm_sha256msg1_epu32(TMSG3_X, TMSG0_X);
+    TMSG3_Y = _mm_sha256msg1_epu32(TMSG3_Y, TMSG0_Y);
+
+    // Rounds 36-39
+    TMP_X = _mm_set_epi64x(0x92722C8581C2C92EULL, 0x766A0ABB650A7354ULL);
+    MSG_X = _mm_add_epi32(TMSG1_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG1_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG1_X, TMSG0_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG1_Y, TMSG0_Y, 4);
+    TMSG2_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    TMSG2_Y = _mm_add_epi32(TMSG2_Y, TMP_Y);
+    TMSG2_X = _mm_sha256msg2_epu32(TMSG2_X, TMSG1_X);
+    TMSG2_Y = _mm_sha256msg2_epu32(TMSG2_Y, TMSG1_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG0_X = _mm_sha256msg1_epu32(TMSG0_X, TMSG1_X);
+    TMSG0_Y = _mm_sha256msg1_epu32(TMSG0_Y, TMSG1_Y);
+
+    // Rounds 40-43
+    TMP_X = _mm_set_epi64x(0xC76C51A3C24B8B70ULL, 0xA81A664BA2BFE8A1ULL);
+    MSG_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG2_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG2_X, TMSG1_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG2_Y, TMSG1_Y, 4);
+    TMSG3_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    TMSG3_Y = _mm_add_epi32(TMSG3_Y, TMP_Y);
+    TMSG3_X = _mm_sha256msg2_epu32(TMSG3_X, TMSG2_X);
+    TMSG3_Y = _mm_sha256msg2_epu32(TMSG3_Y, TMSG2_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG1_X = _mm_sha256msg1_epu32(TMSG1_X, TMSG2_X);
+    TMSG1_Y = _mm_sha256msg1_epu32(TMSG1_Y, TMSG2_Y);
+
+    // Rounds 44-47
+    TMP_X = _mm_set_epi64x(0x106AA070F40E3585ULL, 0xD6990624D192E819ULL);
+    MSG_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG3_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG3_X, TMSG2_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG3_Y, TMSG2_Y, 4);
+    TMSG0_X = _mm_add_epi32(TMSG0_X, TMP_X);
+    TMSG0_Y = _mm_add_epi32(TMSG0_Y, TMP_Y);
+    TMSG0_X = _mm_sha256msg2_epu32(TMSG0_X, TMSG3_X);
+    TMSG0_Y = _mm_sha256msg2_epu32(TMSG0_Y, TMSG3_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG2_X = _mm_sha256msg1_epu32(TMSG2_X, TMSG3_X);
+    TMSG2_Y = _mm_sha256msg1_epu32(TMSG2_Y, TMSG3_Y);
+
+    // Rounds 48-51
+    TMP_X = _mm_set_epi64x(0x34B0BCB52748774CULL, 0x1E376C0819A4C116ULL);
+    MSG_X = _mm_add_epi32(TMSG0_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG0_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG0_X, TMSG3_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG0_Y, TMSG3_Y, 4);
+    TMSG1_X = _mm_add_epi32(TMSG1_X, TMP_X);
+    TMSG1_Y = _mm_add_epi32(TMSG1_Y, TMP_Y);
+    TMSG1_X = _mm_sha256msg2_epu32(TMSG1_X, TMSG0_X);
+    TMSG1_Y = _mm_sha256msg2_epu32(TMSG1_Y, TMSG0_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+    TMSG3_X = _mm_sha256msg1_epu32(TMSG3_X, TMSG0_X);
+    TMSG3_Y = _mm_sha256msg1_epu32(TMSG3_Y, TMSG0_Y);
+
+    // Rounds 52-55
+    TMP_X = _mm_set_epi64x(0x682E6FF35B9CCA4FULL, 0x4ED8AA4A391C0CB3ULL);
+    MSG_X = _mm_add_epi32(TMSG1_X, TMP_X );
+    MSG_Y = _mm_add_epi32(TMSG1_Y, TMP_X );
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG1_X, TMSG0_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG1_Y, TMSG0_Y, 4);
+    TMSG2_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    TMSG2_Y = _mm_add_epi32(TMSG2_Y, TMP_Y);
+    TMSG2_X = _mm_sha256msg2_epu32(TMSG2_X, TMSG1_X);
+    TMSG2_Y = _mm_sha256msg2_epu32(TMSG2_Y, TMSG1_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+
+    // Rounds 56-59
+    TMP_X = _mm_set_epi64x(0x8CC7020884C87814ULL, 0x78A5636F748F82EEULL);
+    MSG_X = _mm_add_epi32(TMSG2_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG2_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    TMP_X = _mm_alignr_epi8(TMSG2_X, TMSG1_X, 4);
+    TMP_Y = _mm_alignr_epi8(TMSG2_Y, TMSG1_Y, 4);
+    TMSG3_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    TMSG3_Y = _mm_add_epi32(TMSG3_Y, TMP_Y);
+    TMSG3_X = _mm_sha256msg2_epu32(TMSG3_X, TMSG2_X);
+    TMSG3_Y = _mm_sha256msg2_epu32(TMSG3_Y, TMSG2_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+
+    // Rounds 60-63
+    TMP_X = _mm_set_epi64x(0xC67178F2BEF9A3F7ULL, 0xA4506CEB90BEFFFAULL);
+    MSG_X = _mm_add_epi32(TMSG3_X, TMP_X);
+    MSG_Y = _mm_add_epi32(TMSG3_Y, TMP_X);
+    STATE1_X = _mm_sha256rnds2_epu32(STATE1_X, STATE0_X, MSG_X);
+    STATE1_Y = _mm_sha256rnds2_epu32(STATE1_Y, STATE0_Y, MSG_Y);
+    MSG_X = _mm_shuffle_epi32(MSG_X, 0x0E);
+    MSG_Y = _mm_shuffle_epi32(MSG_Y, 0x0E);
+    STATE0_X = _mm_sha256rnds2_epu32(STATE0_X, STATE1_X, MSG_X);
+    STATE0_Y = _mm_sha256rnds2_epu32(STATE0_Y, STATE1_Y, MSG_Y);
+
+    // Add values back to state
+    STATE0_X = _mm_add_epi32(STATE0_X, ABEF_SAVE_X);
+    STATE1_X = _mm_add_epi32(STATE1_X, CDGH_SAVE_X);
+    STATE0_Y = _mm_add_epi32(STATE0_Y, ABEF_SAVE_Y);
+    STATE1_Y = _mm_add_epi32(STATE1_Y, CDGH_SAVE_Y);
+
+    TMP_X = _mm_shuffle_epi32(STATE0_X, 0x1B); // FEBA
+    TMP_Y = _mm_shuffle_epi32(STATE0_Y, 0x1B); // FEBA
+    STATE1_X = _mm_shuffle_epi32(STATE1_X, 0xB1); // DCHG
+    STATE1_Y = _mm_shuffle_epi32(STATE1_Y, 0xB1); // DCHG
+    STATE0_X = _mm_blend_epi16(TMP_X, STATE1_X, 0xF0); // DCBA
+    STATE0_Y = _mm_blend_epi16(TMP_Y, STATE1_Y, 0xF0); // DCBA
+    STATE1_X = _mm_alignr_epi8(STATE1_X, TMP_X, 8); // ABEF
+    STATE1_Y = _mm_alignr_epi8(STATE1_Y, TMP_Y, 8); // ABEF
+
+    // Save state
+    _mm_store_si128((__m128i*) &out_X[0], STATE0_X);
+    _mm_store_si128((__m128i*) &out_X[4], STATE1_X);
+    _mm_store_si128((__m128i*) &out_Y[0], STATE0_Y);
+    _mm_store_si128((__m128i*) &out_Y[4], STATE1_Y);
+}
+
+
+#endif
+
--- a/algo/sha/sha256-hash-4way.c
+++ b/algo/sha/sha256-hash-4way.c
@@ -75,8 +75,8 @@ static const uint32_t K256[64] =
   _mm_xor_si128( _mm_and_si128( _mm_xor_si128( Y, Z ), X ), Z ) 

 #define MAJs(X, Y, Z) \
-   _mm_or_si128( _mm_and_si128( X, Y ), \
-                    _mm_and_si128( _mm_or_si128( X, Y ), Z ) )
+  _mm_xor_si128( Y, _mm_and_si128( X_xor_Y = _mm_xor_si128( X, Y ), \
+                                   Y_xor_Z ) )

 #define BSG2_0(x) \
   _mm_xor_si128( _mm_xor_si128( \
@@ -94,37 +94,6 @@ static const uint32_t K256[64] =
   _mm_xor_si128( _mm_xor_si128( \
        mm128_ror_32(x, 17), mm128_ror_32(x, 19) ), _mm_srli_epi32(x, 10) )

-#define SHA2s_4WAY_STEP(A, B, C, D, E, F, G, H, i, j) \
-do { \
-  __m128i K = _mm_set1_epi32( K256[( (j)+(i) )] ); \
-  __m128i T1 = mm128_ror_32( E, 14 ); \
-  __m128i T2 = mm128_ror_32( A,  9 ); \
-  __m128i T3 = _mm_xor_si128( F, G ); \
-  __m128i T4 = _mm_or_si128( A, B ); \
-  __m128i T5 = _mm_and_si128( A, B ); \
-  K  = _mm_add_epi32( K, W[i] ); \
-  T1 = _mm_xor_si128( T1, E ); \
-  T2 = _mm_xor_si128( T2, A ); \
-  T3 = _mm_and_si128( T3, E ); \
-  T4 = _mm_and_si128( T4, C ); \
-  K  = _mm_add_epi32( H, K ); \
-  T1 = mm128_ror_32( T1,  5 ); \
-  T2 = mm128_ror_32( T2, 11 ); \
-  T3 = _mm_xor_si128( T3, G ); \
-  T4 = _mm_or_si128( T4, T5 ); \
-  T1 = _mm_xor_si128( T1, E ); \
-  T2 = _mm_xor_si128( T2, A ); \
-  T1 = mm128_ror_32( T1,  6 ); \
-  T2 = mm128_ror_32( T2,  2 ); \
-  T1 = _mm_add_epi32( T1, T3 ); \
-  T2 = _mm_add_epi32( T2, T4 ); \
-  T1 = _mm_add_epi32( T1, K ); \
-  H  = _mm_add_epi32( T1, T2 ); \
-  D  = _mm_add_epi32( D, T1 ); \
-} while (0)
-
-
-/*
 #define SHA2s_4WAY_STEP(A, B, C, D, E, F, G, H, i, j) \
 do { \
  __m128i T1, T2; \
@@ -132,43 +101,26 @@ do { \
  T1 = _mm_add_epi32( H, mm128_add4_32( BSG2_1(E), CHs(E, F, G), \
                                        K, W[i] ) ); \
  T2 = _mm_add_epi32( BSG2_0(A), MAJs(A, B, C) ); \
+  Y_xor_Z = X_xor_Y; \
  D  = _mm_add_epi32( D,  T1 ); \
  H  = _mm_add_epi32( T1, T2 ); \
 } while (0)
-*/

-
-static void
-sha256_4way_round( sha256_4way_context *ctx, __m128i *in, __m128i r[8] )
+// LE data, no need to byte swap
+static inline void SHA256_4WAY_TRANSFORM( __m128i *out, __m128i *W,
+                                          const __m128i *in )
 {
-   register  __m128i A, B, C, D, E, F, G, H;
-   __m128i W[16];
+   __m128i A, B, C, D, E, F, G, H, X_xor_Y, Y_xor_Z;

-   mm128_block_bswap_32( W, in );
-   mm128_block_bswap_32( W+8, in+8 );
-
-   if ( ctx->initialized )
-   {
-      A = r[0];
-      B = r[1];
-      C = r[2];
-      D = r[3];
-      E = r[4];
-      F = r[5];
-      G = r[6];
-      H = r[7];
-   }
-   else
-   {
-      A = m128_const1_64( 0x6A09E6676A09E667 );
-      B = m128_const1_64( 0xBB67AE85BB67AE85 );
-      C = m128_const1_64( 0x3C6EF3723C6EF372 );
-      D = m128_const1_64( 0xA54FF53AA54FF53A );
-      E = m128_const1_64( 0x510E527F510E527F );
-      F = m128_const1_64( 0x9B05688C9B05688C );
-      G = m128_const1_64( 0x1F83D9AB1F83D9AB );
-      H = m128_const1_64( 0x5BE0CD195BE0CD19 );
-   }
+   A = in[0];
+   B = in[1];
+   C = in[2];
+   D = in[3];
+   E = in[4];
+   F = in[5];
+   G = in[6];
+   H = in[7];
+   Y_xor_Z = _mm_xor_si128( B, C );

   SHA2s_4WAY_STEP( A, B, C, D, E, F, G, H,  0, 0 );
   SHA2s_4WAY_STEP( H, A, B, C, D, E, F, G,  1, 0 );
@@ -223,46 +175,47 @@ sha256_4way_round( sha256_4way_context *ctx, __m128i *in, __m128i r[8] )
      SHA2s_4WAY_STEP( C, D, E, F, G, H, A, B, 14, j );
      SHA2s_4WAY_STEP( B, C, D, E, F, G, H, A, 15, j );
   }
+   
+   out[0] = _mm_add_epi32( in[0], A );
+   out[1] = _mm_add_epi32( in[1], B );
+   out[2] = _mm_add_epi32( in[2], C );
+   out[3] = _mm_add_epi32( in[3], D );
+   out[4] = _mm_add_epi32( in[4], E );
+   out[5] = _mm_add_epi32( in[5], F );
+   out[6] = _mm_add_epi32( in[6], G );
+   out[7] = _mm_add_epi32( in[7], H );
+}

-   if ( ctx->initialized )
-   {
-      r[0] = _mm_add_epi32( r[0], A );
-      r[1] = _mm_add_epi32( r[1], B );
-      r[2] = _mm_add_epi32( r[2], C );
-      r[3] = _mm_add_epi32( r[3], D );
-      r[4] = _mm_add_epi32( r[4], E );
-      r[5] = _mm_add_epi32( r[5], F );
-      r[6] = _mm_add_epi32( r[6], G );
-      r[7] = _mm_add_epi32( r[7], H );
-   }
-   else
-   {
-      ctx->initialized = true;
-      r[0] = _mm_add_epi32( A, m128_const1_64( 0x6A09E6676A09E667 ) );
-      r[1] = _mm_add_epi32( B, m128_const1_64( 0xBB67AE85BB67AE85 ) );
-      r[2] = _mm_add_epi32( C, m128_const1_64( 0x3C6EF3723C6EF372 ) );
-      r[3] = _mm_add_epi32( D, m128_const1_64( 0xA54FF53AA54FF53A ) );
-      r[4] = _mm_add_epi32( E, m128_const1_64( 0x510E527F510E527F ) );
-      r[5] = _mm_add_epi32( F, m128_const1_64( 0x9B05688C9B05688C ) );
-      r[6] = _mm_add_epi32( G, m128_const1_64( 0x1F83D9AB1F83D9AB ) );
-      r[7] = _mm_add_epi32( H, m128_const1_64( 0x5BE0CD195BE0CD19 ) );
-   }
+// LE data, no need to byte swap
+void sha256_4way_transform_le( __m128i *state_out, const __m128i *data,
+                               const __m128i *state_in )
+{
+   __m128i W[16];
+   memcpy_128( W, data, 16 );
+   SHA256_4WAY_TRANSFORM( state_out, W, state_in );
+}
+
+// BE data, need to byte swap input data
+void sha256_4way_transform_be( __m128i *state_out, const __m128i *data,
+                               const __m128i *state_in )
+{
+   __m128i W[16];
+   mm128_block_bswap_32( W, data );
+   mm128_block_bswap_32( W+8, data+8 );
+   SHA256_4WAY_TRANSFORM( state_out, W, state_in );
 }

 void sha256_4way_init( sha256_4way_context *sc )
 {
-   sc->initialized = false;
   sc->count_high = sc->count_low = 0;
-/*
-   sc->val[0] = _mm_set1_epi32( H256[0] );
-   sc->val[1] = _mm_set1_epi32( H256[1] );
-   sc->val[2] = _mm_set1_epi32( H256[2] );
-   sc->val[3] = _mm_set1_epi32( H256[3] );
-   sc->val[4] = _mm_set1_epi32( H256[4] );
-   sc->val[5] = _mm_set1_epi32( H256[5] );
-   sc->val[6] = _mm_set1_epi32( H256[6] );
-   sc->val[7] = _mm_set1_epi32( H256[7] );
-*/
+   sc->val[0] = m128_const1_64( 0x6A09E6676A09E667 );
+   sc->val[1] = m128_const1_64( 0xBB67AE85BB67AE85 );
+   sc->val[2] = m128_const1_64( 0x3C6EF3723C6EF372 );
+   sc->val[3] = m128_const1_64( 0xA54FF53AA54FF53A );
+   sc->val[4] = m128_const1_64( 0x510E527F510E527F );
+   sc->val[5] = m128_const1_64( 0x9B05688C9B05688C );
+   sc->val[6] = m128_const1_64( 0x1F83D9AB1F83D9AB );
+   sc->val[7] = m128_const1_64( 0x5BE0CD195BE0CD19 );
 }

 void sha256_4way_update( sha256_4way_context *sc, const void *data, size_t len )
@@ -286,7 +239,7 @@ void sha256_4way_update( sha256_4way_context *sc, const void *data, size_t len )
      len -= clen;
      if ( ptr == buf_size )
      {
-         sha256_4way_round( sc, sc->buf, sc->val );
+         sha256_4way_transform_be( sc->val, sc->buf, sc->val );
         ptr = 0;
      }
      clow = sc->count_low;
@@ -311,7 +264,7 @@ void sha256_4way_close( sha256_4way_context *sc, void *dst )
    if ( ptr > pad )
    {
         memset_zero_128( sc->buf + (ptr>>2), (buf_size - ptr) >> 2 );
-         sha256_4way_round( sc, sc->buf, sc->val );
+         sha256_4way_transform_be( sc->val, sc->buf, sc->val );
         memset_zero_128( sc->buf, pad >> 2 );
    }
    else
@@ -321,11 +274,9 @@ void sha256_4way_close( sha256_4way_context *sc, void *dst )
    high = (sc->count_high << 3) | (low >> 29);
    low = low << 3;

-    sc->buf[ pad >> 2 ] =
-                 mm128_bswap_32( m128_const1_32( high ) );
-    sc->buf[ ( pad+4 ) >> 2 ] =
-                 mm128_bswap_32( m128_const1_32( low ) );
-    sha256_4way_round( sc, sc->buf, sc->val );
+    sc->buf[  pad     >> 2 ] = m128_const1_32( bswap_32( high ) );
+    sc->buf[( pad+4 ) >> 2 ] = m128_const1_32( bswap_32( low ) );
+    sha256_4way_transform_be( sc->val, sc->buf, sc->val );

    mm128_block_bswap_32( dst, sc->val );
 }
@@ -342,74 +293,147 @@ void sha256_4way_full( void *dst, const void *data, size_t len )

 // SHA-256 8 way

-#define CHx(X, Y, Z) \
-   _mm256_xor_si256( _mm256_and_si256( _mm256_xor_si256( Y, Z ), X ), Z ) 
-
-#define MAJx(X, Y, Z) \
-   _mm256_or_si256( _mm256_and_si256( X, Y ), \
-                    _mm256_and_si256( _mm256_or_si256( X, Y ), Z ) )
-
 #define BSG2_0x(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,  2), mm256_ror_32(x, 13) ), mm256_ror_32( x, 22) )
+   _mm256_xor_si256( _mm256_xor_si256( mm256_ror_32( x,  2 ), \
+                                       mm256_ror_32( x, 13 ) ), \
+                                       mm256_ror_32( x, 22 ) )

 #define BSG2_1x(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,  6), mm256_ror_32(x, 11) ), mm256_ror_32( x, 25) )
+   _mm256_xor_si256( _mm256_xor_si256( mm256_ror_32( x,  6 ), \
+                                       mm256_ror_32( x, 11 ) ), \
+                                       mm256_ror_32( x, 25 ) )

 #define SSG2_0x(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x,  7), mm256_ror_32(x, 18) ), _mm256_srli_epi32(x, 3) ) 
+   _mm256_xor_si256( _mm256_xor_si256( mm256_ror_32( x,  7 ), \
+                                       mm256_ror_32( x, 18 ) ), \
+                                       _mm256_srli_epi32( x, 3 ) ) 

 #define SSG2_1x(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-       mm256_ror_32(x, 17), mm256_ror_32(x, 19) ), _mm256_srli_epi32(x, 10) )
+   _mm256_xor_si256( _mm256_xor_si256( mm256_ror_32( x, 17 ), \
+                                       mm256_ror_32( x, 19 ) ), \
+                                       _mm256_srli_epi32( x, 10 ) )

 #define SHA2x_MEXP( a, b, c, d ) \
     mm256_add4_32( SSG2_1x( W[a] ), W[b], SSG2_0x( W[c] ), W[d] );

-#define SHA2s_8WAY_STEP(A, B, C, D, E, F, G, H, i, j) \
+// With AVX512VL ternary logic optimizations are available.
+// If not optimize by forwarding the result of X^Y in MAJ to the next round
+// to avoid recalculating it as Y^Z. This optimization is not applicable
+// when MAJ is optimized with ternary logic.
+
+#if defined(__AVX512VL__)
+
+#define CHx(X, Y, Z)    _mm256_ternarylogic_epi32( X, Y, Z, 0xca )
+
+#define MAJx(X, Y, Z)   _mm256_ternarylogic_epi32( X, Y, Z, 0xe8 )
+
+#define SHA2s_8WAY_STEP( A, B, C, D, E, F, G, H, i, j ) \
 do { \
-  __m256i T1, T2; \
-  __m256i K = _mm256_set1_epi32( K256[( (j)+(i) )] ); \
-  T1 = _mm256_add_epi32( H, mm256_add4_32( BSG2_1x(E), CHx(E, F, G), \
-                                           K, W[i] ) ); \
-  T2 = _mm256_add_epi32( BSG2_0x(A), MAJx(A, B, C) ); \
+  __m256i T0 = _mm256_add_epi32( _mm256_set1_epi32( K256[ (j)+(i) ] ), \
+                                 W[ i ] ); \
+  __m256i T1 = BSG2_1x( E ); \
+  __m256i T2 = BSG2_0x( A ); \
+  T0 = _mm256_add_epi32( T0, CHx( E, F, G ) ); \
+  T1 = _mm256_add_epi32( T1, H ); \
+  T2 = _mm256_add_epi32( T2, MAJx( A, B, C ) ); \
+  T1 = _mm256_add_epi32( T1, T0 ); \
  D  = _mm256_add_epi32( D,  T1 ); \
  H  = _mm256_add_epi32( T1, T2 ); \
 } while (0)

-static void
-sha256_8way_round( sha256_8way_context *ctx,  __m256i *in, __m256i r[8] )
+#else  // AVX2
+
+#define CHx(X, Y, Z) \
+   _mm256_xor_si256( _mm256_and_si256( _mm256_xor_si256( Y, Z ), X ), Z ) 
+
+// Use saved X_xor_Y from previous round, now called Y_xor_Z,
+// and save new X_xor_Y, for next round.
+#define MAJx(X, Y, Z) \
+  _mm256_xor_si256( Y, _mm256_and_si256( X_xor_Y = _mm256_xor_si256( X, Y ), \
+                                         Y_xor_Z ) )
+
+#define SHA2s_8WAY_STEP( A, B, C, D, E, F, G, H, i, j ) \
+do { \
+  __m256i T0 = _mm256_add_epi32( _mm256_set1_epi32( K256[(j)+(i)] ), W[i] ); \
+  __m256i T1 = BSG2_1x( E ); \
+  __m256i T2 = BSG2_0x( A ); \
+  T0 = _mm256_add_epi32( T0, CHx( E, F, G ) ); \
+  T1 = _mm256_add_epi32( T1, H ); \
+  T2 = _mm256_add_epi32( T2, MAJx( A, B, C ) ); \
+  T1 = _mm256_add_epi32( T1, T0 ); \
+  Y_xor_Z = X_xor_Y; \
+  D  = _mm256_add_epi32( D,  T1 ); \
+  H  = _mm256_add_epi32( T1, T2 ); \
+} while (0)
+
+//  the X_xor_y technique can be extended to eliminate the mov instruction.
+//  Perform double rounds and alternate each round. Doesn't apply to AVX512
+//  and isn't suitable for running 3 round prehash.
+//
+// read Y_xor_Z, update X_xor_Y
+#define MAJ_2step(X, Y, Z, X_xor_Y, Y_xor_Z ) \
+  _mm256_xor_si256( Y, _mm256_and_si256( X_xor_Y = _mm256_xor_si256( X, Y ), \
+                                         Y_xor_Z ) )
+
+// start with toc initialized to y^z:   toc = B ^ C
+// First round reads toc as Y_xor_Z and saves X_xor_Y as tic.
+// Second round reads tic as Y_xor_Z and saves X_xor_Y as toc.
+
+#define SHA256_8WAY_2STEP( A, B, C, D, E, F, G, H, i0, i1, j ) \
+do { \
+  __m256i T0 = _mm256_add_epi32( _mm256_set1_epi32( K256[ (j)+(i0) ] ), \
+                                 W[ i0 ] ); \
+  __m256i T1 = BSG2_1x( E ); \
+  __m256i T2 = BSG2_0x( A ); \
+  T0 = _mm256_add_epi32( T0, CHx( E, F, G ) ); \
+  T1 = _mm256_add_epi32( T1, H ); \
+  T2 = _mm256_add_epi32( T2, MAJ_2step( A, B, C, tic, toc ) ); \
+  T1 = _mm256_add_epi32( T1, T0 ); \
+  D  = _mm256_add_epi32( D,  T1 ); \
+  H  = _mm256_add_epi32( T1, T2 ); \
+\
+  T0 = _mm256_add_epi32( _mm256_set1_epi32( K256[ (j)+(i1) ] ), \
+                                 W[ (i1) ] ); \
+  T1 = BSG2_1x( D ); \
+  T2 = BSG2_0x( H ); \
+  T0 = _mm256_add_epi32( T0, CHx( D, E, F ) ); \
+  T1 = _mm256_add_epi32( T1, G ); \
+  T2 = _mm256_add_epi32( T2, MAJ_2step( H, A, B, toc, tic ) ); \
+  T1 = _mm256_add_epi32( T1, T0 ); \
+  C  = _mm256_add_epi32( C,  T1 ); \
+  G  = _mm256_add_epi32( T1, T2 ); \
+} while (0)
+
+#endif   // AVX512VL else AVX2
+
+static inline void SHA256_8WAY_TRANSFORM( __m256i *out, __m256i *W,
+                                          const  __m256i *in ) \
 {
-   register  __m256i A, B, C, D, E, F, G, H;
-   __m256i W[16];
+   __m256i A, B, C, D, E, F, G, H;

-   mm256_block_bswap_32( W  , in   );
-   mm256_block_bswap_32( W+8, in+8 );
+   A = _mm256_load_si256( in   );
+   B = _mm256_load_si256( in+1 );
+   C = _mm256_load_si256( in+2 );
+   D = _mm256_load_si256( in+3 );
+   E = _mm256_load_si256( in+4 );
+   F = _mm256_load_si256( in+5 );
+   G = _mm256_load_si256( in+6 );
+   H = _mm256_load_si256( in+7 );

-   if ( ctx->initialized )
-   {
-      A = r[0];
-      B = r[1];
-      C = r[2];
-      D = r[3];
-      E = r[4];
-      F = r[5];
-      G = r[6];
-      H = r[7];
-   }
-   else
-   {
-      A = m256_const1_64( 0x6A09E6676A09E667 );
-      B = m256_const1_64( 0xBB67AE85BB67AE85 );
-      C = m256_const1_64( 0x3C6EF3723C6EF372 );
-      D = m256_const1_64( 0xA54FF53AA54FF53A );
-      E = m256_const1_64( 0x510E527F510E527F );
-      F = m256_const1_64( 0x9B05688C9B05688C );
-      G = m256_const1_64( 0x1F83D9AB1F83D9AB );
-      H = m256_const1_64( 0x5BE0CD195BE0CD19 );
-   }
+#if !defined(__AVX512VL__)
+
+   __m256i tic, toc = _mm256_xor_si256( B, C );
+
+   SHA256_8WAY_2STEP( A, B, C, D, E, F, G, H,  0,  1, 0 );
+   SHA256_8WAY_2STEP( G, H, A, B, C, D, E, F,  2,  3, 0 );
+   SHA256_8WAY_2STEP( E, F, G, H, A, B, C, D,  4,  5, 0 );
+   SHA256_8WAY_2STEP( C, D, E, F, G, H, A, B,  6,  7, 0 );
+   SHA256_8WAY_2STEP( A, B, C, D, E, F, G, H,  8,  9, 0 );
+   SHA256_8WAY_2STEP( G, H, A, B, C, D, E, F, 10, 11, 0 );
+   SHA256_8WAY_2STEP( E, F, G, H, A, B, C, D, 12, 13, 0 );
+   SHA256_8WAY_2STEP( C, D, E, F, G, H, A, B, 14, 15, 0 );
+
+#else

   SHA2s_8WAY_STEP( A, B, C, D, E, F, G, H,  0, 0 );
   SHA2s_8WAY_STEP( H, A, B, C, D, E, F, G,  1, 0 );
@@ -428,6 +452,8 @@ sha256_8way_round( sha256_8way_context *ctx,  __m256i *in, __m256i r[8] )
   SHA2s_8WAY_STEP( C, D, E, F, G, H, A, B, 14, 0 );
   SHA2s_8WAY_STEP( B, C, D, E, F, G, H, A, 15, 0 );

+#endif
+
   for ( int j = 16; j < 64; j += 16 )
   {
      W[ 0] = SHA2x_MEXP( 14,  9,  1,  0 );
@@ -447,6 +473,19 @@ sha256_8way_round( sha256_8way_context *ctx,  __m256i *in, __m256i r[8] )
      W[14] = SHA2x_MEXP( 12,  7, 15, 14 );
      W[15] = SHA2x_MEXP( 13,  8,  0, 15 );

+#if !defined(__AVX512VL__)
+
+      SHA256_8WAY_2STEP( A, B, C, D, E, F, G, H,  0,  1, j );
+      SHA256_8WAY_2STEP( G, H, A, B, C, D, E, F,  2,  3, j );
+      SHA256_8WAY_2STEP( E, F, G, H, A, B, C, D,  4,  5, j );
+      SHA256_8WAY_2STEP( C, D, E, F, G, H, A, B,  6,  7, j );
+      SHA256_8WAY_2STEP( A, B, C, D, E, F, G, H,  8,  9, j );
+      SHA256_8WAY_2STEP( G, H, A, B, C, D, E, F, 10, 11, j );
+      SHA256_8WAY_2STEP( E, F, G, H, A, B, C, D, 12, 13, j );
+      SHA256_8WAY_2STEP( C, D, E, F, G, H, A, B, 14, 15, j );
+
+#else
+      
      SHA2s_8WAY_STEP( A, B, C, D, E, F, G, H,  0, j );
      SHA2s_8WAY_STEP( H, A, B, C, D, E, F, G,  1, j );
      SHA2s_8WAY_STEP( G, H, A, B, C, D, E, F,  2, j );
@@ -463,50 +502,52 @@ sha256_8way_round( sha256_8way_context *ctx,  __m256i *in, __m256i r[8] )
      SHA2s_8WAY_STEP( D, E, F, G, H, A, B, C, 13, j );
      SHA2s_8WAY_STEP( C, D, E, F, G, H, A, B, 14, j );
      SHA2s_8WAY_STEP( B, C, D, E, F, G, H, A, 15, j );
+
+#endif      
   }

-   if ( ctx->initialized )
-   {
-      r[0] = _mm256_add_epi32( r[0], A );
-      r[1] = _mm256_add_epi32( r[1], B );
-      r[2] = _mm256_add_epi32( r[2], C );
-      r[3] = _mm256_add_epi32( r[3], D );
-      r[4] = _mm256_add_epi32( r[4], E );
-      r[5] = _mm256_add_epi32( r[5], F );
-      r[6] = _mm256_add_epi32( r[6], G );
-      r[7] = _mm256_add_epi32( r[7], H );
-   }
-   else
-   {
-      ctx->initialized = true;
-      r[0] = _mm256_add_epi32( A, m256_const1_64( 0x6A09E6676A09E667 ) );
-      r[1] = _mm256_add_epi32( B, m256_const1_64( 0xBB67AE85BB67AE85 ) );
-      r[2] = _mm256_add_epi32( C, m256_const1_64( 0x3C6EF3723C6EF372 ) );
-      r[3] = _mm256_add_epi32( D, m256_const1_64( 0xA54FF53AA54FF53A ) );
-      r[4] = _mm256_add_epi32( E, m256_const1_64( 0x510E527F510E527F ) );
-      r[5] = _mm256_add_epi32( F, m256_const1_64( 0x9B05688C9B05688C ) );
-      r[6] = _mm256_add_epi32( G, m256_const1_64( 0x1F83D9AB1F83D9AB ) );
-      r[7] = _mm256_add_epi32( H, m256_const1_64( 0x5BE0CD195BE0CD19 ) );
-   }
+   out[0] = _mm256_add_epi32( in[0], A );
+   out[1] = _mm256_add_epi32( in[1], B );
+   out[2] = _mm256_add_epi32( in[2], C );
+   out[3] = _mm256_add_epi32( in[3], D );
+   out[4] = _mm256_add_epi32( in[4], E );
+   out[5] = _mm256_add_epi32( in[5], F );
+   out[6] = _mm256_add_epi32( in[6], G );
+   out[7] = _mm256_add_epi32( in[7], H );
+}
+
+// accepts LE input data
+void sha256_8way_transform_le( __m256i *state_out, const __m256i *data,
+                               const __m256i *state_in )
+{
+   __m256i W[16];
+   memcpy_256( W, data, 16 );
+   SHA256_8WAY_TRANSFORM( state_out, W, state_in );
+}
+
+// Accepts BE input data, need to bswap
+void sha256_8way_transform_be( __m256i *state_out, const __m256i *data,
+                               const __m256i *state_in )
+{
+   __m256i W[16];
+   mm256_block_bswap_32( W  , data   );
+   mm256_block_bswap_32( W+8, data+8 );
+   SHA256_8WAY_TRANSFORM( state_out, W, state_in );
 }

 void sha256_8way_init( sha256_8way_context *sc )
 {
-   sc->initialized = false;
   sc->count_high = sc->count_low = 0;
-/*
-   sc->val[0] = _mm256_set1_epi32( H256[0] );
-   sc->val[1] = _mm256_set1_epi32( H256[1] );
-   sc->val[2] = _mm256_set1_epi32( H256[2] );
-   sc->val[3] = _mm256_set1_epi32( H256[3] );
-   sc->val[4] = _mm256_set1_epi32( H256[4] );
-   sc->val[5] = _mm256_set1_epi32( H256[5] );
-   sc->val[6] = _mm256_set1_epi32( H256[6] );
-   sc->val[7] = _mm256_set1_epi32( H256[7] );
-*/
+   sc->val[0] = m256_const1_64( 0x6A09E6676A09E667 );
+   sc->val[1] = m256_const1_64( 0xBB67AE85BB67AE85 );
+   sc->val[2] = m256_const1_64( 0x3C6EF3723C6EF372 );
+   sc->val[3] = m256_const1_64( 0xA54FF53AA54FF53A );
+   sc->val[4] = m256_const1_64( 0x510E527F510E527F );
+   sc->val[5] = m256_const1_64( 0x9B05688C9B05688C );
+   sc->val[6] = m256_const1_64( 0x1F83D9AB1F83D9AB );
+   sc->val[7] = m256_const1_64( 0x5BE0CD195BE0CD19 );
 }

-
 // need to handle odd byte length for yespower.
 // Assume only last update is odd.

@@ -531,7 +572,7 @@ void sha256_8way_update( sha256_8way_context *sc, const void *data, size_t len )
      len -= clen;
      if ( ptr == buf_size )
      {
-         sha256_8way_round( sc, sc->buf, sc->val );
+         sha256_8way_transform_be( sc->val, sc->buf, sc->val );
         ptr = 0;
      }
      clow = sc->count_low;
@@ -556,7 +597,7 @@ void sha256_8way_close( sha256_8way_context *sc, void *dst )
    if ( ptr > pad )
    {
         memset_zero_256( sc->buf + (ptr>>2), (buf_size - ptr) >> 2 );
-         sha256_8way_round( sc, sc->buf, sc->val );
+         sha256_8way_transform_be( sc->val, sc->buf, sc->val );
         memset_zero_256( sc->buf, pad >> 2 );
    }
    else
@@ -566,12 +607,10 @@ void sha256_8way_close( sha256_8way_context *sc, void *dst )
    high = (sc->count_high << 3) | (low >> 29);
    low = low << 3;

-    sc->buf[ pad >> 2 ] =
-                 mm256_bswap_32( m256_const1_32( high ) );
-    sc->buf[ ( pad+4 ) >> 2 ] =
-                 mm256_bswap_32( m256_const1_32( low ) );
+    sc->buf[   pad     >> 2 ] = m256_const1_32( bswap_32( high ) );
+    sc->buf[ ( pad+4 ) >> 2 ] = m256_const1_32( bswap_32( low ) );

-    sha256_8way_round( sc, sc->buf, sc->val );
+    sha256_8way_transform_be( sc->val, sc->buf, sc->val );

    mm256_block_bswap_32( dst, sc->val );
 }
@@ -588,32 +627,43 @@ void sha256_8way_full( void *dst, const void *data, size_t len )

 // SHA-256 16 way

-#define CHx16(X, Y, Z) \
-   _mm512_xor_si512( _mm512_and_si512( _mm512_xor_si512( Y, Z ), X ), Z ) 
+#define CHx16(X, Y, Z)    _mm512_ternarylogic_epi32( X, Y, Z, 0xca )

-#define MAJx16(X, Y, Z) \
-   _mm512_or_si512( _mm512_and_si512( X, Y ), \
-                    _mm512_and_si512( _mm512_or_si512( X, Y ), Z ) )
+#define MAJx16(X, Y, Z)   _mm512_ternarylogic_epi32( X, Y, Z, 0xe8 )

-#define BSG2_0x16(x) \
-   _mm512_xor_si512( _mm512_xor_si512( \
-       mm512_ror_32(x,  2), mm512_ror_32(x, 13) ), mm512_ror_32( x, 22) )
+#define BSG2_0x16(x)      mm512_xor3( _mm512_ror_epi32( x,  2 ), \
+                                      _mm512_ror_epi32( x, 13 ), \
+                                      _mm512_ror_epi32( x, 22 ) )

-#define BSG2_1x16(x) \
-   _mm512_xor_si512( _mm512_xor_si512( \
-       mm512_ror_32(x,  6), mm512_ror_32(x, 11) ), mm512_ror_32( x, 25) )
+#define BSG2_1x16(x)      mm512_xor3( _mm512_ror_epi32( x,  6 ), \
+                                      _mm512_ror_epi32( x, 11 ), \
+                                      _mm512_ror_epi32( x, 25 ) )

-#define SSG2_0x16(x) \
-   _mm512_xor_si512( _mm512_xor_si512( \
-       mm512_ror_32(x,  7), mm512_ror_32(x, 18) ), _mm512_srli_epi32(x, 3) ) 
+#define SSG2_0x16(x)      mm512_xor3( _mm512_ror_epi32(  x,  7 ), \
+                                      _mm512_ror_epi32(  x, 18 ), \
+                                      _mm512_srli_epi32( x,  3 ) )

-#define SSG2_1x16(x) \
-   _mm512_xor_si512( _mm512_xor_si512( \
-       mm512_ror_32(x, 17), mm512_ror_32(x, 19) ), _mm512_srli_epi32(x, 10) )
+#define SSG2_1x16(x)      mm512_xor3( _mm512_ror_epi32(  x, 17 ), \
+                                      _mm512_ror_epi32(  x, 19 ), \
+                                      _mm512_srli_epi32( x, 10 ) )

 #define SHA2x16_MEXP( a, b, c, d ) \
     mm512_add4_32( SSG2_1x16( W[a] ), W[b], SSG2_0x16( W[c] ), W[d] );

+#define SHA2s_16WAY_STEP( A, B, C, D, E, F, G, H, i, j ) \
+do { \
+  __m512i T0 = _mm512_add_epi32( _mm512_set1_epi32( K256[(j)+(i)] ), W[i] ); \
+  __m512i T1 = BSG2_1x16( E ); \
+  __m512i T2 = BSG2_0x16( A ); \
+  T0 = _mm512_add_epi32( T0, CHx16( E, F, G ) ); \
+  T1 = _mm512_add_epi32( T1, H ); \
+  T2 = _mm512_add_epi32( T2, MAJx16( A, B, C ) ); \
+  T1 = _mm512_add_epi32( T1, T0 ); \
+  D  = _mm512_add_epi32( D,  T1 ); \
+  H  = _mm512_add_epi32( T1, T2 ); \
+} while (0)
+   
+/*
 #define SHA2s_16WAY_STEP(A, B, C, D, E, F, G, H, i, j) \
 do { \
  __m512i T1, T2; \
@@ -624,38 +674,21 @@ do { \
  D  = _mm512_add_epi32( D,  T1 ); \
  H  = _mm512_add_epi32( T1, T2 ); \
 } while (0)
+*/

-static void
-sha256_16way_round( sha256_16way_context *ctx,  __m512i *in, __m512i r[8] )
+
+static inline void SHA256_16WAY_TRANSFORM( __m512i *out, __m512i *W,
+                                           const  __m512i *in ) \
 {
-   register  __m512i A, B, C, D, E, F, G, H;
-   __m512i W[16];
-
-   mm512_block_bswap_32( W  , in   );
-   mm512_block_bswap_32( W+8, in+8 );
-
-   if ( ctx->initialized )
-   {
-      A = r[0];
-      B = r[1];
-      C = r[2];
-      D = r[3];
-      E = r[4];
-      F = r[5];
-      G = r[6];
-      H = r[7];
-   }
-   else
-   {
-      A = m512_const1_64( 0x6A09E6676A09E667 );
-      B = m512_const1_64( 0xBB67AE85BB67AE85 );
-      C = m512_const1_64( 0x3C6EF3723C6EF372 );
-      D = m512_const1_64( 0xA54FF53AA54FF53A );
-      E = m512_const1_64( 0x510E527F510E527F );
-      F = m512_const1_64( 0x9B05688C9B05688C );
-      G = m512_const1_64( 0x1F83D9AB1F83D9AB );
-      H = m512_const1_64( 0x5BE0CD195BE0CD19 );
-   }
+   __m512i A, B, C, D, E, F, G, H;
+   A = _mm512_load_si512( in   );
+   B = _mm512_load_si512( in+1 );
+   C = _mm512_load_si512( in+2 );
+   D = _mm512_load_si512( in+3 );
+   E = _mm512_load_si512( in+4 );
+   F = _mm512_load_si512( in+5 );
+   G = _mm512_load_si512( in+6 );
+   H = _mm512_load_si512( in+7 );

   SHA2s_16WAY_STEP( A, B, C, D, E, F, G, H,  0, 0 );
   SHA2s_16WAY_STEP( H, A, B, C, D, E, F, G,  1, 0 );
@@ -711,38 +744,167 @@ sha256_16way_round( sha256_16way_context *ctx,  __m512i *in, __m512i r[8] )
      SHA2s_16WAY_STEP( B, C, D, E, F, G, H, A, 15, j );
   }

-   if ( ctx->initialized )
+   out[0] = _mm512_add_epi32( in[0], A );
+   out[1] = _mm512_add_epi32( in[1], B );
+   out[2] = _mm512_add_epi32( in[2], C );
+   out[3] = _mm512_add_epi32( in[3], D );
+   out[4] = _mm512_add_epi32( in[4], E );
+   out[5] = _mm512_add_epi32( in[5], F );
+   out[6] = _mm512_add_epi32( in[6], G );
+   out[7] = _mm512_add_epi32( in[7], H );
+}
+
+// accepts LE input data
+void sha256_16way_transform_le( __m512i *state_out, const __m512i *data,
+                                const __m512i *state_in )
+{
+   __m512i W[16];
+   memcpy_512( W, data, 16 );
+   SHA256_16WAY_TRANSFORM( state_out, W, state_in );
+}
+
+// Accepts BE input data, need to bswap
+void sha256_16way_transform_be( __m512i *state_out, const __m512i *data,
+                                const __m512i *state_in )
+{
+   __m512i W[16];
+   mm512_block_bswap_32( W  , data   );
+   mm512_block_bswap_32( W+8, data+8 );
+   SHA256_16WAY_TRANSFORM( state_out, W, state_in );
+}
+ 
+// Aggresive prehashing, LE byte order
+void sha256_16way_prehash_3rounds( __m512i *state_mid, const __m512i *W,
+                             const __m512i *state_in )
+{
+   __m512i A, B, C, D, E, F, G, H;
+
+   A = _mm512_load_si512( state_in     );
+   B = _mm512_load_si512( state_in + 1 );
+   C = _mm512_load_si512( state_in + 2 );
+   D = _mm512_load_si512( state_in + 3 );
+   E = _mm512_load_si512( state_in + 4 );
+   F = _mm512_load_si512( state_in + 5 );
+   G = _mm512_load_si512( state_in + 6 );
+   H = _mm512_load_si512( state_in + 7 );
+
+   SHA2s_16WAY_STEP( A, B, C, D, E, F, G, H,  0, 0 );
+   SHA2s_16WAY_STEP( H, A, B, C, D, E, F, G,  1, 0 );
+   SHA2s_16WAY_STEP( G, H, A, B, C, D, E, F,  2, 0 );
+
+   _mm512_store_si512( state_mid    , A );
+   _mm512_store_si512( state_mid + 1, B );
+   _mm512_store_si512( state_mid + 2, C );
+   _mm512_store_si512( state_mid + 3, D );
+   _mm512_store_si512( state_mid + 4, E );
+   _mm512_store_si512( state_mid + 5, F );
+   _mm512_store_si512( state_mid + 6, G );
+   _mm512_store_si512( state_mid + 7, H );
+}   
+
+void sha256_16way_final_rounds( __m512i *state_out, const __m512i *data,
+                          const __m512i *state_in, const __m512i *state_mid )
+{
+   __m512i A, B, C, D, E, F, G, H;
+   __m512i W[16];
+
+   memcpy_512( W, data, 16 );
+
+   A = _mm512_load_si512( state_mid     );
+   B = _mm512_load_si512( state_mid + 1 );
+   C = _mm512_load_si512( state_mid + 2 );
+   D = _mm512_load_si512( state_mid + 3 );
+   E = _mm512_load_si512( state_mid + 4 );
+   F = _mm512_load_si512( state_mid + 5 );
+   G = _mm512_load_si512( state_mid + 6 );
+   H = _mm512_load_si512( state_mid + 7 );
+
+//   SHA2s_16WAY_STEP( A, B, C, D, E, F, G, H,  0, 0 );
+//   SHA2s_16WAY_STEP( H, A, B, C, D, E, F, G,  1, 0 );
+//   SHA2s_16WAY_STEP( G, H, A, B, C, D, E, F,  2, 0 );
+   SHA2s_16WAY_STEP( F, G, H, A, B, C, D, E,  3, 0 );
+   SHA2s_16WAY_STEP( E, F, G, H, A, B, C, D,  4, 0 );
+   SHA2s_16WAY_STEP( D, E, F, G, H, A, B, C,  5, 0 );
+   SHA2s_16WAY_STEP( C, D, E, F, G, H, A, B,  6, 0 );
+   SHA2s_16WAY_STEP( B, C, D, E, F, G, H, A,  7, 0 );
+   SHA2s_16WAY_STEP( A, B, C, D, E, F, G, H,  8, 0 );
+   SHA2s_16WAY_STEP( H, A, B, C, D, E, F, G,  9, 0 );
+   SHA2s_16WAY_STEP( G, H, A, B, C, D, E, F, 10, 0 );
+   SHA2s_16WAY_STEP( F, G, H, A, B, C, D, E, 11, 0 );
+   SHA2s_16WAY_STEP( E, F, G, H, A, B, C, D, 12, 0 );
+   SHA2s_16WAY_STEP( D, E, F, G, H, A, B, C, 13, 0 );
+   SHA2s_16WAY_STEP( C, D, E, F, G, H, A, B, 14, 0 );
+   SHA2s_16WAY_STEP( B, C, D, E, F, G, H, A, 15, 0 );
+
+   for ( int j = 16; j < 64; j += 16 )
   {
-      r[0] = _mm512_add_epi32( r[0], A );
-      r[1] = _mm512_add_epi32( r[1], B );
-      r[2] = _mm512_add_epi32( r[2], C );
-      r[3] = _mm512_add_epi32( r[3], D );
-      r[4] = _mm512_add_epi32( r[4], E );
-      r[5] = _mm512_add_epi32( r[5], F );
-      r[6] = _mm512_add_epi32( r[6], G );
-      r[7] = _mm512_add_epi32( r[7], H );
-   }
-   else
-   {
-      ctx->initialized = true;
-      r[0] = _mm512_add_epi32( A, m512_const1_64( 0x6A09E6676A09E667 ) );
-      r[1] = _mm512_add_epi32( B, m512_const1_64( 0xBB67AE85BB67AE85 ) );
-      r[2] = _mm512_add_epi32( C, m512_const1_64( 0x3C6EF3723C6EF372 ) );
-      r[3] = _mm512_add_epi32( D, m512_const1_64( 0xA54FF53AA54FF53A ) );
-      r[4] = _mm512_add_epi32( E, m512_const1_64( 0x510E527F510E527F ) );
-      r[5] = _mm512_add_epi32( F, m512_const1_64( 0x9B05688C9B05688C ) );
-      r[6] = _mm512_add_epi32( G, m512_const1_64( 0x1F83D9AB1F83D9AB ) );
-      r[7] = _mm512_add_epi32( H, m512_const1_64( 0x5BE0CD195BE0CD19 ) );
+      W[ 0] = SHA2x16_MEXP( 14,  9,  1,  0 );
+      W[ 1] = SHA2x16_MEXP( 15, 10,  2,  1 );
+      W[ 2] = SHA2x16_MEXP(  0, 11,  3,  2 );
+      W[ 3] = SHA2x16_MEXP(  1, 12,  4,  3 );
+      W[ 4] = SHA2x16_MEXP(  2, 13,  5,  4 );
+      W[ 5] = SHA2x16_MEXP(  3, 14,  6,  5 );
+      W[ 6] = SHA2x16_MEXP(  4, 15,  7,  6 );
+      W[ 7] = SHA2x16_MEXP(  5,  0,  8,  7 );
+      W[ 8] = SHA2x16_MEXP(  6,  1,  9,  8 );
+      W[ 9] = SHA2x16_MEXP(  7,  2, 10,  9 );
+      W[10] = SHA2x16_MEXP(  8,  3, 11, 10 );
+      W[11] = SHA2x16_MEXP(  9,  4, 12, 11 );
+      W[12] = SHA2x16_MEXP( 10,  5, 13, 12 );
+      W[13] = SHA2x16_MEXP( 11,  6, 14, 13 );
+      W[14] = SHA2x16_MEXP( 12,  7, 15, 14 );
+      W[15] = SHA2x16_MEXP( 13,  8,  0, 15 );
+
+      SHA2s_16WAY_STEP( A, B, C, D, E, F, G, H,  0, j );
+      SHA2s_16WAY_STEP( H, A, B, C, D, E, F, G,  1, j );
+      SHA2s_16WAY_STEP( G, H, A, B, C, D, E, F,  2, j );
+      SHA2s_16WAY_STEP( F, G, H, A, B, C, D, E,  3, j );
+      SHA2s_16WAY_STEP( E, F, G, H, A, B, C, D,  4, j );
+      SHA2s_16WAY_STEP( D, E, F, G, H, A, B, C,  5, j );
+      SHA2s_16WAY_STEP( C, D, E, F, G, H, A, B,  6, j );
+      SHA2s_16WAY_STEP( B, C, D, E, F, G, H, A,  7, j );
+      SHA2s_16WAY_STEP( A, B, C, D, E, F, G, H,  8, j );
+      SHA2s_16WAY_STEP( H, A, B, C, D, E, F, G,  9, j );
+      SHA2s_16WAY_STEP( G, H, A, B, C, D, E, F, 10, j );
+      SHA2s_16WAY_STEP( F, G, H, A, B, C, D, E, 11, j );
+      SHA2s_16WAY_STEP( E, F, G, H, A, B, C, D, 12, j );
+      SHA2s_16WAY_STEP( D, E, F, G, H, A, B, C, 13, j );
+      SHA2s_16WAY_STEP( C, D, E, F, G, H, A, B, 14, j );
+      SHA2s_16WAY_STEP( B, C, D, E, F, G, H, A, 15, j );
   }
+
+   A = _mm512_add_epi32( A, _mm512_load_si512( state_in     ) );
+   B = _mm512_add_epi32( B, _mm512_load_si512( state_in + 1 ) );
+   C = _mm512_add_epi32( C, _mm512_load_si512( state_in + 2 ) );
+   D = _mm512_add_epi32( D, _mm512_load_si512( state_in + 3 ) );
+   E = _mm512_add_epi32( E, _mm512_load_si512( state_in + 4 ) );
+   F = _mm512_add_epi32( F, _mm512_load_si512( state_in + 5 ) );
+   G = _mm512_add_epi32( G, _mm512_load_si512( state_in + 6 ) );
+   H = _mm512_add_epi32( H, _mm512_load_si512( state_in + 7 ) );
+   
+   _mm512_store_si512( state_out    ,  A );
+   _mm512_store_si512( state_out + 1,  B );
+   _mm512_store_si512( state_out + 2,  C );
+   _mm512_store_si512( state_out + 3,  D );
+   _mm512_store_si512( state_out + 4,  E );
+   _mm512_store_si512( state_out + 5,  F );
+   _mm512_store_si512( state_out + 6,  G );
+   _mm512_store_si512( state_out + 7,  H );
 }

 void sha256_16way_init( sha256_16way_context *sc )
 {
-   sc->initialized = false;
   sc->count_high = sc->count_low = 0;
+   sc->val[0] = m512_const1_64( 0x6A09E6676A09E667 );
+   sc->val[1] = m512_const1_64( 0xBB67AE85BB67AE85 );
+   sc->val[2] = m512_const1_64( 0x3C6EF3723C6EF372 );
+   sc->val[3] = m512_const1_64( 0xA54FF53AA54FF53A );
+   sc->val[4] = m512_const1_64( 0x510E527F510E527F );
+   sc->val[5] = m512_const1_64( 0x9B05688C9B05688C );
+   sc->val[6] = m512_const1_64( 0x1F83D9AB1F83D9AB );
+   sc->val[7] = m512_const1_64( 0x5BE0CD195BE0CD19 );
 }

-
 void sha256_16way_update( sha256_16way_context *sc, const void *data,
                           size_t len )
 {
@@ -765,7 +927,7 @@ void sha256_16way_update( sha256_16way_context *sc, const void *data,
      len -= clen;
      if ( ptr == buf_size )
      {
-         sha256_16way_round( sc, sc->buf, sc->val );
+         sha256_16way_transform_be( sc->val, sc->buf, sc->val );
         ptr = 0;
      }
      clow = sc->count_low;
@@ -790,7 +952,7 @@ void sha256_16way_close( sha256_16way_context *sc, void *dst )
    if ( ptr > pad )
    {
         memset_zero_512( sc->buf + (ptr>>2), (buf_size - ptr) >> 2 );
-         sha256_16way_round( sc, sc->buf, sc->val );
+         sha256_16way_transform_be( sc->val, sc->buf, sc->val );
         memset_zero_512( sc->buf, pad >> 2 );
    }
    else
@@ -800,12 +962,10 @@ void sha256_16way_close( sha256_16way_context *sc, void *dst )
    high = (sc->count_high << 3) | (low >> 29);
    low = low << 3;

-    sc->buf[ pad >> 2 ] =
-                 mm512_bswap_32( m512_const1_32( high ) );
-    sc->buf[ ( pad+4 ) >> 2 ] =
-                 mm512_bswap_32( m512_const1_32( low ) );
+    sc->buf[   pad     >> 2 ] = m512_const1_32( bswap_32( high ) );
+    sc->buf[ ( pad+4 ) >> 2 ] = m512_const1_32( bswap_32( low ) );

-    sha256_16way_round( sc, sc->buf, sc->val );
+    sha256_16way_transform_be( sc->val, sc->buf, sc->val );

    mm512_block_bswap_32( dst, sc->val );
 }
--- a/algo/sha/sha256-hash-opt.c
+++ b/algo/sha/sha256-hash-opt.c
@@ -0,0 +1,388 @@
+/*   Intel SHA extensions using C intrinsics               */
+/*   Written and place in public domain by Jeffrey Walton  */
+/*   Based on code from Intel, and by Sean Gulley for      */
+/*   the miTLS project.                                    */
+
+// A stripped down version with byte swapping removed. 
+
+#if defined(__SHA__)
+
+#include "sha256-hash.h"
+
+void sha256_opt_transform_le( uint32_t *state_out, const void *input,
+                           const uint32_t *state_in )
+{
+    __m128i STATE0, STATE1;
+    __m128i MSG, TMP;
+    __m128i TMSG0, TMSG1, TMSG2, TMSG3;
+    __m128i ABEF_SAVE, CDGH_SAVE;
+
+    // Load initial values
+    TMP = _mm_load_si128((__m128i*) &state_in[0]);
+    STATE1 = _mm_load_si128((__m128i*) &state_in[4]);
+//    MASK = _mm_set_epi64x(0x0c0d0e0f08090a0bULL, 0x0405060700010203ULL);
+
+    TMP = _mm_shuffle_epi32(TMP, 0xB1); // CDAB
+    STATE1 = _mm_shuffle_epi32(STATE1, 0x1B); // EFGH
+    STATE0 = _mm_alignr_epi8(TMP, STATE1, 8); // ABEF
+    STATE1 = _mm_blend_epi16(STATE1, TMP, 0xF0); // CDGH
+
+    // Save current hash
+    ABEF_SAVE = STATE0;
+    CDGH_SAVE = STATE1;
+
+    // Rounds 0-3
+    TMSG0 = _mm_load_si128((const __m128i*) (input+0));
+//    TMSG0 = _mm_shuffle_epi8(MSG, MASK);
+    MSG = _mm_add_epi32(TMSG0, _mm_set_epi64x(0xE9B5DBA5B5C0FBCFULL, 0x71374491428A2F98ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+
+    // Rounds 4-7
+    TMSG1 = _mm_load_si128((const __m128i*) (input+16));
+//    TMSG1 = _mm_shuffle_epi8(TMSG1, MASK);
+    MSG = _mm_add_epi32(TMSG1, _mm_set_epi64x(0xAB1C5ED5923F82A4ULL, 0x59F111F13956C25BULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1);
+
+    // Rounds 8-11
+    TMSG2 = _mm_load_si128((const __m128i*) (input+32));
+//    TMSG2 = _mm_shuffle_epi8(TMSG2, MASK);
+    MSG = _mm_add_epi32(TMSG2, _mm_set_epi64x(0x550C7DC3243185BEULL, 0x12835B01D807AA98ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2);
+
+    // Rounds 12-15
+    TMSG3 = _mm_load_si128((const __m128i*) (input+48));
+//    TMSG3 = _mm_shuffle_epi8(TMSG3, MASK);
+    MSG = _mm_add_epi32(TMSG3, _mm_set_epi64x(0xC19BF1749BDC06A7ULL, 0x80DEB1FE72BE5D74ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG3, TMSG2, 4);
+    TMSG0 = _mm_add_epi32(TMSG0, TMP);
+    TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3);
+
+    // Rounds 16-19
+    MSG = _mm_add_epi32(TMSG0, _mm_set_epi64x(0x240CA1CC0FC19DC6ULL, 0xEFBE4786E49B69C1ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG0, TMSG3, 4);
+    TMSG1 = _mm_add_epi32(TMSG1, TMP);
+    TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0);
+
+    // Rounds 20-23
+    MSG = _mm_add_epi32(TMSG1, _mm_set_epi64x(0x76F988DA5CB0A9DCULL, 0x4A7484AA2DE92C6FULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG1, TMSG0, 4);
+    TMSG2 = _mm_add_epi32(TMSG2, TMP);
+    TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1);
+
+    // Rounds 24-27
+    MSG = _mm_add_epi32(TMSG2, _mm_set_epi64x(0xBF597FC7B00327C8ULL, 0xA831C66D983E5152ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG2, TMSG1, 4);
+    TMSG3 = _mm_add_epi32(TMSG3, TMP);
+    TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2);
+
+    // Rounds 28-31
+    MSG = _mm_add_epi32(TMSG3, _mm_set_epi64x(0x1429296706CA6351ULL,  0xD5A79147C6E00BF3ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG3, TMSG2, 4);
+    TMSG0 = _mm_add_epi32(TMSG0, TMP);
+    TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3);
+
+    // Rounds 32-35
+    MSG = _mm_add_epi32(TMSG0, _mm_set_epi64x(0x53380D134D2C6DFCULL, 0x2E1B213827B70A85ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG0, TMSG3, 4);
+    TMSG1 = _mm_add_epi32(TMSG1, TMP);
+    TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0);
+
+    // Rounds 36-39
+    MSG = _mm_add_epi32(TMSG1, _mm_set_epi64x(0x92722C8581C2C92EULL, 0x766A0ABB650A7354ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG1, TMSG0, 4);
+    TMSG2 = _mm_add_epi32(TMSG2, TMP);
+    TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1);
+
+    // Rounds 40-43
+    MSG = _mm_add_epi32(TMSG2, _mm_set_epi64x(0xC76C51A3C24B8B70ULL, 0xA81A664BA2BFE8A1ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG2, TMSG1, 4);
+    TMSG3 = _mm_add_epi32(TMSG3, TMP);
+    TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2);
+
+    // Rounds 44-47
+    MSG = _mm_add_epi32(TMSG3, _mm_set_epi64x(0x106AA070F40E3585ULL, 0xD6990624D192E819ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG3, TMSG2, 4);
+    TMSG0 = _mm_add_epi32(TMSG0, TMP);
+    TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3);
+
+    // Rounds 48-51
+    MSG = _mm_add_epi32(TMSG0, _mm_set_epi64x(0x34B0BCB52748774CULL, 0x1E376C0819A4C116ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG0, TMSG3, 4);
+    TMSG1 = _mm_add_epi32(TMSG1, TMP);
+    TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0);
+
+    // Rounds 52-55
+    MSG = _mm_add_epi32(TMSG1, _mm_set_epi64x(0x682E6FF35B9CCA4FULL, 0x4ED8AA4A391C0CB3ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG1, TMSG0, 4);
+    TMSG2 = _mm_add_epi32(TMSG2, TMP);
+    TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+
+    // Rounds 56-59
+    MSG = _mm_add_epi32(TMSG2, _mm_set_epi64x(0x8CC7020884C87814ULL, 0x78A5636F748F82EEULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG2, TMSG1, 4);
+    TMSG3 = _mm_add_epi32(TMSG3, TMP);
+    TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+
+    // Rounds 60-63
+    MSG = _mm_add_epi32(TMSG3, _mm_set_epi64x(0xC67178F2BEF9A3F7ULL, 0xA4506CEB90BEFFFAULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+
+    // Add values back to state
+    STATE0 = _mm_add_epi32(STATE0, ABEF_SAVE);
+    STATE1 = _mm_add_epi32(STATE1, CDGH_SAVE);
+
+    TMP = _mm_shuffle_epi32(STATE0, 0x1B); // FEBA
+    STATE1 = _mm_shuffle_epi32(STATE1, 0xB1); // DCHG
+    STATE0 = _mm_blend_epi16(TMP, STATE1, 0xF0); // DCBA
+    STATE1 = _mm_alignr_epi8(STATE1, TMP, 8); // ABEF
+
+    // Save state
+    _mm_store_si128((__m128i*) &state_out[0], STATE0);
+    _mm_store_si128((__m128i*) &state_out[4], STATE1);
+}
+
+
+void sha256_opt_transform_be( uint32_t *state_out, const void *input,
+                           const uint32_t *state_in )
+{
+    __m128i STATE0, STATE1;
+    __m128i MSG, TMP, MASK;
+    __m128i TMSG0, TMSG1, TMSG2, TMSG3;
+    __m128i ABEF_SAVE, CDGH_SAVE;
+
+    // Load initial values
+    TMP = _mm_load_si128((__m128i*) &state_in[0]);
+    STATE1 = _mm_load_si128((__m128i*) &state_in[4]);
+    MASK = _mm_set_epi64x(0x0c0d0e0f08090a0bULL, 0x0405060700010203ULL);
+
+    TMP = _mm_shuffle_epi32(TMP, 0xB1); // CDAB
+    STATE1 = _mm_shuffle_epi32(STATE1, 0x1B); // EFGH
+    STATE0 = _mm_alignr_epi8(TMP, STATE1, 8); // ABEF
+    STATE1 = _mm_blend_epi16(STATE1, TMP, 0xF0); // CDGH
+
+    // Save current hash
+    ABEF_SAVE = STATE0;
+    CDGH_SAVE = STATE1;
+
+    // Rounds 0-3
+    TMSG0 = _mm_load_si128((const __m128i*) (input+0));
+    TMSG0 = _mm_shuffle_epi8( TMSG0, MASK );
+    MSG = _mm_add_epi32(TMSG0, _mm_set_epi64x(0xE9B5DBA5B5C0FBCFULL, 0x71374491428A2F98ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+
+    // Rounds 4-7
+    TMSG1 = _mm_load_si128((const __m128i*) (input+16));
+    TMSG1 = _mm_shuffle_epi8(TMSG1, MASK);
+    MSG = _mm_add_epi32(TMSG1, _mm_set_epi64x(0xAB1C5ED5923F82A4ULL, 0x59F111F13956C25BULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1);
+    // Rounds 8-11
+    TMSG2 = _mm_load_si128((const __m128i*) (input+32));
+    TMSG2 = _mm_shuffle_epi8(TMSG2, MASK);
+    MSG = _mm_add_epi32(TMSG2, _mm_set_epi64x(0x550C7DC3243185BEULL, 0x12835B01D807AA98ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2);
+
+    // Rounds 12-15
+    TMSG3 = _mm_load_si128((const __m128i*) (input+48));
+    TMSG3 = _mm_shuffle_epi8(TMSG3, MASK);
+    MSG = _mm_add_epi32(TMSG3, _mm_set_epi64x(0xC19BF1749BDC06A7ULL, 0x80DEB1FE72BE5D74ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG3, TMSG2, 4);
+    TMSG0 = _mm_add_epi32(TMSG0, TMP);
+    TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3);
+
+    // Rounds 16-19
+    MSG = _mm_add_epi32(TMSG0, _mm_set_epi64x(0x240CA1CC0FC19DC6ULL, 0xEFBE4786E49B69C1ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG0, TMSG3, 4);
+    TMSG1 = _mm_add_epi32(TMSG1, TMP);
+    TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0);
+
+    // Rounds 20-23
+    MSG = _mm_add_epi32(TMSG1, _mm_set_epi64x(0x76F988DA5CB0A9DCULL, 0x4A7484AA2DE92C6FULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG1, TMSG0, 4);
+    TMSG2 = _mm_add_epi32(TMSG2, TMP);
+    TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1);
+
+    // Rounds 24-27
+    MSG = _mm_add_epi32(TMSG2, _mm_set_epi64x(0xBF597FC7B00327C8ULL, 0xA831C66D983E5152ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG2, TMSG1, 4);
+    TMSG3 = _mm_add_epi32(TMSG3, TMP);
+    TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2);
+
+    // Rounds 28-31
+    MSG = _mm_add_epi32(TMSG3, _mm_set_epi64x(0x1429296706CA6351ULL,  0xD5A79147C6E00BF3ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG3, TMSG2, 4);
+    TMSG0 = _mm_add_epi32(TMSG0, TMP);
+    TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3);
+
+    // Rounds 32-35
+    MSG = _mm_add_epi32(TMSG0, _mm_set_epi64x(0x53380D134D2C6DFCULL, 0x2E1B213827B70A85ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG0, TMSG3, 4);
+    TMSG1 = _mm_add_epi32(TMSG1, TMP);
+    TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0);
+
+    // Rounds 36-39
+    MSG = _mm_add_epi32(TMSG1, _mm_set_epi64x(0x92722C8581C2C92EULL, 0x766A0ABB650A7354ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG1, TMSG0, 4);
+    TMSG2 = _mm_add_epi32(TMSG2, TMP);
+    TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1);
+
+    // Rounds 40-43
+    MSG = _mm_add_epi32(TMSG2, _mm_set_epi64x(0xC76C51A3C24B8B70ULL, 0xA81A664BA2BFE8A1ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG2, TMSG1, 4);
+    TMSG3 = _mm_add_epi32(TMSG3, TMP);
+    TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2);
+
+    // Rounds 44-47
+    MSG = _mm_add_epi32(TMSG3, _mm_set_epi64x(0x106AA070F40E3585ULL, 0xD6990624D192E819ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG3, TMSG2, 4);
+    TMSG0 = _mm_add_epi32(TMSG0, TMP);
+    TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3);
+
+    // Rounds 48-51
+    MSG = _mm_add_epi32(TMSG0, _mm_set_epi64x(0x34B0BCB52748774CULL, 0x1E376C0819A4C116ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG0, TMSG3, 4);
+    TMSG1 = _mm_add_epi32(TMSG1, TMP);
+    TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+    TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0);
+
+    // Rounds 52-55
+    MSG = _mm_add_epi32(TMSG1, _mm_set_epi64x(0x682E6FF35B9CCA4FULL, 0x4ED8AA4A391C0CB3ULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG1, TMSG0, 4);
+    TMSG2 = _mm_add_epi32(TMSG2, TMP);
+    TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+
+    // Rounds 56-59
+    MSG = _mm_add_epi32(TMSG2, _mm_set_epi64x(0x8CC7020884C87814ULL, 0x78A5636F748F82EEULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    TMP = _mm_alignr_epi8(TMSG2, TMSG1, 4);
+    TMSG3 = _mm_add_epi32(TMSG3, TMP);
+    TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+
+    // Rounds 60-63
+    MSG = _mm_add_epi32(TMSG3, _mm_set_epi64x(0xC67178F2BEF9A3F7ULL, 0xA4506CEB90BEFFFAULL));
+    STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG);
+    MSG = _mm_shuffle_epi32(MSG, 0x0E);
+    STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, MSG);
+
+    // Add values back to state
+    STATE0 = _mm_add_epi32(STATE0, ABEF_SAVE);
+    STATE1 = _mm_add_epi32(STATE1, CDGH_SAVE);
+
+    TMP = _mm_shuffle_epi32(STATE0, 0x1B); // FEBA
+    STATE1 = _mm_shuffle_epi32(STATE1, 0xB1); // DCHG
+    STATE0 = _mm_blend_epi16(TMP, STATE1, 0xF0); // DCBA
+    STATE1 = _mm_alignr_epi8(STATE1, TMP, 8); // ABEF
+
+    // Save state
+    _mm_store_si128((__m128i*) &state_out[0], STATE0);
+    _mm_store_si128((__m128i*) &state_out[4], STATE1);
+}
+
+#endif
--- a/algo/sha/sha256-hash.c
+++ b/algo/sha/sha256-hash.c
@@ -0,0 +1,142 @@
+#include "sha256-hash.h"
+
+static const uint32_t SHA256_IV[8] =
+{
+   0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
+   0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
+};
+
+/*
+static const uint8_t SHA256_PAD[64] =
+{
+   0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+};
+*/
+
+void sha256_ctx_init( sha256_context *ctx )
+{
+   memcpy( ctx->state, SHA256_IV, sizeof SHA256_IV );
+   ctx->count = 0;
+}
+
+void sha256_update( sha256_context *ctx, const void *data, size_t len )
+{
+   int ptr = ctx->count & 0x3f;
+   const uint8_t *src = data;
+
+   ctx->count += (uint64_t)len;
+
+   if ( len < 64 - ptr )
+   {
+      memcpy( ctx->buf + ptr, src, len );
+      return;
+   }
+
+   memcpy( ctx->buf + ptr, src, 64 - ptr );
+   sha256_transform_be( ctx->state, (uint32_t*)ctx->buf, ctx->state );
+   src += 64 - ptr;
+   len -= 64 - ptr;
+
+   while ( len >= 64 )
+   {
+      sha256_transform_be( ctx->state, (uint32_t*)src, ctx->state );
+      src += 64;
+      len -= 64;
+   }
+
+   memcpy( ctx->buf, src, len );
+}
+
+#if 0
+void sha256_final( sha256_context *ctx, uint32_t *hash )
+{
+   size_t r;
+
+
+   /* Figure out how many bytes we have buffered. */
+   r = ctx->count & 0x3f;
+//   r = ( ctx->count >> 3 ) & 0x3f;
+
+//printf("final: count= %d, r= %d\n", ctx->count, r );
+   
+   /* Pad to 56 mod 64, transforming if we finish a block en route. */
+   if ( r < 56 )
+   {
+      /* Pad to 56 mod 64. */
+      memcpy( &ctx->buf[r], SHA256_PAD, 56 - r );
+   }
+   else
+   {
+      /* Finish the current block and mix. */
+      memcpy( &ctx->buf[r], SHA256_PAD, 64 - r );
+      sha256_transform_be( ctx->state, (uint32_t*)ctx->buf, ctx->state );
+
+//      SHA256_Transform(ctx->state, ctx->buf, &tmp32[0], &tmp32[64]);
+
+      /* The start of the final block is all zeroes. */
+      memset( &ctx->buf[0], 0, 56 );
+   }
+
+   /* Add the terminating bit-count. */
+   ctx->buf[56] = bswap_64( ctx->count << 3 );
+//   ctx->buf[56] = bswap_64( ctx->count );
+//   be64enc( &ctx->buf[56], ctx->count );
+
+   /* Mix in the final block. */
+   sha256_transform_be( ctx->state, (uint32_t*)ctx->buf, ctx->state );
+
+//   SHA256_Transform(ctx->state, ctx->buf, &tmp32[0], &tmp32[64]);
+
+   for ( int i = 0; i < 8; i++ )  hash[i] = bswap_32( ctx->state[i] );
+   
+//   for ( int i = 0; i < 8; i++ )  be32enc( hash + 4*i, ctx->state + i );
+
+/*
+//   be32enc_vect(digest, ctx->state, 4);
+//   be32enc_vect(uint8_t * dst, const uint32_t * src, size_t len)
+   // Encode vector, two words at a time. 
+   do {
+      be32enc(&dst[0], src[0]);
+      be32enc(&dst[4], src[1]);
+      src += 2;
+      dst += 8;
+   } while (--len);
+*/
+
+}
+#endif
+
+void sha256_final( sha256_context *ctx, void *hash )
+{
+   int ptr = ctx->count & 0x3f;
+
+   ctx->buf[ ptr++ ] = 0x80;
+   
+   if ( ptr > 56 )
+   {
+      memset( ctx->buf + ptr, 0, 64 - ptr );
+      sha256_transform_be( ctx->state, (uint32_t*)ctx->buf, ctx->state );
+      memset( ctx->buf, 0, 56 );
+   }
+   else
+      memset( ctx->buf + ptr, 0, 56 - ptr );
+
+   *(uint64_t*)(&ctx->buf[56]) = bswap_64( ctx->count << 3 );   
+
+   sha256_transform_be( ctx->state, (uint32_t*)ctx->buf, ctx->state );
+
+   for ( int i = 0; i < 8; i++ )
+      ( (uint32_t*)hash )[i] = bswap_32( ctx->state[i] );
+}
+
+void sha256_full( void *hash, const void *data, size_t len )
+{
+   sha256_context ctx;
+   sha256_ctx_init( &ctx );
+   sha256_update( &ctx, data, len );
+   sha256_final( &ctx, hash );
+}
+
--- a/algo/sha/sha256-hash.h
+++ b/algo/sha/sha256-hash.h
@@ -0,0 +1,56 @@
+#ifndef SHA256_HASH_H__
+#define SHA256_HASH_H__ 1
+
+#include <stddef.h>
+#include "simd-utils.h"
+#include "cpuminer-config.h"
+#include "sph_sha2.h"
+
+
+// generic interface 
+
+typedef struct {
+   unsigned char buf[64];    /* first field, for alignment */
+   uint32_t state[8];
+   uint64_t count;
+} sha256_context __attribute__((aligned(64)));
+
+void sha256_full( void *hash, const void *data, size_t len );
+void sha256_update( sha256_context *ctx, const void *data, size_t len );
+void sha256_final( sha256_context *ctx, void *hash );
+void sha256_ctx_init( sha256_context *ctx );
+void sha256_transform_le( uint32_t *state_out, const uint32_t *data,
+                          const uint32_t *state_in );
+void sha256_transform_be( uint32_t *state_out, const uint32_t *data,
+                          const uint32_t *state_in );
+
+#if defined(__SHA__)
+
+void sha256_opt_transform_le( uint32_t *state_out, const void *input,
+                           const uint32_t *state_in );
+
+void sha256_opt_transform_be( uint32_t *state_out, const void *input,
+                           const uint32_t *state_in );
+
+// 2 way with interleaved instructions
+void sha256_ni2way_transform_le( uint32_t *out_X, uint32_t*out_Y,
+                              const void *msg_X, const void *msg_Y,
+                              const uint32_t *in_X, const uint32_t *in_Y );
+
+void sha256_ni2way_transform_be( uint32_t *out_X, uint32_t*out_Y,
+                              const void *msg_X, const void *msg_Y,
+                              const uint32_t *in_X, const uint32_t *in_Y );
+
+// Select target
+// with SHA...
+#define sha256_transform_le sha256_opt_transform_le
+#define sha256_transform_be sha256_opt_transform_be
+
+#else
+
+// without SHA...
+#define sha256_transform_le sph_sha256_transform_le
+#define sha256_transform_be sph_sha256_transform_be
+
+#endif
+#endif
--- a/algo/sha/sha256d-4way.c
+++ b/algo/sha/sha256d-4way.c
@@ -0,0 +1,255 @@
+#include "sha256t-gate.h"
+#include <stdlib.h>
+#include <stdint.h>
+#include <string.h>
+#include <stdio.h>
+#include "sha-hash-4way.h"
+
+#if defined(SHA256D_16WAY)
+
+int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr )
+{
+   __m512i  block[16]    __attribute__ ((aligned (64)));
+   __m512i  hash32[8]    __attribute__ ((aligned (32)));
+   __m512i  initstate[8] __attribute__ ((aligned (32)));
+   __m512i  midstate[8]  __attribute__ ((aligned (32)));
+   __m512i  midstate2[8] __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   __m512i  vdata[20]    __attribute__ ((aligned (32)));
+   uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
+   uint32_t *pdata = work->data;
+   const uint32_t *ptarget = work->target;
+   const uint32_t targ32_d7 = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce - 16;
+   uint32_t n = first_nonce;
+   __m512i *noncev = vdata + 19;
+   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;
+   const __m512i last_byte = m512_const1_32( 0x80000000 );
+   const __m512i sixteen = m512_const1_32( 16 );
+
+   for ( int i = 0; i < 19; i++ )
+       vdata[i] = m512_const1_32( pdata[i] );
+
+   *noncev = _mm512_set_epi32( n+15, n+14, n+13, n+12, n+11, n+10, n+9, n+8,
+                               n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n+1, n );
+
+   // initialize state
+   initstate[0] = m512_const1_64( 0x6A09E6676A09E667 );
+   initstate[1] = m512_const1_64( 0xBB67AE85BB67AE85 );
+   initstate[2] = m512_const1_64( 0x3C6EF3723C6EF372 );
+   initstate[3] = m512_const1_64( 0xA54FF53AA54FF53A );
+   initstate[4] = m512_const1_64( 0x510E527F510E527F );
+   initstate[5] = m512_const1_64( 0x9B05688C9B05688C );
+   initstate[6] = m512_const1_64( 0x1F83D9AB1F83D9AB );
+   initstate[7] = m512_const1_64( 0x5BE0CD195BE0CD19 );
+
+   // hash first 64 byte block of data
+   sha256_16way_transform_le( midstate, vdata, initstate );
+
+   // Do 3 rounds on the first 12 bytes of the next block
+   sha256_16way_prehash_3rounds( midstate2, vdata + 16, midstate );
+
+   do
+   {
+      // 1. final 16 bytes of data, with padding
+      memcpy_512( block, vdata + 16, 4 );
+      block[ 4] = last_byte;
+      memset_zero_512( block + 5, 10 );
+      block[15] = m512_const1_32( 80*8 ); // bit count
+      sha256_16way_final_rounds( hash32, block, midstate, midstate2 );
+
+      // 2. 32 byte hash from 1.
+      memcpy_512( block, hash32, 8 );
+      block[ 8] = last_byte;
+      memset_zero_512( block + 9, 6 );
+      block[15] = m512_const1_32( 32*8 ); // bit count
+      sha256_16way_transform_le( hash32, block, initstate );
+
+      // byte swap final hash for testing
+      mm512_block_bswap_32( hash32, hash32 );
+
+      for ( int lane = 0; lane < 16; lane++ )
+      if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
+      {
+         extr_lane_16x32( lane_hash, hash32, lane, 256 );
+         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         {
+            pdata[19] = n + lane;
+            submit_solution( work, lane_hash, mythr );
+         }
+       }
+       *noncev = _mm512_add_epi32( *noncev, sixteen );
+       n += 16;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+   pdata[19] = n;
+   *hashes_done = n - first_nonce;
+   return 0;
+}
+
+#endif
+
+#if defined(SHA256D_8WAY)
+
+int scanhash_sha256d_8way( struct work *work, const uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr )
+{
+   __m256i  block[16]    __attribute__ ((aligned (64)));
+   __m256i  hash32[8]    __attribute__ ((aligned (32)));
+   __m256i  initstate[8] __attribute__ ((aligned (32)));
+   __m256i  midstate[8]  __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   __m256i  vdata[20]    __attribute__ ((aligned (32)));
+   uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
+   uint32_t *pdata = work->data;
+   const uint32_t *ptarget = work->target;
+   const uint32_t targ32_d7 = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce - 8;
+   uint32_t n = first_nonce;
+   __m256i *noncev = vdata + 19;
+   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;
+   const __m256i last_byte = m256_const1_32( 0x80000000 );
+   const __m256i eight = m256_const1_32( 8 );
+
+   for ( int i = 0; i < 19; i++ )
+       vdata[i] = m256_const1_32( pdata[i] );
+
+   *noncev = _mm256_set_epi32( n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n+1, n );
+
+   // initialize state
+   initstate[0] = m256_const1_64( 0x6A09E6676A09E667 );
+   initstate[1] = m256_const1_64( 0xBB67AE85BB67AE85 );
+   initstate[2] = m256_const1_64( 0x3C6EF3723C6EF372 );
+   initstate[3] = m256_const1_64( 0xA54FF53AA54FF53A );
+   initstate[4] = m256_const1_64( 0x510E527F510E527F );
+   initstate[5] = m256_const1_64( 0x9B05688C9B05688C );
+   initstate[6] = m256_const1_64( 0x1F83D9AB1F83D9AB );
+   initstate[7] = m256_const1_64( 0x5BE0CD195BE0CD19 );
+
+   // hash first 64 bytes of data
+   sha256_8way_transform_le( midstate, vdata, initstate );
+
+   do
+   {
+      // 1. final 16 bytes of data, with padding
+      memcpy_256( block, vdata + 16, 4 );
+      block[ 4] = last_byte;
+      memset_zero_256( block + 5, 10 );
+      block[15] = m256_const1_32( 80*8 ); // bit count
+      sha256_8way_transform_le( hash32, block, midstate );
+
+      // 2. 32 byte hash from 1.
+      memcpy_256( block, hash32, 8 );
+      block[ 8] = last_byte;
+      memset_zero_256( block + 9, 6 );
+      block[15] = m256_const1_32( 32*8 ); // bit count
+      sha256_8way_transform_le( hash32, block, initstate );
+
+      // byte swap final hash for testing
+      mm256_block_bswap_32( hash32, hash32 );
+
+      for ( int lane = 0; lane < 8; lane++ )
+      if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
+      {
+         extr_lane_8x32( lane_hash, hash32, lane, 256 );
+         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         {
+            pdata[19] = n + lane;
+            submit_solution( work, lane_hash, mythr );
+         }
+       }
+       *noncev = _mm256_add_epi32( *noncev, eight );
+       n += 8;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+   pdata[19] = n;
+   *hashes_done = n - first_nonce;
+   return 0;
+}
+
+#endif
+
+#if defined(SHA256D_4WAY)
+
+int scanhash_sha256d_4way( struct work *work, const uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr )
+{
+   __m128i  block[16]    __attribute__ ((aligned (64)));
+   __m128i  hash32[8]    __attribute__ ((aligned (32)));
+   __m128i  initstate[8] __attribute__ ((aligned (32)));
+   __m128i  midstate[8]  __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   __m128i  vdata[20]    __attribute__ ((aligned (32)));
+   uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
+   uint32_t *pdata = work->data;
+   const uint32_t *ptarget = work->target;
+   const uint32_t targ32_d7 = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce - 4;
+   uint32_t n = first_nonce;
+   __m128i *noncev = vdata + 19;
+   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;
+   const __m128i last_byte = m128_const1_32( 0x80000000 );
+   const __m128i four = m128_const1_32( 4 );
+
+   for ( int i = 0; i < 19; i++ )
+       vdata[i] = m128_const1_32( pdata[i] );
+
+   *noncev = _mm_set_epi32( n+ 3, n+ 2, n+1, n );
+
+   // initialize state
+   initstate[0] = m128_const1_64( 0x6A09E6676A09E667 );
+   initstate[1] = m128_const1_64( 0xBB67AE85BB67AE85 );
+   initstate[2] = m128_const1_64( 0x3C6EF3723C6EF372 );
+   initstate[3] = m128_const1_64( 0xA54FF53AA54FF53A );
+   initstate[4] = m128_const1_64( 0x510E527F510E527F );
+   initstate[5] = m128_const1_64( 0x9B05688C9B05688C );
+   initstate[6] = m128_const1_64( 0x1F83D9AB1F83D9AB );
+   initstate[7] = m128_const1_64( 0x5BE0CD195BE0CD19 );
+
+   // hash first 64 bytes of data
+   sha256_4way_transform_le( midstate, vdata, initstate );
+
+   do
+   {
+      // 1. final 16 bytes of data, with padding
+      memcpy_128( block, vdata + 16, 4 );
+      block[ 4] = last_byte;
+      memset_zero_128( block + 5, 10 );
+      block[15] = m128_const1_32( 80*8 ); // bit count
+      sha256_4way_transform_le( hash32, block, midstate );
+
+      // 2. 32 byte hash from 1.
+      memcpy_128( block, hash32, 8 );
+      block[ 8] = last_byte;
+      memset_zero_128( block + 9, 6 );
+      block[15] = m128_const1_32( 32*8 ); // bit count
+      sha256_4way_transform_le( hash32, block, initstate );
+
+      // byte swap final hash for testing
+      mm128_block_bswap_32( hash32, hash32 );
+
+      for ( int lane = 0; lane < 4; lane++ )
+      if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
+      {
+         extr_lane_4x32( lane_hash, hash32, lane, 256 );
+         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         {
+            pdata[19] = n + lane;
+            submit_solution( work, lane_hash, mythr );
+         }
+       }
+       *noncev = _mm_add_epi32( *noncev, four );
+       n += 4;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+   pdata[19] = n;
+   *hashes_done = n - first_nonce;
+   return 0;
+}
+
+#endif
+
--- a/algo/sha/sha256d.c
+++ b/algo/sha/sha256d.c
@@ -0,0 +1,8 @@
+#include "sha256d.h"
+
+void sha256d( void *hash, const void *data, int len )
+{
+   sha256_full( hash, data, len );
+   sha256_full( hash, hash,  32 );
+}
+
--- a/algo/sha/sha256d.h
+++ b/algo/sha/sha256d.h
@@ -0,0 +1,7 @@
+#include "algo-gate-api.h"
+#include <string.h>
+#include <inttypes.h>
+#include "sha256-hash.h"
+
+void sha256d( void *hash, const void *data, int len );
+
--- a/algo/sha/sha256q-4way.c
+++ b/algo/sha/sha256q-4way.c
@@ -5,6 +5,79 @@
 #include <stdio.h>
 #include "sha-hash-4way.h"

+#if defined(SHA256T_16WAY)
+
+static __thread sha256_16way_context sha256_ctx16 __attribute__ ((aligned (64)));
+
+void sha256q_16way_hash( void* output, const void* input )
+{
+   uint32_t vhash[8*16] __attribute__ ((aligned (64)));
+   sha256_16way_context ctx;
+   memcpy( &ctx, &sha256_ctx16, sizeof ctx );
+
+   sha256_16way_update( &ctx, input + (64<<4), 16 );
+   sha256_16way_close( &ctx, vhash );
+
+   sha256_16way_init( &ctx );
+   sha256_16way_update( &ctx, vhash, 32 );
+   sha256_16way_close( &ctx, vhash );
+
+   sha256_16way_init( &ctx );
+   sha256_16way_update( &ctx, vhash, 32 );
+   sha256_16way_close( &ctx, vhash );
+
+   sha256_16way_init( &ctx );
+   sha256_16way_update( &ctx, vhash, 32 );
+   sha256_16way_close( &ctx, output );
+}
+
+int scanhash_sha256q_16way( struct work *work, const uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t vdata[20*16] __attribute__ ((aligned (64)));
+   uint32_t hash32[8*16] __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash32_d7 = &(hash32[7<<4]);
+   uint32_t *pdata = work->data;
+   const uint32_t *ptarget = work->target;
+   const uint32_t targ32_d7 = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce - 16;
+   uint32_t n = first_nonce;
+   __m512i  *noncev = (__m512i*)vdata + 19;   // aligned
+   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;
+
+   mm512_bswap32_intrlv80_16x32( vdata, pdata );
+   *noncev = _mm512_set_epi32( n+15, n+14, n+13, n+12, n+11, n+10, n+9, n+8,
+                               n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n+1, n );
+   sha256_16way_init( &sha256_ctx16 );
+   sha256_16way_update( &sha256_ctx16, vdata, 64 );
+
+   do
+   {
+     pdata[19] = n;
+     sha256q_16way_hash( hash32, vdata );
+     for ( int lane = 0; lane < 16; lane++ )
+     if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
+     {
+        extr_lane_16x32( lane_hash, hash32, lane, 256 );
+        if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+        {
+           pdata[19] = bswap_32( n + lane );
+           submit_solution( work, lane_hash, mythr );
+        }
+      }
+      *noncev = _mm512_add_epi32( *noncev, m512_const1_32( 16 ) );
+      n += 16;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+   pdata[19] = n;
+   *hashes_done = n - first_nonce;
+   return 0;
+}
+
+#endif
+
 #if defined(SHA256T_8WAY)

 static __thread sha256_8way_context sha256_ctx8 __attribute__ ((aligned (64)));
@@ -31,68 +104,47 @@ void sha256q_8way_hash( void* output, const void* input )
   sha256_8way_close( &ctx, output );
 }

-int scanhash_sha256q_8way( struct work *work, uint32_t max_nonce,
-	                   uint64_t *hashes_done, struct thr_info *mythr )
+int scanhash_sha256q_8way( struct work *work, const uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr )
 {
-   uint32_t vdata[20*8] __attribute__ ((aligned (64)));
-   uint32_t hash[8*8] __attribute__ ((aligned (32)));
+   uint32_t vdata[20*8]  __attribute__ ((aligned (64)));
+   uint32_t hash32[8*8]    __attribute__ ((aligned (32)));
   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash32_d7 = &(hash32[7<<3]);
   uint32_t *pdata = work->data;
-   uint32_t *ptarget = work->target;
-   const uint32_t Htarg = ptarget[7];
+   const uint32_t *ptarget = work->target;
+   const uint32_t targ32_d7 = ptarget[7];
   const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce - 8;
   uint32_t n = first_nonce;
   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
-   int thr_id = mythr->id;  // thr_id arg is deprecated
+   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;

-   const uint64_t htmax[] = {          0,
-                                     0xF,
-                                    0xFF,
-                                   0xFFF,
-                                  0xFFFF,
-                              0x10000000 };
-   const uint32_t masks[] = {  0xFFFFFFFF,
-                               0xFFFFFFF0,
-                               0xFFFFFF00,
-                               0xFFFFF000,
-                               0xFFFF0000,
-                                        0 };
-
-   // Need big endian data
   mm256_bswap32_intrlv80_8x32( vdata, pdata );
+   *noncev = _mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n );
   sha256_8way_init( &sha256_ctx8 );
   sha256_8way_update( &sha256_ctx8, vdata, 64 );

-   for ( int m = 0; m < 6; m++ ) if ( Htarg <= htmax[m] )
+   do
   {
-      uint32_t mask = masks[m];
-      do
-      {
-         *noncev = mm256_bswap_32(
-		            _mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n ) );
-
-	      pdata[19] = n;
-         sha256q_8way_hash( hash, vdata );
-
-         uint32_t *hash7 = &(hash[7<<3]); 
-	 
-         for ( int lane = 0; lane < 8; lane++ )
-         if ( !( hash7[ lane ] & mask ) )
-         { 
-            // deinterleave hash for lane
-	         extr_lane_8x32( lane_hash, hash, lane, 256 );
-
-	         if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
-            {
-	           pdata[19] = n + lane;
-              submit_solution( work, lane_hash, mythr );
-            }
-	      }
-         n += 8;
-      } while ( (n < max_nonce-10) && !work_restart[thr_id].restart );
-      break;
-   }
-   *hashes_done = n - first_nonce + 1;
+     pdata[19] = n;
+     sha256q_8way_hash( hash32, vdata );
+     for ( int lane = 0; lane < 8; lane++ )
+     if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
+     {
+        extr_lane_8x32( lane_hash, hash32, lane, 256 );
+        if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+        {
+           pdata[19] = bswap_32( n + lane );
+           submit_solution( work, lane_hash, mythr );
+        }
+      }
+      *noncev = _mm256_add_epi32( *noncev, m256_const1_32( 8 ) );
+      n += 8;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+   pdata[19] = n;
+   *hashes_done = n - first_nonce;
   return 0;
 }

--- a/algo/sha/sha256q.c
+++ b/algo/sha/sha256q.c
@@ -1,108 +1,66 @@
 #include "sha256t-gate.h"
-
-#if !defined(SHA256T_16WAY) && !defined(SHA256T_8WAY) && !defined(SHA256T_4WAY)
-
 #include <stdlib.h>
 #include <stdint.h>
 #include <string.h>
 #include <stdio.h>
-#include <openssl/sha.h>
+#include "algo/sha/sha256-hash.h"

-static __thread SHA256_CTX sha256q_ctx __attribute__ ((aligned (64)));
+static __thread sha256_context sha256q_ctx __attribute__ ((aligned (64)));

 void sha256q_midstate( const void* input )
 {
-    SHA256_Init( &sha256q_ctx );
-    SHA256_Update( &sha256q_ctx, input, 64 );
+   sha256_ctx_init( &sha256q_ctx );
+   sha256_update( &sha256q_ctx, input, 64 );
 }

-void sha256q_hash( void* output, const void* input )
+int sha256q_hash( void* output, const void* input )
 {
   uint32_t _ALIGN(64) hash[16];
   const int midlen = 64;            // bytes
   const int tail   = 80 - midlen;   // 16

-   SHA256_CTX ctx __attribute__ ((aligned (64)));
+   sha256_context ctx __attribute__ ((aligned (64)));
   memcpy( &ctx, &sha256q_ctx, sizeof sha256q_ctx );

-   SHA256_Update( &ctx, input + midlen, tail );
-   SHA256_Final( (unsigned char*)hash, &ctx );
+   sha256_update( &ctx, input + midlen, tail );
+   sha256_final( &ctx, hash );

-   SHA256_Init( &ctx );
-   SHA256_Update( &ctx, hash, 32 );
-   SHA256_Final( (unsigned char*)hash, &ctx );
-
-   SHA256_Init( &ctx );
-   SHA256_Update( &ctx, hash, 32 );
-   SHA256_Final( (unsigned char*)hash, &ctx );
-
-   SHA256_Init( &ctx );
-   SHA256_Update( &ctx, hash, 32 );
-   SHA256_Final( (unsigned char*)hash, &ctx );
-
-   memcpy( output, hash, 32 );
+   sha256_full( hash,   hash, 32 );
+   sha256_full( hash,   hash, 32 );
+   sha256_full( output, hash, 32 );
+   
+   return 1;
 }

 int scanhash_sha256q( struct work *work, uint32_t max_nonce,
                      uint64_t *hashes_done, struct thr_info *mythr )
 {
+   uint32_t edata[20] __attribute__((aligned(64)));
+   uint32_t hash[8] __attribute__((aligned(64)));
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
-   uint32_t n = pdata[19] - 1;
   const uint32_t first_nonce = pdata[19];
-   const uint32_t Htarg = ptarget[7];
-#ifdef _MSC_VER
-   uint32_t __declspec(align(32)) hash64[8];
-#else
-   uint32_t hash64[8] __attribute__((aligned(32)));
-#endif
-   uint32_t endiandata[32];
-   int thr_id = mythr->id;  // thr_id arg is deprecated
+   const uint32_t last_nonce = max_nonce - 1;
+   uint32_t n = first_nonce;
+   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;

-   uint64_t htmax[] = {
-		0,
-		0xF,
-		0xFF,
-		0xFFF,
-		0xFFFF,
-		0x10000000
-	};
-   uint32_t masks[] = {
-		0xFFFFFFFF,
-		0xFFFFFFF0,
-		0xFFFFFF00,
-		0xFFFFF000,
-		0xFFFF0000,
-		0
-	};
+   mm128_bswap32_80( edata, pdata );
+   sha256q_midstate( edata );

-   // we need bigendian data...
-   casti_m128i( endiandata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
-   casti_m128i( endiandata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
-   casti_m128i( endiandata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
-   casti_m128i( endiandata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
-   casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
-
-   sha256q_midstate( endiandata );
-
-   for ( int m = 0; m < 6; m++ )
+   do
   {
-      if ( Htarg <= htmax[m] )
+      edata[19] = n;
+      if ( likely( sha256q_hash( hash, edata ) ) )
+      if ( unlikely( valid_hash( hash, ptarget ) && !bench ) )
      {
-         uint32_t mask = masks[m];
-         do {
-            pdata[19] = ++n;
-            be32enc(&endiandata[19], n);
-            sha256q_hash( hash64, endiandata );
-            if ( !( hash64[7] & mask ) )
-            if ( fulltest( hash64, ptarget ) && !opt_benchmark )
-               submit_solution( work, hash64, mythr );
-         } while ( n < max_nonce && !work_restart[thr_id].restart );
-         break;
+         pdata[19] = bswap_32( n );
+         submit_solution( work, hash, mythr );
      }
-   }
-   *hashes_done = n - first_nonce + 1;
+      n++;
+   } while ( n < last_nonce && !work_restart[thr_id].restart );
+   *hashes_done = n - first_nonce;
   pdata[19] = n;
   return 0;
 }
-#endif
+
--- a/algo/sha/sha256t-4way.c
+++ b/algo/sha/sha256t-4way.c
@@ -5,87 +5,177 @@
 #include <stdio.h>
 #include "sha-hash-4way.h"

-#if defined(SHA256T_8WAY)
+#if defined(SHA256T_16WAY)

-static __thread sha256_8way_context sha256_ctx8 __attribute__ ((aligned (64)));
-
-void sha256t_8way_hash( void* output, const void* input )
+int scanhash_sha256t_16way( struct work *work, const uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr )
 {
-   uint32_t vhash[8*8] __attribute__ ((aligned (64)));
-   sha256_8way_context ctx;
-   memcpy( &ctx, &sha256_ctx8, sizeof ctx );
+   __m512i  block[16]    __attribute__ ((aligned (64)));
+   __m512i  hash32[8]    __attribute__ ((aligned (32)));
+   __m512i  initstate[8] __attribute__ ((aligned (32)));
+   __m512i  midstate[8]  __attribute__ ((aligned (32)));
+   __m512i  midstate2[8] __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   __m512i  vdata[20]    __attribute__ ((aligned (32)));
+   uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
+   uint32_t *pdata = work->data;
+   const uint32_t *ptarget = work->target;
+   const uint32_t targ32_d7 = ptarget[7];
+   const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce - 16;
+   uint32_t n = first_nonce;
+   __m512i *noncev = vdata + 19; 
+   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;
+   const __m512i last_byte = m512_const1_32( 0x80000000 );
+   const __m512i sixteen = m512_const1_32( 16 );

-   sha256_8way_update( &ctx, input + (64<<3), 16 );
-   sha256_8way_close( &ctx, vhash );
+   for ( int i = 0; i < 19; i++ )
+       vdata[i] = m512_const1_32( pdata[i] );

-   sha256_8way_init( &ctx );
-   sha256_8way_update( &ctx, vhash, 32 );
-   sha256_8way_close( &ctx, vhash );
+   *noncev = _mm512_set_epi32( n+15, n+14, n+13, n+12, n+11, n+10, n+9, n+8,
+                               n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n+1, n );

-   sha256_8way_init( &ctx );
-   sha256_8way_update( &ctx, vhash, 32 );
-   sha256_8way_close( &ctx, output );
+   // initialize state
+   initstate[0] = m512_const1_64( 0x6A09E6676A09E667 );
+   initstate[1] = m512_const1_64( 0xBB67AE85BB67AE85 );
+   initstate[2] = m512_const1_64( 0x3C6EF3723C6EF372 );
+   initstate[3] = m512_const1_64( 0xA54FF53AA54FF53A );
+   initstate[4] = m512_const1_64( 0x510E527F510E527F );
+   initstate[5] = m512_const1_64( 0x9B05688C9B05688C );
+   initstate[6] = m512_const1_64( 0x1F83D9AB1F83D9AB );
+   initstate[7] = m512_const1_64( 0x5BE0CD195BE0CD19 );
+
+   // hash first 64 byte block of data
+   sha256_16way_transform_le( midstate, vdata, initstate );
+
+   // Do 3 rounds on the first 12 bytes of the next block
+   sha256_16way_prehash_3rounds( midstate2, vdata + 16, midstate );
+
+   do
+   {
+      // 1. final 16 bytes of data, with padding
+      memcpy_512( block, vdata + 16, 4 );
+      block[ 4] = last_byte;
+      memset_zero_512( block + 5, 10 );  
+      block[15] = m512_const1_32( 80*8 ); // bit count
+      sha256_16way_final_rounds( hash32, block, midstate, midstate2 );
+
+      // 2. 32 byte hash from 1.
+      memcpy_512( block, hash32, 8 );
+      block[ 8] = last_byte;
+      memset_zero_512( block + 9, 6 );
+      block[15] = m512_const1_32( 32*8 ); // bit count
+      sha256_16way_transform_le( hash32, block, initstate );
+
+      // 3. 32 byte hash from 2.
+      memcpy_512( block, hash32, 8 );
+      sha256_16way_transform_le( hash32, block, initstate );
+
+      // byte swap final hash for testing
+      mm512_block_bswap_32( hash32, hash32 );    
+
+      for ( int lane = 0; lane < 16; lane++ )
+      if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
+      {
+         extr_lane_16x32( lane_hash, hash32, lane, 256 );
+         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
+         {
+            pdata[19] = n + lane;
+            submit_solution( work, lane_hash, mythr );
+         }
+       }
+       *noncev = _mm512_add_epi32( *noncev, sixteen );
+       n += 16;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+   pdata[19] = n;
+   *hashes_done = n - first_nonce;
+   return 0;
 }

+
+#endif
+
+#if defined(SHA256T_8WAY)
+
 int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
                           uint64_t *hashes_done, struct thr_info *mythr )
 {
-   uint32_t vdata[20*8]  __attribute__ ((aligned (64)));
-   uint32_t hash[8*8]    __attribute__ ((aligned (32)));
+   __m256i  block[16]    __attribute__ ((aligned (64)));
+   __m256i  hash32[8]    __attribute__ ((aligned (32)));
+   __m256i  initstate[8] __attribute__ ((aligned (32)));
+   __m256i  midstate[8]  __attribute__ ((aligned (32)));
   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
-   uint32_t *hash7 = &(hash[7<<3]);
+   __m256i  vdata[20]    __attribute__ ((aligned (32)));
+   uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
   uint32_t *pdata = work->data;
   const uint32_t *ptarget = work->target;
-   const uint32_t Htarg = ptarget[7];
+   const uint32_t targ32_d7 = ptarget[7];
   const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce - 8;
   uint32_t n = first_nonce;
-   __m256i  *noncev = (__m256i*)vdata + 19;   // aligned
+   __m256i *noncev = vdata + 19;
   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;
+   const __m256i last_byte = m256_const1_32( 0x80000000 );
+   const __m256i eight = m256_const1_32( 8 );

-   const uint64_t htmax[] = {          0,
-                                     0xF,
-                                    0xFF,
-                                   0xFFF,
-                                  0xFFFF,
-                              0x10000000 };
-   const uint32_t masks[] = {  0xFFFFFFFF,
-                               0xFFFFFFF0,
-                               0xFFFFFF00,
-                               0xFFFFF000,
-                               0xFFFF0000,
-                                        0 };
+   for ( int i = 0; i < 19; i++ )
+       vdata[i] = m256_const1_32( pdata[i] );

+   *noncev = _mm256_set_epi32( n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n+1, n );

-   // Need big endian data
-   mm256_bswap32_intrlv80_8x32( vdata, pdata );
-   sha256_8way_init( &sha256_ctx8 );
-   sha256_8way_update( &sha256_ctx8, vdata, 64 );
+   // initialize state
+   initstate[0] = m256_const1_64( 0x6A09E6676A09E667 );
+   initstate[1] = m256_const1_64( 0xBB67AE85BB67AE85 );
+   initstate[2] = m256_const1_64( 0x3C6EF3723C6EF372 );
+   initstate[3] = m256_const1_64( 0xA54FF53AA54FF53A );
+   initstate[4] = m256_const1_64( 0x510E527F510E527F );
+   initstate[5] = m256_const1_64( 0x9B05688C9B05688C );
+   initstate[6] = m256_const1_64( 0x1F83D9AB1F83D9AB );
+   initstate[7] = m256_const1_64( 0x5BE0CD195BE0CD19 );

-   for ( int m = 0; m < 6; m++ ) if ( Htarg <= htmax[m] )
+   // hash first 64 bytes of data
+   sha256_8way_transform_le( midstate, vdata, initstate );
+
+   do
   {
-      const uint32_t mask = masks[m];
-      do
+      // 1. final 16 bytes of data, with padding
+      memcpy_256( block, vdata + 16, 4 );
+      block[ 4] = last_byte;
+      memset_zero_256( block + 5, 10 );
+      block[15] = m256_const1_32( 80*8 ); // bit count
+      sha256_8way_transform_le( hash32, block, midstate );
+
+      // 2. 32 byte hash from 1.
+      memcpy_256( block, hash32, 8 );
+      block[ 8] = last_byte;
+      memset_zero_256( block + 9, 6 );
+      block[15] = m256_const1_32( 32*8 ); // bit count
+      sha256_8way_transform_le( hash32, block, initstate );
+
+      // 3. 32 byte hash from 2.
+      memcpy_256( block, hash32, 8 );
+      sha256_8way_transform_le( hash32, block, initstate );
+
+      // byte swap final hash for testing
+      mm256_block_bswap_32( hash32, hash32 );
+
+      for ( int lane = 0; lane < 8; lane++ )
+      if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
      {
-        *noncev = mm256_bswap_32( _mm256_set_epi32(
-                                          n+7,n+6,n+5,n+4,n+3,n+2,n+1,n ) );
-         pdata[19] = n;
-         sha256t_8way_hash( hash, vdata );
-         for ( int lane = 0; lane < 8; lane++ )
-         if ( !( hash7[ lane ] & mask ) )
+         extr_lane_8x32( lane_hash, hash32, lane, 256 );
+         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
         {
-            // deinterleave hash for lane
-            extr_lane_8x32( lane_hash, hash, lane, 256 );
-            if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
-            {
-              pdata[19] = n + lane;
-              submit_solution( work, lane_hash, mythr );
-	         }
+            pdata[19] = n + lane;
+            submit_solution( work, lane_hash, mythr );
         }
-         n += 8;
-      } while ( (n < max_nonce-10) && !work_restart[thr_id].restart );
-      break;
-   }
-   *hashes_done = n - first_nonce + 1;
+       }
+       *noncev = _mm256_add_epi32( *noncev, eight );
+       n += 8;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+   pdata[19] = n;
+   *hashes_done = n - first_nonce;
   return 0;
 }

@@ -93,82 +183,84 @@ int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,

 #if defined(SHA256T_4WAY)

-static __thread sha256_4way_context sha256_ctx4 __attribute__ ((aligned (64)));
-
-void sha256t_4way_hash( void* output, const void* input )
-{
-   uint32_t vhash[8*4] __attribute__ ((aligned (64)));
-   sha256_4way_context ctx;
-   memcpy( &ctx, &sha256_ctx4, sizeof ctx );
-
-   sha256_4way_update( &ctx, input + (64<<2), 16 );
-   sha256_4way_close( &ctx, vhash );
-
-   sha256_4way_init( &ctx );
-   sha256_4way_update( &ctx, vhash, 32 );
-   sha256_4way_close( &ctx, vhash );
-
-   sha256_4way_init( &ctx );
-   sha256_4way_update( &ctx, vhash, 32 );
-   sha256_4way_close( &ctx, output );
-}
-
 int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
-	                   uint64_t *hashes_done, struct thr_info *mythr )
+                           uint64_t *hashes_done, struct thr_info *mythr )
 {
-   uint32_t vdata[20*4] __attribute__ ((aligned (64)));
-   uint32_t hash[8*4] __attribute__ ((aligned (32)));
-   uint32_t lane_hash[8] __attribute__ ((aligned (64)));
-   uint32_t *hash7 = &(hash[7<<2]);
+   __m128i  block[16]    __attribute__ ((aligned (64)));
+   __m128i  hash32[8]    __attribute__ ((aligned (32)));
+   __m128i  initstate[8] __attribute__ ((aligned (32)));
+   __m128i  midstate[8]  __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   __m128i  vdata[20]    __attribute__ ((aligned (32)));
+   uint32_t *hash32_d7 =  (uint32_t*)&( hash32[7] );
   uint32_t *pdata = work->data;
   const uint32_t *ptarget = work->target;
-   const uint32_t Htarg = ptarget[7];
+   const uint32_t targ32_d7 = ptarget[7];
   const uint32_t first_nonce = pdata[19];
+   const uint32_t last_nonce = max_nonce - 4;
   uint32_t n = first_nonce;
-   __m128i  *noncev = (__m128i*)vdata + 19;   // aligned
+   __m128i *noncev = vdata + 19;
   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;
+   const __m128i last_byte = m128_const1_32( 0x80000000 );
+   const __m128i four = m128_const1_32( 4 );

-   const uint64_t htmax[] = {          0,
-                                     0xF,
-                                    0xFF,
-                                   0xFFF,
-                                  0xFFFF,
-                              0x10000000 };
-   const uint32_t masks[] = { 0xFFFFFFFF,
-                              0xFFFFFFF0,
-                              0xFFFFFF00,
-                              0xFFFFF000,
-                              0xFFFF0000,
-                                       0 };
+   for ( int i = 0; i < 19; i++ )
+       vdata[i] = m128_const1_32( pdata[i] );

-   mm128_bswap32_intrlv80_4x32( vdata, pdata );
-   sha256_4way_init( &sha256_ctx4 );
-   sha256_4way_update( &sha256_ctx4, vdata, 64 );
+   *noncev = _mm_set_epi32( n+ 3, n+ 2, n+1, n );

-   for ( int m = 0; m < 6; m++ ) if ( Htarg <= htmax[m] )
+   // initialize state
+   initstate[0] = m128_const1_64( 0x6A09E6676A09E667 );
+   initstate[1] = m128_const1_64( 0xBB67AE85BB67AE85 );
+   initstate[2] = m128_const1_64( 0x3C6EF3723C6EF372 );
+   initstate[3] = m128_const1_64( 0xA54FF53AA54FF53A );
+   initstate[4] = m128_const1_64( 0x510E527F510E527F );
+   initstate[5] = m128_const1_64( 0x9B05688C9B05688C );
+   initstate[6] = m128_const1_64( 0x1F83D9AB1F83D9AB );
+   initstate[7] = m128_const1_64( 0x5BE0CD195BE0CD19 );
+
+   // hash first 64 bytes of data
+   sha256_4way_transform_le( midstate, vdata, initstate );
+
+   do
   {
-      const uint32_t mask = masks[m];
-      do {
-         *noncev = mm128_bswap_32( _mm_set_epi32( n+3,n+2,n+1,n ) );
-         pdata[19] = n;
+      // 1. final 16 bytes of data, with padding
+      memcpy_128( block, vdata + 16, 4 );
+      block[ 4] = last_byte;
+      memset_zero_128( block + 5, 10 );
+      block[15] = m128_const1_32( 80*8 ); // bit count
+      sha256_4way_transform_le( hash32, block, midstate );

-         sha256t_4way_hash( hash, vdata );
+      // 2. 32 byte hash from 1.
+      memcpy_128( block, hash32, 8 );
+      block[ 8] = last_byte;
+      memset_zero_128( block + 9, 6 );
+      block[15] = m128_const1_32( 32*8 ); // bit count
+      sha256_4way_transform_le( hash32, block, initstate );

-         for ( int lane = 0; lane < 4; lane++ )
-         if ( !( hash7[ lane ] & mask ) )
+      // 3. 32 byte hash from 2.
+      memcpy_128( block, hash32, 8 );
+      sha256_4way_transform_le( hash32, block, initstate );
+
+      // byte swap final hash for testing
+      mm128_block_bswap_32( hash32, hash32 );
+
+      for ( int lane = 0; lane < 4; lane++ )
+      if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
+      {
+         extr_lane_4x32( lane_hash, hash32, lane, 256 );
+         if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
         {
-            extr_lane_4x32( lane_hash, hash, lane, 256 );
-            if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
-            {
-              pdata[19] = n + lane;
-              submit_solution( work, lane_hash, mythr );
-	         }
-         } 
-         n += 4;
-      } while ( (n < max_nonce - 4) && !work_restart[thr_id].restart );
-      break;
-   }
-   *hashes_done = n - first_nonce + 1;
+            pdata[19] = n + lane;
+            submit_solution( work, lane_hash, mythr );
+         }
+       }
+       *noncev = _mm_add_epi32( *noncev, four );
+       n += 4;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+   pdata[19] = n;
+   *hashes_done = n - first_nonce;
   return 0;
 }

--- a/algo/sha/sha256t-gate.c
+++ b/algo/sha/sha256t-gate.c
@@ -2,38 +2,37 @@

 bool register_sha256t_algo( algo_gate_t* gate )
 {
-#if defined(SHA256T_8WAY)
-    gate->optimizations = SSE2_OPT | AVX2_OPT | SHA_OPT;
-    gate->scanhash   = (void*)&scanhash_sha256t_8way;
-    gate->hash       = (void*)&sha256t_8way_hash;
-#elif defined(SHA256T_4WAY)
-    gate->optimizations = SSE2_OPT | AVX2_OPT | SHA_OPT;
-    gate->scanhash   = (void*)&scanhash_sha256t_4way;
-    gate->hash       = (void*)&sha256t_4way_hash;
-#else
+    gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
+#if defined(SHA256T_16WAY)
+    gate->scanhash   = (void*)&scanhash_sha256t_16way;
+#elif defined(__SHA__)
    gate->optimizations = SHA_OPT;
    gate->scanhash   = (void*)&scanhash_sha256t;
-    gate->hash       = (void*)&sha256t_hash;
+#elif defined(SHA256T_8WAY)
+    gate->scanhash   = (void*)&scanhash_sha256t_8way;
+#else
+    gate->scanhash   = (void*)&scanhash_sha256t_4way;
 #endif
    return true;
 }

 bool register_sha256q_algo( algo_gate_t* gate )
 {
-#if defined(SHA256T_8WAY)
-    gate->optimizations = SSE2_OPT | AVX2_OPT | SHA_OPT;
-    gate->scanhash   = (void*)&scanhash_sha256q_8way;
-    gate->hash       = (void*)&sha256q_8way_hash;
-#elif defined(SHA256T_4WAY)
-    gate->optimizations = SSE2_OPT | AVX2_OPT | SHA_OPT;
-    gate->scanhash   = (void*)&scanhash_sha256q_4way;
-    gate->hash       = (void*)&sha256q_4way_hash;
-#else
+    gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
+#if defined(SHA256T_16WAY)
+    gate->scanhash   = (void*)&scanhash_sha256q_16way;
+    gate->hash       = (void*)&sha256q_16way_hash;
+#elif defined(__SHA__)
    gate->optimizations = SHA_OPT;
    gate->scanhash   = (void*)&scanhash_sha256q;
    gate->hash       = (void*)&sha256q_hash;
+#elif defined(SHA256T_8WAY)
+    gate->scanhash   = (void*)&scanhash_sha256q_8way;
+    gate->hash       = (void*)&sha256q_8way_hash;
+#else
+    gate->scanhash   = (void*)&scanhash_sha256q_4way;
+    gate->hash       = (void*)&sha256q_4way_hash;
 #endif
    return true;
-
 }

--- a/algo/sha/sha256t-gate.h
+++ b/algo/sha/sha256t-gate.h
@@ -4,21 +4,28 @@
 #include <stdint.h>
 #include "algo-gate-api.h"

-// Override multi way on ryzen, SHA is better.
-#if !defined(__SHA__)
- #if defined(__AVX2__)
-  #define SHA256T_8WAY
- #elif defined(__SSE2__)
-  #define SHA256T_4WAY
- #endif
+#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+  #define SHA256T_16WAY 1
+#elif defined(__AVX2__)
+  #define SHA256T_8WAY 1
+#else
+  #define SHA256T_4WAY 1
 #endif

 bool register_sha256t_algo( algo_gate_t* gate );
 bool register_sha256q_algo( algo_gate_t* gate );

+#if defined(SHA256T_16WAY)
+
+int scanhash_sha256t_16way( struct work *work, uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr );
+void sha256q_16way_hash( void *output, const void *input );
+int scanhash_sha256q_16way( struct work *work, uint32_t max_nonce,
+                           uint64_t *hashes_done, struct thr_info *mythr );
+#endif
+
 #if defined(SHA256T_8WAY)

-void sha256t_8way_hash( void *output, const void *input );
 int scanhash_sha256t_8way( struct work *work, uint32_t max_nonce,
                           uint64_t *hashes_done, struct thr_info *mythr );
 void sha256q_8way_hash( void *output, const void *input );
@@ -28,7 +35,6 @@ int scanhash_sha256q_8way( struct work *work, uint32_t max_nonce,

 #if defined(SHA256T_4WAY)

-void sha256t_4way_hash( void *output, const void *input );
 int scanhash_sha256t_4way( struct work *work, uint32_t max_nonce,
                           uint64_t *hashes_done, struct thr_info *mythr );
 void sha256q_4way_hash( void *output, const void *input );
@@ -36,10 +42,14 @@ int scanhash_sha256q_4way( struct work *work, uint32_t max_nonce,
                           uint64_t *hashes_done, struct thr_info *mythr );
 #endif

-void sha256t_hash( void *output, const void *input );
+#if defined(__SHA__)
+
 int scanhash_sha256t( struct work *work, uint32_t max_nonce,
                      uint64_t *hashes_done, struct thr_info *mythr );
-void sha256q_hash( void *output, const void *input );
+
+#endif
+
+int sha256q_hash( void *output, const void *input );
 int scanhash_sha256q( struct work *work, uint32_t max_nonce,
                      uint64_t *hashes_done, struct thr_info *mythr );

--- a/algo/sha/sha256t.c
+++ b/algo/sha/sha256t.c
@@ -1,105 +1,102 @@
 #include "sha256t-gate.h"
-
-#if !defined(SHA256T_16WAY) && !defined(SHA256T_8WAY) && !defined(SHA256T_4WAY)
-
 #include <stdlib.h>
 #include <stdint.h>
 #include <string.h>
 #include <stdio.h>
-#include <openssl/sha.h>
+//#include "algo/sha/sph_sha2.h"
+#include "sha256-hash.h"

-static __thread SHA256_CTX sha256t_ctx __attribute__ ((aligned (64)));
+#if defined(__SHA__)

-void sha256t_midstate( const void* input )
-{
-    SHA256_Init( &sha256t_ctx );
-    SHA256_Update( &sha256t_ctx, input, 64 );
-}
+// Only used on CPUs with SHA

-void sha256t_hash( void* output, const void* input )
-{
-   uint32_t _ALIGN(64) hash[16];
-   const int midlen = 64;            // bytes
-   const int tail   = 80 - midlen;   // 16
-
-   SHA256_CTX ctx __attribute__ ((aligned (64)));
-   memcpy( &ctx, &sha256t_ctx, sizeof sha256t_ctx );
-
-   SHA256_Update( &ctx, input + midlen, tail );
-   SHA256_Final( (unsigned char*)hash, &ctx );
-
-   SHA256_Init( &ctx );
-   SHA256_Update( &ctx, hash, 32 );
-   SHA256_Final( (unsigned char*)hash, &ctx );
-
-   SHA256_Init( &ctx );
-   SHA256_Update( &ctx, hash, 32 );
-   SHA256_Final( (unsigned char*)hash, &ctx );
-
-   memcpy( output, hash, 32 );
-}

 int scanhash_sha256t( struct work *work, uint32_t max_nonce,
                      uint64_t *hashes_done, struct thr_info *mythr )
 {
+   uint32_t block0[16]    __attribute__ ((aligned (64)));
+   uint32_t block1[16]    __attribute__ ((aligned (64)));
+   uint32_t hash0[8]    __attribute__ ((aligned (32)));
+   uint32_t hash1[8]    __attribute__ ((aligned (32)));
+   uint32_t initstate[8] __attribute__ ((aligned (32)));
+   uint32_t midstate[8]  __attribute__ ((aligned (32)));
   uint32_t *pdata = work->data;
   uint32_t *ptarget = work->target;
-   uint32_t n = pdata[19] - 1;
   const uint32_t first_nonce = pdata[19];
-   const uint32_t Htarg = ptarget[7];
-#ifdef _MSC_VER
-   uint32_t __declspec(align(32)) hash64[8];
-#else
-   uint32_t hash64[8] __attribute__((aligned(32)));
-#endif
-   uint32_t endiandata[32];
-   int thr_id = mythr->id;  // thr_id arg is deprecated
+   const uint32_t last_nonce = max_nonce - 1;
+   uint32_t n = first_nonce;
+   const int thr_id = mythr->id;
+   const bool bench = opt_benchmark;
+   __m128i shuf_bswap32 =
+           _mm_set_epi64x( 0x0c0d0e0f08090a0bULL, 0x0405060700010203ULL );

-   uint64_t htmax[] = {
-		0,
-		0xF,
-		0xFF,
-		0xFFF,
-		0xFFFF,
-		0x10000000
-	};
-   uint32_t masks[] = {
-		0xFFFFFFFF,
-		0xFFFFFFF0,
-		0xFFFFFF00,
-		0xFFFFF000,
-		0xFFFF0000,
-		0
-	};
+   // initialize state
+   initstate[0] = 0x6A09E667;
+   initstate[1] = 0xBB67AE85;
+   initstate[2] = 0x3C6EF372;
+   initstate[3] = 0xA54FF53A;
+   initstate[4] = 0x510E527F;
+   initstate[5] = 0x9B05688C;
+   initstate[6] = 0x1F83D9AB;
+   initstate[7] = 0x5BE0CD19;

-   // we need bigendian data...
-   casti_m128i( endiandata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
-   casti_m128i( endiandata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
-   casti_m128i( endiandata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
-   casti_m128i( endiandata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
-   casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
+   // hash first 64 bytes of data
+   sha256_opt_transform_le( midstate, pdata, initstate );

-   sha256t_midstate( endiandata );
-
-   for ( int m = 0; m < 6; m++ )
+   do
   {
-      if ( Htarg <= htmax[m] )
+      // 1. final 16 bytes of data, with padding
+      memcpy( block0, pdata + 16, 16 );
+      memcpy( block1, pdata + 16, 16 );
+      block0[ 3] = n;
+      block1[ 3] = n+1;
+      block0[ 4] = block1[ 4] = 0x80000000;
+      memset( block0 + 5, 0, 40 );
+      memset( block1 + 5, 0, 40 );
+      block0[15] = block1[15] = 80*8; // bit count
+      sha256_ni2way_transform_le( hash0, hash1, block0, block1, midstate, midstate );
+
+      // 2. 32 byte hash from 1.
+      memcpy( block0, hash0, 32 );
+      memcpy( block1, hash1, 32 );
+      block0[ 8] = block1[ 8] = 0x80000000;
+      memset( block0 + 9, 0, 24 );
+      memset( block1 + 9, 0, 24 );
+      block0[15] = block1[15] = 32*8; // bit count
+      sha256_ni2way_transform_le( hash0, hash1, block0, block1, initstate, initstate );
+
+      // 3. 32 byte hash from 2.
+      memcpy( block0, hash0, 32 );
+      memcpy( block1, hash1, 32 );
+      sha256_ni2way_transform_le( hash0, hash1, block0, block1, initstate, initstate );
+
+      // byte swap final hash for testing
+      casti_m128i( hash0, 0 ) =
+               _mm_shuffle_epi8( casti_m128i( hash0, 0 ), shuf_bswap32 );
+      casti_m128i( hash0, 1 ) =
+               _mm_shuffle_epi8( casti_m128i( hash0, 1 ), shuf_bswap32 );
+      casti_m128i( hash1, 0 ) =
+               _mm_shuffle_epi8( casti_m128i( hash1, 0 ), shuf_bswap32 );
+      casti_m128i( hash1, 1 ) =
+               _mm_shuffle_epi8( casti_m128i( hash1, 1 ), shuf_bswap32 );
+
+      if ( unlikely( valid_hash( hash0, ptarget ) && !bench ) )
      {
-         uint32_t mask = masks[m];
-         do {
-            pdata[19] = ++n;
-            be32enc(&endiandata[19], n);
-            sha256t_hash( hash64, endiandata );
-            if ( !(hash64[7] & mask) )
-            if ( fulltest( hash64, ptarget ) && !opt_benchmark )
-               submit_solution( work, hash64, mythr );
-         } while ( n < max_nonce && !work_restart[thr_id].restart );
-         break;
+         pdata[19] = n;
+         submit_solution( work, hash0, mythr );
      }
-   }
-   *hashes_done = n - first_nonce + 1;
+      if ( unlikely( valid_hash( hash1, ptarget ) && !bench ) )
+      {
+         pdata[19] = n+1;
+         submit_solution( work, hash1, mythr );
+      }
+      n += 2;
+   } while ( (n < last_nonce) && !work_restart[thr_id].restart );
+
   pdata[19] = n;
+   *hashes_done = n - first_nonce;
   return 0;
 }
+
 #endif

--- a/algo/sha/sha512-hash-4way.c
+++ b/algo/sha/sha512-hash-4way.c
@@ -95,84 +95,36 @@ static const uint64_t K512[80] =

 // SHA-512 8 way 64 bit

-#define CH8W(X, Y, Z) \
-   _mm512_xor_si512( _mm512_and_si512( _mm512_xor_si512( Y, Z ), X ), Z ) 
+#define CH8W( X, Y, Z )    _mm512_ternarylogic_epi64( X, Y, Z, 0xca )

-#define MAJ8W(X, Y, Z) \
-   _mm512_or_si512( _mm512_and_si512( X, Y ), \
-                    _mm512_and_si512( _mm512_or_si512( X, Y ), Z ) )
+#define MAJ8W( X, Y, Z )   _mm512_ternarylogic_epi64( X, Y, Z, 0xe8 )

-#define BSG8W_5_0(x) \
-   _mm512_xor_si512( _mm512_xor_si512( \
-        mm512_ror_64(x, 28), mm512_ror_64(x, 34) ), mm512_ror_64(x, 39) )
+#define BSG8W_5_0( x )     mm512_xor3( _mm512_ror_epi64( x, 28 ), \
+                                       _mm512_ror_epi64( x, 34 ), \
+                                       _mm512_ror_epi64( x, 39 ) )

-#define BSG8W_5_1(x) \
-   _mm512_xor_si512( _mm512_xor_si512( \
-        mm512_ror_64(x, 14), mm512_ror_64(x, 18) ), mm512_ror_64(x, 41) )
+#define BSG8W_5_1( x )     mm512_xor3( _mm512_ror_epi64( x, 14 ), \
+                                       _mm512_ror_epi64( x, 18 ), \
+                                       _mm512_ror_epi64( x, 41 ) )

-#define SSG8W_5_0(x) \
-   _mm512_xor_si512( _mm512_xor_si512( \
-        mm512_ror_64(x,  1), mm512_ror_64(x,  8) ), _mm512_srli_epi64(x, 7) ) 
+#define SSG8W_5_0( x )     mm512_xor3( _mm512_ror_epi64( x,  1 ), \
+                                       _mm512_ror_epi64( x,  8 ), \
+                                       _mm512_srli_epi64( x, 7 ) ) 

-#define SSG8W_5_1(x) \
-   _mm512_xor_si512( _mm512_xor_si512( \
-        mm512_ror_64(x, 19), mm512_ror_64(x, 61) ), _mm512_srli_epi64(x, 6) )
+#define SSG8W_5_1( x )     mm512_xor3( _mm512_ror_epi64( x, 19 ), \
+                                       _mm512_ror_epi64( x, 61 ), \
+                                       _mm512_srli_epi64( x, 6 ) )

-static inline __m512i ssg8w_512_add( __m512i w0, __m512i w1 )
-{
-   __m512i w0a, w1a, w0b, w1b;
-   w0a = mm512_ror_64( w0, 1 );
-   w1a = mm512_ror_64( w1,19 );
-   w0b = mm512_ror_64( w0, 8 );
-   w1b = mm512_ror_64( w1,61 );
-   w0a = _mm512_xor_si512( w0a, w0b );
-   w1a = _mm512_xor_si512( w1a, w1b );
-   w0b = _mm512_srli_epi64( w0, 7 );
-   w1b = _mm512_srli_epi64( w1, 6 );
-   w0a = _mm512_xor_si512( w0a, w0b );
-   w1a = _mm512_xor_si512( w1a, w1b );
-   return _mm512_add_epi64( w0a, w1a );
-}
-
-
-#define SSG8W_512x2_0( w0, w1, i ) do \
-{ \
-   __m512i X0a, X1a, X0b, X1b; \
-  X0a = mm512_ror_64( W[i-15], 1 ); \
-  X1a = mm512_ror_64( W[i-14], 1 ); \
-  X0b = mm512_ror_64( W[i-15], 8 ); \
-  X1b = mm512_ror_64( W[i-14], 8 ); \
-  X0a = _mm512_xor_si512( X0a, X0b ); \
-  X1a = _mm512_xor_si512( X1a, X1b ); \
-  X0b = _mm512_srli_epi64( W[i-15], 7 ); \
-  X1b = _mm512_srli_epi64( W[i-14], 7 ); \
-  w0  = _mm512_xor_si512( X0a, X0b ); \
-  w1  = _mm512_xor_si512( X1a, X1b ); \
-} while(0)
-
-#define SSG8W_512x2_1( w0, w1, i ) do \
-{ \
-   __m512i X0a, X1a, X0b, X1b; \
-  X0a = mm512_ror_64( W[i-2],19 ); \
-  X1a = mm512_ror_64( W[i-1],19 ); \
-  X0b = mm512_ror_64( W[i-2],61 ); \
-  X1b = mm512_ror_64( W[i-1],61 ); \
-  X0a = _mm512_xor_si512( X0a, X0b ); \
-  X1a = _mm512_xor_si512( X1a, X1b ); \
-  X0b = _mm512_srli_epi64( W[i-2], 6 ); \
-  X1b = _mm512_srli_epi64( W[i-1], 6 ); \
-  w0  = _mm512_xor_si512( X0a, X0b ); \
-  w1  = _mm512_xor_si512( X1a, X1b ); \
-} while(0)
-
-#define SHA3_8WAY_STEP(A, B, C, D, E, F, G, H, i) \
+#define SHA3_8WAY_STEP( A, B, C, D, E, F, G, H, i ) \
 do { \
-  __m512i T1, T2; \
-  __m512i K = _mm512_set1_epi64( K512[ i ] ); \
-  T1 = _mm512_add_epi64( H, mm512_add4_64( BSG8W_5_1(E), CH8W(E, F, G), \
-                                           K, W[i] ) ); \
-  T2 = _mm512_add_epi64( BSG8W_5_0(A), MAJ8W(A, B, C) ); \
-  D  = _mm512_add_epi64( D, T1 ); \
+  __m512i T0 = _mm512_add_epi64( _mm512_set1_epi64( K512[i] ), W[ i ] ); \
+  __m512i T1 = BSG8W_5_1( E ); \
+  __m512i T2 = BSG8W_5_0( A ); \
+  T0 = _mm512_add_epi64( T0, CH8W( E, F, G ) ); \
+  T1 = _mm512_add_epi64( T1, H ); \
+  T2 = _mm512_add_epi64( T2, MAJ8W( A, B, C ) ); \
+  T1 = _mm512_add_epi64( T1, T0 ); \
+  D  = _mm512_add_epi64( D,  T1 ); \
  H  = _mm512_add_epi64( T1, T2 ); \
 } while (0)

@@ -187,8 +139,8 @@ sha512_8way_round( sha512_8way_context *ctx,  __m512i *in, __m512i r[8] )
   mm512_block_bswap_64( W+8, in+8 );

   for ( i = 16; i < 80; i++ )
-      W[i] = _mm512_add_epi64( ssg8w_512_add( W[i-15], W[i-2] ),
-                               _mm512_add_epi64( W[ i- 7 ], W[ i-16 ] ) );
+      W[i] = mm512_add4_64( SSG8W_5_0( W[i-15] ), SSG8W_5_1( W[i-2] ),
+                             W[ i- 7 ], W[ i-16 ] );

   if ( ctx->initialized )
   {
@@ -319,14 +271,13 @@ void sha512_8way_close( sha512_8way_context *sc, void *dst )

 // SHA-512 4 way 64 bit

-/*
 #define CH(X, Y, Z) \
   _mm256_xor_si256( _mm256_and_si256( _mm256_xor_si256( Y, Z ), X ), Z ) 

 #define MAJ(X, Y, Z) \
-   _mm256_or_si256( _mm256_and_si256( X, Y ), \
-                    _mm256_and_si256( _mm256_or_si256( X, Y ), Z ) )
-
+  _mm256_xor_si256( Y, _mm256_and_si256( X_xor_Y = _mm256_xor_si256( X, Y ), \
+                                         Y_xor_Z ) )
+                    
 #define BSG5_0(x) \
  mm256_ror_64( _mm256_xor_si256( mm256_ror_64( \
                   _mm256_xor_si256( mm256_ror_64( x,  5 ), x ), 6 ), x ), 28 )
@@ -334,16 +285,7 @@ void sha512_8way_close( sha512_8way_context *sc, void *dst )
 #define BSG5_1(x) \
  mm256_ror_64( _mm256_xor_si256( mm256_ror_64( \
                   _mm256_xor_si256( mm256_ror_64( x, 23 ), x ), 4 ), x ), 14 )
-*/
-/*
-#define BSG5_0(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-        mm256_ror_64(x, 28), mm256_ror_64(x, 34) ), mm256_ror_64(x, 39) )

-#define BSG5_1(x) \
-   _mm256_xor_si256( _mm256_xor_si256( \
-        mm256_ror_64(x, 14), mm256_ror_64(x, 18) ), mm256_ror_64(x, 41) )
-*/
 /*
 #define SSG5_0(x) \
   _mm256_xor_si256( _mm256_xor_si256( \
@@ -371,98 +313,25 @@ static inline __m256i ssg512_add( __m256i w0, __m256i w1 )
   return _mm256_add_epi64( w0a, w1a );
 }

-/*
-#define SSG512x2_0( w0, w1, i ) do \
-{ \
-   __m256i X0a, X1a, X0b, X1b; \
-  X0a = mm256_ror_64( W[i-15], 1 ); \
-  X1a = mm256_ror_64( W[i-14], 1 ); \
-  X0b = mm256_ror_64( W[i-15], 8 ); \
-  X1b = mm256_ror_64( W[i-14], 8 ); \
-  X0a = _mm256_xor_si256( X0a, X0b ); \
-  X1a = _mm256_xor_si256( X1a, X1b ); \
-  X0b = _mm256_srli_epi64( W[i-15], 7 ); \
-  X1b = _mm256_srli_epi64( W[i-14], 7 ); \
-  w0  = _mm256_xor_si256( X0a, X0b ); \
-  w1  = _mm256_xor_si256( X1a, X1b ); \
-} while(0)
-
-#define SSG512x2_1( w0, w1, i ) do \
-{ \
-   __m256i X0a, X1a, X0b, X1b; \
-  X0a = mm256_ror_64( W[i-2],19 ); \
-  X1a = mm256_ror_64( W[i-1],19 ); \
-  X0b = mm256_ror_64( W[i-2],61 ); \
-  X1b = mm256_ror_64( W[i-1],61 ); \
-  X0a = _mm256_xor_si256( X0a, X0b ); \
-  X1a = _mm256_xor_si256( X1a, X1b ); \
-  X0b = _mm256_srli_epi64( W[i-2], 6 ); \
-  X1b = _mm256_srli_epi64( W[i-1], 6 ); \
-  w0  = _mm256_xor_si256( X0a, X0b ); \
-  w1  = _mm256_xor_si256( X1a, X1b ); \
-} while(0)
-*/
-
-#define SHA3_4WAY_STEP(A, B, C, D, E, F, G, H, i) \
+#define SHA3_4WAY_STEP( A, B, C, D, E, F, G, H, i ) \
 do { \
-  __m256i K = _mm256_set1_epi64x( K512[ i ] ); \
-  __m256i T1 = mm256_ror_64( E, 23 ); \
-  __m256i T2 = mm256_ror_64( A,  5 ); \
-  __m256i T3 = _mm256_xor_si256( F, G ); \
-  __m256i T4 = _mm256_or_si256( A, B ); \
-  __m256i T5 = _mm256_and_si256( A, B ); \
-  K = _mm256_add_epi64( K, W[i] ); \
-  T1 = _mm256_xor_si256( T1, E ); \
-  T2 = _mm256_xor_si256( T2, A ); \
-  T3 = _mm256_and_si256( T3, E ); \
-  T4 = _mm256_and_si256( T4, C ); \
-  K = _mm256_add_epi64( H, K ); \
-  T1 = mm256_ror_64( T1, 4 ); \
-  T2 = mm256_ror_64( T2, 6 ); \
-  T3 = _mm256_xor_si256( T3, G ); \
-  T4 = _mm256_or_si256( T4, T5 ); \
-  T1 = _mm256_xor_si256( T1, E ); \
-  T2 = _mm256_xor_si256( T2, A ); \
-  T1 = mm256_ror_64( T1, 14 ); \
-  T2 = mm256_ror_64( T2, 28 ); \
-  T1 = _mm256_add_epi64( T1, T3 ); \
-  T2 = _mm256_add_epi64( T2, T4 ); \
-  T1 = _mm256_add_epi64( T1, K ); \
-  H  = _mm256_add_epi64( T1, T2 ); \
-  D  = _mm256_add_epi64( D, T1 ); \
-} while (0)
-
-/*
-#define SHA3_4WAY_STEP(A, B, C, D, E, F, G, H, i) \
-do { \
-  __m256i K = _mm256_add_epi64( W[i], _mm256_set1_epi64x( K512[ i ] ) ); \
-  __m256i T1 = BSG5_1(E); \
-  __m256i T2 = BSG5_0(A); \
-  T1 = mm256_add4_64( T1, H, CH(E, F, G), K ); \
-  T2 = _mm256_add_epi64( T2, MAJ(A, B, C) ); \
-  D  = _mm256_add_epi64( D, T1 ); \
+  __m256i T0 = _mm256_add_epi64( _mm256_set1_epi64x( K512[i] ), W[ i ] ); \
+  __m256i T1 = BSG5_1( E ); \
+  __m256i T2 = BSG5_0( A ); \
+  T0 = _mm256_add_epi64( T0, CH( E, F, G ) ); \
+  T1 = _mm256_add_epi64( T1, H ); \
+  T2 = _mm256_add_epi64( T2, MAJ( A, B, C ) ); \
+  T1 = _mm256_add_epi64( T1, T0 ); \
+  Y_xor_Z = X_xor_Y; \
+  D  = _mm256_add_epi64( D,  T1 ); \
  H  = _mm256_add_epi64( T1, T2 ); \
 } while (0)
-*/
-
-/*
-#define SHA3_4WAY_STEP(A, B, C, D, E, F, G, H, i) \
-do { \
-  __m256i T1, T2; \
-  __m256i K = _mm256_set1_epi64x( K512[ i ] ); \
-  T1 = _mm256_add_epi64( H, mm256_add4_64( BSG5_1(E), CH(E, F, G), \
-                                           K, W[i] ) ); \
-  T2 = _mm256_add_epi64( BSG5_0(A), MAJ(A, B, C) ); \
-  D  = _mm256_add_epi64( D, T1 ); \
-  H  = _mm256_add_epi64( T1, T2 ); \
-} while (0)
-*/

 static void
 sha512_4way_round( sha512_4way_context *ctx,  __m256i *in, __m256i r[8] )
 {
   int i;
-   register __m256i A, B, C, D, E, F, G, H;
+   register __m256i A, B, C, D, E, F, G, H, X_xor_Y, Y_xor_Z;
   __m256i W[80];

   mm256_block_bswap_64( W  , in );
@@ -495,6 +364,8 @@ sha512_4way_round( sha512_4way_context *ctx,  __m256i *in, __m256i r[8] )
      H = m256_const1_64( 0x5BE0CD19137E2179 );
   }

+   Y_xor_Z = _mm256_xor_si256( B, C );
+
   for ( i = 0; i < 80; i += 8 )
   {
      SHA3_4WAY_STEP( A, B, C, D, E, F, G, H, i + 0 );
--- a/algo/sha/sph_sha2.c
+++ b/algo/sha/sph_sha2.c
@@ -40,8 +40,8 @@
 #endif

 #define CH(X, Y, Z)    ((((Y) ^ (Z)) & (X)) ^ (Z))
-#define MAJ(X, Y, Z)   (((Y) & (Z)) | (((Y) | (Z)) & (X)))
-
+//#define MAJ(X, Y, Z)   (((Y) & (Z)) | (((Y) | (Z)) & (X)))
+#define MAJ( X, Y, Z )   ( Y  ^ ( ( X_xor_Y = X ^ Y ) & ( Y_xor_Z ) ) )
 #define ROTR    SPH_ROTR32

 #define BSG2_0(x)      (ROTR(x, 2) ^ ROTR(x, 13) ^ ROTR(x, 22))
@@ -71,8 +71,8 @@ static const sph_u32 H256[8] = {
 * of the compression function.
 */

-#if SPH_SMALL_FOOTPRINT_SHA2

+/*
 static const sph_u32 K[64] = {
 	SPH_C32(0x428A2F98), SPH_C32(0x71374491),
 	SPH_C32(0xB5C0FBCF), SPH_C32(0xE9B5DBA5),
@@ -107,6 +107,9 @@ static const sph_u32 K[64] = {
 	SPH_C32(0x90BEFFFA), SPH_C32(0xA4506CEB),
 	SPH_C32(0xBEF9A3F7), SPH_C32(0xC67178F2)
 };
+*/
+
+#if SPH_SMALL_FOOTPRINT_SHA2

 #define SHA2_MEXP1(in, pc)   do { \
 		W[pc] = in(pc); \
@@ -124,6 +127,7 @@ static const sph_u32 K[64] = {
 		t1 = SPH_T32(h + BSG2_1(e) + CH(e, f, g) \
 			+ K[pcount + (pc)] + W[(pc) & 0x0F]); \
 		t2 = SPH_T32(BSG2_0(a) + MAJ(a, b, c)); \
+      Y_xor_Z = X_xor_Y; \
 		d = SPH_T32(d + t1); \
 		h = SPH_T32(t1 + t2); \
 	} while (0)
@@ -134,7 +138,7 @@ static const sph_u32 K[64] = {
 	SHA2_STEPn(2, a, b, c, d, e, f, g, h, in, pc)

 #define SHA2_ROUND_BODY(in, r)   do { \
-		sph_u32 A, B, C, D, E, F, G, H; \
+		sph_u32 A, B, C, D, E, F, G, H, X_xor_Y, Y_xor_Z; \
 		sph_u32 W[16]; \
 		unsigned pcount; \
 \
@@ -147,6 +151,7 @@ static const sph_u32 K[64] = {
 		G = (r)[6]; \
 		H = (r)[7]; \
 		pcount = 0; \
+      Y_xor_Z = B ^ C; \
 		SHA2_STEP1(A, B, C, D, E, F, G, H, in,  0); \
 		SHA2_STEP1(H, A, B, C, D, E, F, G, in,  1); \
 		SHA2_STEP1(G, H, A, B, C, D, E, F, in,  2); \
@@ -191,10 +196,10 @@ static const sph_u32 K[64] = {
 		(r)[7] = SPH_T32((r)[7] + H); \
 	} while (0)

-#else
+#else  // large footprint (default)

 #define SHA2_ROUND_BODY(in, r)   do { \
-		sph_u32 A, B, C, D, E, F, G, H, T1, T2; \
+		sph_u32 A, B, C, D, E, F, G, H, T1, T2, X_xor_Y, Y_xor_Z;; \
 		sph_u32 W00, W01, W02, W03, W04, W05, W06, W07; \
 		sph_u32 W08, W09, W10, W11, W12, W13, W14, W15; \
 \
@@ -206,388 +211,453 @@ static const sph_u32 K[64] = {
 		F = (r)[5]; \
 		G = (r)[6]; \
 		H = (r)[7]; \
+      Y_xor_Z = B ^ C; \
 		W00 = in(0); \
 		T1 = SPH_T32(H + BSG2_1(E) + CH(E, F, G) \
 			+ SPH_C32(0x428A2F98) + W00); \
 		T2 = SPH_T32(BSG2_0(A) + MAJ(A, B, C)); \
+      Y_xor_Z = X_xor_Y; \
 		D = SPH_T32(D + T1); \
 		H = SPH_T32(T1 + T2); \
 		W01 = in(1); \
 		T1 = SPH_T32(G + BSG2_1(D) + CH(D, E, F) \
 			+ SPH_C32(0x71374491) + W01); \
 		T2 = SPH_T32(BSG2_0(H) + MAJ(H, A, B)); \
+      Y_xor_Z = X_xor_Y; \
 		C = SPH_T32(C + T1); \
 		G = SPH_T32(T1 + T2); \
 		W02 = in(2); \
 		T1 = SPH_T32(F + BSG2_1(C) + CH(C, D, E) \
 			+ SPH_C32(0xB5C0FBCF) + W02); \
 		T2 = SPH_T32(BSG2_0(G) + MAJ(G, H, A)); \
+      Y_xor_Z = X_xor_Y; \
 		B = SPH_T32(B + T1); \
 		F = SPH_T32(T1 + T2); \
 		W03 = in(3); \
 		T1 = SPH_T32(E + BSG2_1(B) + CH(B, C, D) \
 			+ SPH_C32(0xE9B5DBA5) + W03); \
 		T2 = SPH_T32(BSG2_0(F) + MAJ(F, G, H)); \
+      Y_xor_Z = X_xor_Y; \
 		A = SPH_T32(A + T1); \
 		E = SPH_T32(T1 + T2); \
 		W04 = in(4); \
 		T1 = SPH_T32(D + BSG2_1(A) + CH(A, B, C) \
 			+ SPH_C32(0x3956C25B) + W04); \
 		T2 = SPH_T32(BSG2_0(E) + MAJ(E, F, G)); \
+      Y_xor_Z = X_xor_Y; \
 		H = SPH_T32(H + T1); \
 		D = SPH_T32(T1 + T2); \
 		W05 = in(5); \
 		T1 = SPH_T32(C + BSG2_1(H) + CH(H, A, B) \
 			+ SPH_C32(0x59F111F1) + W05); \
 		T2 = SPH_T32(BSG2_0(D) + MAJ(D, E, F)); \
+      Y_xor_Z = X_xor_Y; \
 		G = SPH_T32(G + T1); \
 		C = SPH_T32(T1 + T2); \
 		W06 = in(6); \
 		T1 = SPH_T32(B + BSG2_1(G) + CH(G, H, A) \
 			+ SPH_C32(0x923F82A4) + W06); \
 		T2 = SPH_T32(BSG2_0(C) + MAJ(C, D, E)); \
+      Y_xor_Z = X_xor_Y; \
 		F = SPH_T32(F + T1); \
 		B = SPH_T32(T1 + T2); \
 		W07 = in(7); \
 		T1 = SPH_T32(A + BSG2_1(F) + CH(F, G, H) \
 			+ SPH_C32(0xAB1C5ED5) + W07); \
 		T2 = SPH_T32(BSG2_0(B) + MAJ(B, C, D)); \
+      Y_xor_Z = X_xor_Y; \
 		E = SPH_T32(E + T1); \
 		A = SPH_T32(T1 + T2); \
 		W08 = in(8); \
 		T1 = SPH_T32(H + BSG2_1(E) + CH(E, F, G) \
 			+ SPH_C32(0xD807AA98) + W08); \
 		T2 = SPH_T32(BSG2_0(A) + MAJ(A, B, C)); \
+      Y_xor_Z = X_xor_Y; \
 		D = SPH_T32(D + T1); \
 		H = SPH_T32(T1 + T2); \
 		W09 = in(9); \
 		T1 = SPH_T32(G + BSG2_1(D) + CH(D, E, F) \
 			+ SPH_C32(0x12835B01) + W09); \
 		T2 = SPH_T32(BSG2_0(H) + MAJ(H, A, B)); \
+      Y_xor_Z = X_xor_Y; \
 		C = SPH_T32(C + T1); \
 		G = SPH_T32(T1 + T2); \
 		W10 = in(10); \
 		T1 = SPH_T32(F + BSG2_1(C) + CH(C, D, E) \
 			+ SPH_C32(0x243185BE) + W10); \
 		T2 = SPH_T32(BSG2_0(G) + MAJ(G, H, A)); \
+      Y_xor_Z = X_xor_Y; \
 		B = SPH_T32(B + T1); \
 		F = SPH_T32(T1 + T2); \
 		W11 = in(11); \
 		T1 = SPH_T32(E + BSG2_1(B) + CH(B, C, D) \
 			+ SPH_C32(0x550C7DC3) + W11); \
 		T2 = SPH_T32(BSG2_0(F) + MAJ(F, G, H)); \
+      Y_xor_Z = X_xor_Y; \
 		A = SPH_T32(A + T1); \
 		E = SPH_T32(T1 + T2); \
 		W12 = in(12); \
 		T1 = SPH_T32(D + BSG2_1(A) + CH(A, B, C) \
 			+ SPH_C32(0x72BE5D74) + W12); \
 		T2 = SPH_T32(BSG2_0(E) + MAJ(E, F, G)); \
+      Y_xor_Z = X_xor_Y; \
 		H = SPH_T32(H + T1); \
 		D = SPH_T32(T1 + T2); \
 		W13 = in(13); \
 		T1 = SPH_T32(C + BSG2_1(H) + CH(H, A, B) \
 			+ SPH_C32(0x80DEB1FE) + W13); \
 		T2 = SPH_T32(BSG2_0(D) + MAJ(D, E, F)); \
+      Y_xor_Z = X_xor_Y; \
 		G = SPH_T32(G + T1); \
 		C = SPH_T32(T1 + T2); \
 		W14 = in(14); \
 		T1 = SPH_T32(B + BSG2_1(G) + CH(G, H, A) \
 			+ SPH_C32(0x9BDC06A7) + W14); \
 		T2 = SPH_T32(BSG2_0(C) + MAJ(C, D, E)); \
+      Y_xor_Z = X_xor_Y; \
 		F = SPH_T32(F + T1); \
 		B = SPH_T32(T1 + T2); \
 		W15 = in(15); \
 		T1 = SPH_T32(A + BSG2_1(F) + CH(F, G, H) \
 			+ SPH_C32(0xC19BF174) + W15); \
 		T2 = SPH_T32(BSG2_0(B) + MAJ(B, C, D)); \
+      Y_xor_Z = X_xor_Y; \
 		E = SPH_T32(E + T1); \
 		A = SPH_T32(T1 + T2); \
 		W00 = SPH_T32(SSG2_1(W14) + W09 + SSG2_0(W01) + W00); \
 		T1 = SPH_T32(H + BSG2_1(E) + CH(E, F, G) \
 			+ SPH_C32(0xE49B69C1) + W00); \
 		T2 = SPH_T32(BSG2_0(A) + MAJ(A, B, C)); \
+      Y_xor_Z = X_xor_Y; \
 		D = SPH_T32(D + T1); \
 		H = SPH_T32(T1 + T2); \
 		W01 = SPH_T32(SSG2_1(W15) + W10 + SSG2_0(W02) + W01); \
 		T1 = SPH_T32(G + BSG2_1(D) + CH(D, E, F) \
 			+ SPH_C32(0xEFBE4786) + W01); \
 		T2 = SPH_T32(BSG2_0(H) + MAJ(H, A, B)); \
+      Y_xor_Z = X_xor_Y; \
 		C = SPH_T32(C + T1); \
 		G = SPH_T32(T1 + T2); \
 		W02 = SPH_T32(SSG2_1(W00) + W11 + SSG2_0(W03) + W02); \
 		T1 = SPH_T32(F + BSG2_1(C) + CH(C, D, E) \
 			+ SPH_C32(0x0FC19DC6) + W02); \
 		T2 = SPH_T32(BSG2_0(G) + MAJ(G, H, A)); \
+      Y_xor_Z = X_xor_Y; \
 		B = SPH_T32(B + T1); \
 		F = SPH_T32(T1 + T2); \
 		W03 = SPH_T32(SSG2_1(W01) + W12 + SSG2_0(W04) + W03); \
 		T1 = SPH_T32(E + BSG2_1(B) + CH(B, C, D) \
 			+ SPH_C32(0x240CA1CC) + W03); \
 		T2 = SPH_T32(BSG2_0(F) + MAJ(F, G, H)); \
+      Y_xor_Z = X_xor_Y; \
 		A = SPH_T32(A + T1); \
 		E = SPH_T32(T1 + T2); \
 		W04 = SPH_T32(SSG2_1(W02) + W13 + SSG2_0(W05) + W04); \
 		T1 = SPH_T32(D + BSG2_1(A) + CH(A, B, C) \
 			+ SPH_C32(0x2DE92C6F) + W04); \
 		T2 = SPH_T32(BSG2_0(E) + MAJ(E, F, G)); \
+      Y_xor_Z = X_xor_Y; \
 		H = SPH_T32(H + T1); \
 		D = SPH_T32(T1 + T2); \
 		W05 = SPH_T32(SSG2_1(W03) + W14 + SSG2_0(W06) + W05); \
 		T1 = SPH_T32(C + BSG2_1(H) + CH(H, A, B) \
 			+ SPH_C32(0x4A7484AA) + W05); \
 		T2 = SPH_T32(BSG2_0(D) + MAJ(D, E, F)); \
+      Y_xor_Z = X_xor_Y; \
 		G = SPH_T32(G + T1); \
 		C = SPH_T32(T1 + T2); \
 		W06 = SPH_T32(SSG2_1(W04) + W15 + SSG2_0(W07) + W06); \
 		T1 = SPH_T32(B + BSG2_1(G) + CH(G, H, A) \
 			+ SPH_C32(0x5CB0A9DC) + W06); \
 		T2 = SPH_T32(BSG2_0(C) + MAJ(C, D, E)); \
+      Y_xor_Z = X_xor_Y; \
 		F = SPH_T32(F + T1); \
 		B = SPH_T32(T1 + T2); \
 		W07 = SPH_T32(SSG2_1(W05) + W00 + SSG2_0(W08) + W07); \
 		T1 = SPH_T32(A + BSG2_1(F) + CH(F, G, H) \
 			+ SPH_C32(0x76F988DA) + W07); \
 		T2 = SPH_T32(BSG2_0(B) + MAJ(B, C, D)); \
+      Y_xor_Z = X_xor_Y; \
 		E = SPH_T32(E + T1); \
 		A = SPH_T32(T1 + T2); \
 		W08 = SPH_T32(SSG2_1(W06) + W01 + SSG2_0(W09) + W08); \
 		T1 = SPH_T32(H + BSG2_1(E) + CH(E, F, G) \
 			+ SPH_C32(0x983E5152) + W08); \
 		T2 = SPH_T32(BSG2_0(A) + MAJ(A, B, C)); \
+      Y_xor_Z = X_xor_Y; \
 		D = SPH_T32(D + T1); \
 		H = SPH_T32(T1 + T2); \
 		W09 = SPH_T32(SSG2_1(W07) + W02 + SSG2_0(W10) + W09); \
 		T1 = SPH_T32(G + BSG2_1(D) + CH(D, E, F) \
 			+ SPH_C32(0xA831C66D) + W09); \
 		T2 = SPH_T32(BSG2_0(H) + MAJ(H, A, B)); \
+      Y_xor_Z = X_xor_Y; \
 		C = SPH_T32(C + T1); \
 		G = SPH_T32(T1 + T2); \
 		W10 = SPH_T32(SSG2_1(W08) + W03 + SSG2_0(W11) + W10); \
 		T1 = SPH_T32(F + BSG2_1(C) + CH(C, D, E) \
 			+ SPH_C32(0xB00327C8) + W10); \
 		T2 = SPH_T32(BSG2_0(G) + MAJ(G, H, A)); \
+      Y_xor_Z = X_xor_Y; \
 		B = SPH_T32(B + T1); \
 		F = SPH_T32(T1 + T2); \
 		W11 = SPH_T32(SSG2_1(W09) + W04 + SSG2_0(W12) + W11); \
 		T1 = SPH_T32(E + BSG2_1(B) + CH(B, C, D) \
 			+ SPH_C32(0xBF597FC7) + W11); \
 		T2 = SPH_T32(BSG2_0(F) + MAJ(F, G, H)); \
+      Y_xor_Z = X_xor_Y; \
 		A = SPH_T32(A + T1); \
 		E = SPH_T32(T1 + T2); \
 		W12 = SPH_T32(SSG2_1(W10) + W05 + SSG2_0(W13) + W12); \
 		T1 = SPH_T32(D + BSG2_1(A) + CH(A, B, C) \
 			+ SPH_C32(0xC6E00BF3) + W12); \
 		T2 = SPH_T32(BSG2_0(E) + MAJ(E, F, G)); \
+      Y_xor_Z = X_xor_Y; \
 		H = SPH_T32(H + T1); \
 		D = SPH_T32(T1 + T2); \
 		W13 = SPH_T32(SSG2_1(W11) + W06 + SSG2_0(W14) + W13); \
 		T1 = SPH_T32(C + BSG2_1(H) + CH(H, A, B) \
 			+ SPH_C32(0xD5A79147) + W13); \
 		T2 = SPH_T32(BSG2_0(D) + MAJ(D, E, F)); \
+      Y_xor_Z = X_xor_Y; \
 		G = SPH_T32(G + T1); \
 		C = SPH_T32(T1 + T2); \
 		W14 = SPH_T32(SSG2_1(W12) + W07 + SSG2_0(W15) + W14); \
 		T1 = SPH_T32(B + BSG2_1(G) + CH(G, H, A) \
 			+ SPH_C32(0x06CA6351) + W14); \
 		T2 = SPH_T32(BSG2_0(C) + MAJ(C, D, E)); \
+      Y_xor_Z = X_xor_Y; \
 		F = SPH_T32(F + T1); \
 		B = SPH_T32(T1 + T2); \
 		W15 = SPH_T32(SSG2_1(W13) + W08 + SSG2_0(W00) + W15); \
 		T1 = SPH_T32(A + BSG2_1(F) + CH(F, G, H) \
 			+ SPH_C32(0x14292967) + W15); \
 		T2 = SPH_T32(BSG2_0(B) + MAJ(B, C, D)); \
+      Y_xor_Z = X_xor_Y; \
 		E = SPH_T32(E + T1); \
 		A = SPH_T32(T1 + T2); \
 		W00 = SPH_T32(SSG2_1(W14) + W09 + SSG2_0(W01) + W00); \
 		T1 = SPH_T32(H + BSG2_1(E) + CH(E, F, G) \
 			+ SPH_C32(0x27B70A85) + W00); \
 		T2 = SPH_T32(BSG2_0(A) + MAJ(A, B, C)); \
+      Y_xor_Z = X_xor_Y; \
 		D = SPH_T32(D + T1); \
 		H = SPH_T32(T1 + T2); \
 		W01 = SPH_T32(SSG2_1(W15) + W10 + SSG2_0(W02) + W01); \
 		T1 = SPH_T32(G + BSG2_1(D) + CH(D, E, F) \
 			+ SPH_C32(0x2E1B2138) + W01); \
 		T2 = SPH_T32(BSG2_0(H) + MAJ(H, A, B)); \
+      Y_xor_Z = X_xor_Y; \
 		C = SPH_T32(C + T1); \
 		G = SPH_T32(T1 + T2); \
 		W02 = SPH_T32(SSG2_1(W00) + W11 + SSG2_0(W03) + W02); \
 		T1 = SPH_T32(F + BSG2_1(C) + CH(C, D, E) \
 			+ SPH_C32(0x4D2C6DFC) + W02); \
 		T2 = SPH_T32(BSG2_0(G) + MAJ(G, H, A)); \
+      Y_xor_Z = X_xor_Y; \
 		B = SPH_T32(B + T1); \
 		F = SPH_T32(T1 + T2); \
 		W03 = SPH_T32(SSG2_1(W01) + W12 + SSG2_0(W04) + W03); \
 		T1 = SPH_T32(E + BSG2_1(B) + CH(B, C, D) \
 			+ SPH_C32(0x53380D13) + W03); \
 		T2 = SPH_T32(BSG2_0(F) + MAJ(F, G, H)); \
+      Y_xor_Z = X_xor_Y; \
 		A = SPH_T32(A + T1); \
 		E = SPH_T32(T1 + T2); \
 		W04 = SPH_T32(SSG2_1(W02) + W13 + SSG2_0(W05) + W04); \
 		T1 = SPH_T32(D + BSG2_1(A) + CH(A, B, C) \
 			+ SPH_C32(0x650A7354) + W04); \
 		T2 = SPH_T32(BSG2_0(E) + MAJ(E, F, G)); \
+      Y_xor_Z = X_xor_Y; \
 		H = SPH_T32(H + T1); \
 		D = SPH_T32(T1 + T2); \
 		W05 = SPH_T32(SSG2_1(W03) + W14 + SSG2_0(W06) + W05); \
 		T1 = SPH_T32(C + BSG2_1(H) + CH(H, A, B) \
 			+ SPH_C32(0x766A0ABB) + W05); \
 		T2 = SPH_T32(BSG2_0(D) + MAJ(D, E, F)); \
+      Y_xor_Z = X_xor_Y; \
 		G = SPH_T32(G + T1); \
 		C = SPH_T32(T1 + T2); \
 		W06 = SPH_T32(SSG2_1(W04) + W15 + SSG2_0(W07) + W06); \
 		T1 = SPH_T32(B + BSG2_1(G) + CH(G, H, A) \
 			+ SPH_C32(0x81C2C92E) + W06); \
 		T2 = SPH_T32(BSG2_0(C) + MAJ(C, D, E)); \
+      Y_xor_Z = X_xor_Y; \
 		F = SPH_T32(F + T1); \
 		B = SPH_T32(T1 + T2); \
 		W07 = SPH_T32(SSG2_1(W05) + W00 + SSG2_0(W08) + W07); \
 		T1 = SPH_T32(A + BSG2_1(F) + CH(F, G, H) \
 			+ SPH_C32(0x92722C85) + W07); \
 		T2 = SPH_T32(BSG2_0(B) + MAJ(B, C, D)); \
+      Y_xor_Z = X_xor_Y; \
 		E = SPH_T32(E + T1); \
 		A = SPH_T32(T1 + T2); \
 		W08 = SPH_T32(SSG2_1(W06) + W01 + SSG2_0(W09) + W08); \
 		T1 = SPH_T32(H + BSG2_1(E) + CH(E, F, G) \
 			+ SPH_C32(0xA2BFE8A1) + W08); \
 		T2 = SPH_T32(BSG2_0(A) + MAJ(A, B, C)); \
+      Y_xor_Z = X_xor_Y; \
 		D = SPH_T32(D + T1); \
 		H = SPH_T32(T1 + T2); \
 		W09 = SPH_T32(SSG2_1(W07) + W02 + SSG2_0(W10) + W09); \
 		T1 = SPH_T32(G + BSG2_1(D) + CH(D, E, F) \
 			+ SPH_C32(0xA81A664B) + W09); \
 		T2 = SPH_T32(BSG2_0(H) + MAJ(H, A, B)); \
+      Y_xor_Z = X_xor_Y; \
 		C = SPH_T32(C + T1); \
 		G = SPH_T32(T1 + T2); \
 		W10 = SPH_T32(SSG2_1(W08) + W03 + SSG2_0(W11) + W10); \
 		T1 = SPH_T32(F + BSG2_1(C) + CH(C, D, E) \
 			+ SPH_C32(0xC24B8B70) + W10); \
 		T2 = SPH_T32(BSG2_0(G) + MAJ(G, H, A)); \
+      Y_xor_Z = X_xor_Y; \
 		B = SPH_T32(B + T1); \
 		F = SPH_T32(T1 + T2); \
 		W11 = SPH_T32(SSG2_1(W09) + W04 + SSG2_0(W12) + W11); \
 		T1 = SPH_T32(E + BSG2_1(B) + CH(B, C, D) \
 			+ SPH_C32(0xC76C51A3) + W11); \
 		T2 = SPH_T32(BSG2_0(F) + MAJ(F, G, H)); \
+      Y_xor_Z = X_xor_Y; \
 		A = SPH_T32(A + T1); \
 		E = SPH_T32(T1 + T2); \
 		W12 = SPH_T32(SSG2_1(W10) + W05 + SSG2_0(W13) + W12); \
 		T1 = SPH_T32(D + BSG2_1(A) + CH(A, B, C) \
 			+ SPH_C32(0xD192E819) + W12); \
 		T2 = SPH_T32(BSG2_0(E) + MAJ(E, F, G)); \
+      Y_xor_Z = X_xor_Y; \
 		H = SPH_T32(H + T1); \
 		D = SPH_T32(T1 + T2); \
 		W13 = SPH_T32(SSG2_1(W11) + W06 + SSG2_0(W14) + W13); \
 		T1 = SPH_T32(C + BSG2_1(H) + CH(H, A, B) \
 			+ SPH_C32(0xD6990624) + W13); \
 		T2 = SPH_T32(BSG2_0(D) + MAJ(D, E, F)); \
+      Y_xor_Z = X_xor_Y; \
 		G = SPH_T32(G + T1); \
 		C = SPH_T32(T1 + T2); \
 		W14 = SPH_T32(SSG2_1(W12) + W07 + SSG2_0(W15) + W14); \
 		T1 = SPH_T32(B + BSG2_1(G) + CH(G, H, A) \
 			+ SPH_C32(0xF40E3585) + W14); \
 		T2 = SPH_T32(BSG2_0(C) + MAJ(C, D, E)); \
+      Y_xor_Z = X_xor_Y; \
 		F = SPH_T32(F + T1); \
 		B = SPH_T32(T1 + T2); \
 		W15 = SPH_T32(SSG2_1(W13) + W08 + SSG2_0(W00) + W15); \
 		T1 = SPH_T32(A + BSG2_1(F) + CH(F, G, H) \
 			+ SPH_C32(0x106AA070) + W15); \
 		T2 = SPH_T32(BSG2_0(B) + MAJ(B, C, D)); \
+      Y_xor_Z = X_xor_Y; \
 		E = SPH_T32(E + T1); \
 		A = SPH_T32(T1 + T2); \
 		W00 = SPH_T32(SSG2_1(W14) + W09 + SSG2_0(W01) + W00); \
 		T1 = SPH_T32(H + BSG2_1(E) + CH(E, F, G) \
 			+ SPH_C32(0x19A4C116) + W00); \
 		T2 = SPH_T32(BSG2_0(A) + MAJ(A, B, C)); \
+      Y_xor_Z = X_xor_Y; \
 		D = SPH_T32(D + T1); \
 		H = SPH_T32(T1 + T2); \
 		W01 = SPH_T32(SSG2_1(W15) + W10 + SSG2_0(W02) + W01); \
 		T1 = SPH_T32(G + BSG2_1(D) + CH(D, E, F) \
 			+ SPH_C32(0x1E376C08) + W01); \
 		T2 = SPH_T32(BSG2_0(H) + MAJ(H, A, B)); \
+      Y_xor_Z = X_xor_Y; \
 		C = SPH_T32(C + T1); \
 		G = SPH_T32(T1 + T2); \
 		W02 = SPH_T32(SSG2_1(W00) + W11 + SSG2_0(W03) + W02); \
 		T1 = SPH_T32(F + BSG2_1(C) + CH(C, D, E) \
 			+ SPH_C32(0x2748774C) + W02); \
 		T2 = SPH_T32(BSG2_0(G) + MAJ(G, H, A)); \
+      Y_xor_Z = X_xor_Y; \
 		B = SPH_T32(B + T1); \
 		F = SPH_T32(T1 + T2); \
 		W03 = SPH_T32(SSG2_1(W01) + W12 + SSG2_0(W04) + W03); \
 		T1 = SPH_T32(E + BSG2_1(B) + CH(B, C, D) \
 			+ SPH_C32(0x34B0BCB5) + W03); \
 		T2 = SPH_T32(BSG2_0(F) + MAJ(F, G, H)); \
+      Y_xor_Z = X_xor_Y; \
 		A = SPH_T32(A + T1); \
 		E = SPH_T32(T1 + T2); \
 		W04 = SPH_T32(SSG2_1(W02) + W13 + SSG2_0(W05) + W04); \
 		T1 = SPH_T32(D + BSG2_1(A) + CH(A, B, C) \
 			+ SPH_C32(0x391C0CB3) + W04); \
 		T2 = SPH_T32(BSG2_0(E) + MAJ(E, F, G)); \
+      Y_xor_Z = X_xor_Y; \
 		H = SPH_T32(H + T1); \
 		D = SPH_T32(T1 + T2); \
 		W05 = SPH_T32(SSG2_1(W03) + W14 + SSG2_0(W06) + W05); \
 		T1 = SPH_T32(C + BSG2_1(H) + CH(H, A, B) \
 			+ SPH_C32(0x4ED8AA4A) + W05); \
 		T2 = SPH_T32(BSG2_0(D) + MAJ(D, E, F)); \
+      Y_xor_Z = X_xor_Y; \
 		G = SPH_T32(G + T1); \
 		C = SPH_T32(T1 + T2); \
 		W06 = SPH_T32(SSG2_1(W04) + W15 + SSG2_0(W07) + W06); \
 		T1 = SPH_T32(B + BSG2_1(G) + CH(G, H, A) \
 			+ SPH_C32(0x5B9CCA4F) + W06); \
 		T2 = SPH_T32(BSG2_0(C) + MAJ(C, D, E)); \
+      Y_xor_Z = X_xor_Y; \
 		F = SPH_T32(F + T1); \
 		B = SPH_T32(T1 + T2); \
 		W07 = SPH_T32(SSG2_1(W05) + W00 + SSG2_0(W08) + W07); \
 		T1 = SPH_T32(A + BSG2_1(F) + CH(F, G, H) \
 			+ SPH_C32(0x682E6FF3) + W07); \
 		T2 = SPH_T32(BSG2_0(B) + MAJ(B, C, D)); \
+      Y_xor_Z = X_xor_Y; \
 		E = SPH_T32(E + T1); \
 		A = SPH_T32(T1 + T2); \
 		W08 = SPH_T32(SSG2_1(W06) + W01 + SSG2_0(W09) + W08); \
 		T1 = SPH_T32(H + BSG2_1(E) + CH(E, F, G) \
 			+ SPH_C32(0x748F82EE) + W08); \
 		T2 = SPH_T32(BSG2_0(A) + MAJ(A, B, C)); \
+      Y_xor_Z = X_xor_Y; \
 		D = SPH_T32(D + T1); \
 		H = SPH_T32(T1 + T2); \
 		W09 = SPH_T32(SSG2_1(W07) + W02 + SSG2_0(W10) + W09); \
 		T1 = SPH_T32(G + BSG2_1(D) + CH(D, E, F) \
 			+ SPH_C32(0x78A5636F) + W09); \
 		T2 = SPH_T32(BSG2_0(H) + MAJ(H, A, B)); \
+      Y_xor_Z = X_xor_Y; \
 		C = SPH_T32(C + T1); \
 		G = SPH_T32(T1 + T2); \
 		W10 = SPH_T32(SSG2_1(W08) + W03 + SSG2_0(W11) + W10); \
 		T1 = SPH_T32(F + BSG2_1(C) + CH(C, D, E) \
 			+ SPH_C32(0x84C87814) + W10); \
 		T2 = SPH_T32(BSG2_0(G) + MAJ(G, H, A)); \
+      Y_xor_Z = X_xor_Y; \
 		B = SPH_T32(B + T1); \
 		F = SPH_T32(T1 + T2); \
 		W11 = SPH_T32(SSG2_1(W09) + W04 + SSG2_0(W12) + W11); \
 		T1 = SPH_T32(E + BSG2_1(B) + CH(B, C, D) \
 			+ SPH_C32(0x8CC70208) + W11); \
 		T2 = SPH_T32(BSG2_0(F) + MAJ(F, G, H)); \
+      Y_xor_Z = X_xor_Y; \
 		A = SPH_T32(A + T1); \
 		E = SPH_T32(T1 + T2); \
 		W12 = SPH_T32(SSG2_1(W10) + W05 + SSG2_0(W13) + W12); \
 		T1 = SPH_T32(D + BSG2_1(A) + CH(A, B, C) \
 			+ SPH_C32(0x90BEFFFA) + W12); \
 		T2 = SPH_T32(BSG2_0(E) + MAJ(E, F, G)); \
+      Y_xor_Z = X_xor_Y; \
 		H = SPH_T32(H + T1); \
 		D = SPH_T32(T1 + T2); \
 		W13 = SPH_T32(SSG2_1(W11) + W06 + SSG2_0(W14) + W13); \
 		T1 = SPH_T32(C + BSG2_1(H) + CH(H, A, B) \
 			+ SPH_C32(0xA4506CEB) + W13); \
 		T2 = SPH_T32(BSG2_0(D) + MAJ(D, E, F)); \
+      Y_xor_Z = X_xor_Y; \
 		G = SPH_T32(G + T1); \
 		C = SPH_T32(T1 + T2); \
 		W14 = SPH_T32(SSG2_1(W12) + W07 + SSG2_0(W15) + W14); \
 		T1 = SPH_T32(B + BSG2_1(G) + CH(G, H, A) \
 			+ SPH_C32(0xBEF9A3F7) + W14); \
 		T2 = SPH_T32(BSG2_0(C) + MAJ(C, D, E)); \
+      Y_xor_Z = X_xor_Y; \
 		F = SPH_T32(F + T1); \
 		B = SPH_T32(T1 + T2); \
 		W15 = SPH_T32(SSG2_1(W13) + W08 + SSG2_0(W00) + W15); \
 		T1 = SPH_T32(A + BSG2_1(F) + CH(F, G, H) \
 			+ SPH_C32(0xC67178F2) + W15); \
 		T2 = SPH_T32(BSG2_0(B) + MAJ(B, C, D)); \
+      Y_xor_Z = X_xor_Y; \
 		E = SPH_T32(E + T1); \
 		A = SPH_T32(T1 + T2); \
 		(r)[0] = SPH_T32((r)[0] + A); \
@@ -600,7 +670,7 @@ static const sph_u32 K[64] = {
 		(r)[7] = SPH_T32((r)[7] + H); \
 	} while (0)

-#endif
+#endif  // small footprint else

 /*
 * One round of SHA-224 / SHA-256. The data must be aligned for 32-bit access.
@@ -613,6 +683,25 @@ sha2_round(const unsigned char *data, sph_u32 r[8])
 #undef SHA2_IN
 }

+void sph_sha256_transform_le( uint32_t *state_out, const uint32_t *data,
+                              const uint32_t *state_in )
+{
+memcpy( state_out, state_in, 32 );
+#define SHA2_IN(x)   (data[x])
+   SHA2_ROUND_BODY( SHA2_IN, state_out );
+#undef SHA2_IN
+}
+
+void sph_sha256_transform_be( uint32_t *state_out, const uint32_t *data, 
+                              const uint32_t *state_in )
+{  
+memcpy( state_out, state_in, 32 );
+#define SHA2_IN(x)   sph_dec32be_aligned( data+(x) )
+   SHA2_ROUND_BODY( SHA2_IN, state_out );
+#undef SHA2_IN
+
+}
+
 /* see sph_sha2.h */
 void
 sph_sha224_init(void *cc)
@@ -653,7 +742,7 @@ void
 sph_sha224_close(void *cc, void *dst)
 {
 	sha224_close(cc, dst, 7);
-	sph_sha224_init(cc);
+//	sph_sha224_init(cc);
 }

 /* see sph_sha2.h */
@@ -661,7 +750,7 @@ void
 sph_sha224_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
 {
 	sha224_addbits_and_close(cc, ub, n, dst, 7);
-	sph_sha224_init(cc);
+//	sph_sha224_init(cc);
 }

 /* see sph_sha2.h */
@@ -677,14 +766,22 @@ void
 sph_sha256_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
 {
 	sha224_addbits_and_close(cc, ub, n, dst, 8);
-	sph_sha256_init(cc);
+//	sph_sha256_init(cc);
 }

-/* see sph_sha2.h */
-void
-sph_sha224_comp(const sph_u32 msg[16], sph_u32 val[8])
+void sph_sha256_full( void *dst, const void *data, size_t len )
 {
-#define SHA2_IN(x)   msg[x]
-	SHA2_ROUND_BODY(SHA2_IN, val);
-#undef SHA2_IN
-}
+   sph_sha256_context cc;
+   sph_sha256_init( &cc );
+   sph_sha256( &cc, data, len );
+   sph_sha256_close( &cc, dst );
+}   
+
+/* see sph_sha2.h */
+//void
+//sph_sha224_comp(const sph_u32 msg[16], sph_u32 val[8])
+//{
+//#define SHA2_IN(x)   msg[x]
+//	SHA2_ROUND_BODY(SHA2_IN, val);
+//#undef SHA2_IN
+//}
--- a/algo/sha/sph_sha2.h
+++ b/algo/sha/sph_sha2.h
@@ -73,7 +73,7 @@ typedef struct {
 	sph_u32 count_high, count_low;
 #endif
 #endif
-} sph_sha224_context;
+} sph_sha224_context __attribute__((aligned(64)));

 /**
 * This structure is a context for SHA-256 computations. It is identical
@@ -205,6 +205,17 @@ void sph_sha256_comp(const sph_u32 msg[16], sph_u32 val[8]);
 #define sph_sha256_comp   sph_sha224_comp
 #endif

+void sph_sha256_full( void *dst, const void *data, size_t len );
+
+// These shouldn't be called directly, use sha256-hash.h generic functions
+// sha256_transform_le & sha256_transform_be instead.
+void sph_sha256_transform_le( uint32_t *state_out, const uint32_t *data,
+                              const uint32_t *state_in );
+
+void sph_sha256_transform_be( uint32_t *state_out, const uint32_t *data,
+                              const uint32_t *state_in );
+
+
 #if SPH_64

 /**
--- a/algo/sha/sph_sha2big.c
+++ b/algo/sha/sph_sha2big.c
@@ -38,7 +38,8 @@
 #if SPH_64

 #define CH(X, Y, Z)    ((((Y) ^ (Z)) & (X)) ^ (Z))
-#define MAJ(X, Y, Z)   (((X) & (Y)) | (((X) | (Y)) & (Z)))
+//#define MAJ(X, Y, Z)   (((X) & (Y)) | (((X) | (Y)) & (Z)))
+#define MAJ( X, Y, Z )   ( Y  ^ ( ( X ^ Y ) & ( Y ^ Z ) ) )

 #define ROTR64    SPH_ROTR64

--- a/algo/shabal/shabal-hash-4way.c
+++ b/algo/shabal/shabal-hash-4way.c
@@ -70,6 +70,8 @@ extern "C"{
           C8, C9, CA, CB, CC, CD, CE, CF; \
   __m256i M0, M1, M2, M3, M4, M5, M6, M7, \
           M8, M9, MA, MB, MC, MD, ME, MF; \
+   const __m256i FIVE  = _mm256_set1_epi32( 5 ); \
+   const __m256i THREE = _mm256_set1_epi32( 3 ); \
   sph_u32 Wlow, Whigh;

 #define READ_STATE8(state) do \
@@ -310,12 +312,12 @@ do { \

 #define PERM_ELT8(xa0, xa1, xb0, xb1, xb2, xb3, xc, xm) \
 do { \
-   xa0 = _mm256_xor_si256( xm, _mm256_xor_si256( xb1, _mm256_xor_si256(  \
+   xa0 = mm256_xor3( xm, xb1, _mm256_xor_si256(  \
            _mm256_andnot_si256( xb3, xb2 ), \
-            _mm256_mullo_epi32( _mm256_xor_si256( xa0, _mm256_xor_si256( xc, \
-               _mm256_mullo_epi32(  mm256_rol_32( xa1, 15 ), _mm256_set1_epi32(5UL) ) \
-                   ) ), _mm256_set1_epi32(3UL) ) ) ) ); \
-   xb0 = mm256_not( _mm256_xor_si256( xa0, mm256_rol_32( xb0, 1 ) ) ); \
+            _mm256_mullo_epi32( mm256_xor3( xa0, xc, \
+               _mm256_mullo_epi32( mm256_rol_32( xa1, 15 ), \
+                                   FIVE ) ), THREE ) ) ); \
+   xb0 = mm256_xnor( xa0, mm256_rol_32( xb0, 1 ) ); \
 } while (0)

 #define PERM_STEP_0_8   do { \
@@ -666,7 +668,9 @@ shabal512_8way_addbits_and_close(void *cc, unsigned ub, unsigned n, void *dst)
 	        C8, C9, CA, CB, CC, CD, CE, CF; \
 	__m128i M0, M1, M2, M3, M4, M5, M6, M7, \
 	        M8, M9, MA, MB, MC, MD, ME, MF; \
-	sph_u32 Wlow, Whigh;
+   const __m128i FIVE  = _mm_set1_epi32( 5 ); \
+   const __m128i THREE = _mm_set1_epi32( 3 ); \
+   sph_u32 Wlow, Whigh;

 #define READ_STATE(state) do \
 { \
@@ -930,8 +934,8 @@ do { \
   xa0 = _mm_xor_si128( xm, _mm_xor_si128( xb1, _mm_xor_si128(  \
            _mm_andnot_si128( xb3, xb2 ), \
            _mm_mullo_epi32( _mm_xor_si128( xa0, _mm_xor_si128( xc, \
-               _mm_mullo_epi32(  mm128_rol_32( xa1, 15 ), _mm_set1_epi32(5UL) ) \
-                   ) ), _mm_set1_epi32(3UL) ) ) ) ); \
+               _mm_mullo_epi32(  mm128_rol_32( xa1, 15 ), FIVE ) \
+                   ) ), THREE ) ) ) ); \
   xb0 = mm128_not( _mm_xor_si128( xa0, mm128_rol_32( xb0, 1 ) ) ); \
 } while (0)

--- a/algo/shavite/shavite-hash-2way.c
+++ b/algo/shavite/shavite-hash-2way.c
@@ -20,8 +20,21 @@ static const uint32_t IV512[] =


 #define mm256_ror2x256hi_1x32( a, b ) \
-   _mm256_blend_epi32( mm256_ror128_32( a ), \
-                       mm256_ror128_32( b ), 0x88 )
+   _mm256_blend_epi32( mm256_shuflr128_32( a ), \
+                       mm256_shuflr128_32( b ), 0x88 )
+
+#if defined(__VAES__)
+
+#define mm256_aesenc_2x128( x, k ) \
+   _mm256_aesenc_epi128( x, _mm256_castsi128_si256( k ) )
+
+#else
+
+#define mm256_aesenc_2x128( x, k ) \
+   mm256_concat_128( _mm_aesenc_si128( mm128_extr_hi128_256( x ), k ), \
+                     _mm_aesenc_si128( mm128_extr_lo128_256( x ), k ) )
+
+#endif

 static void
 c512_2way( shavite512_2way_context *ctx, const void *msg )
@@ -65,7 +78,7 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )
   {
      // round 1, 5, 9

-     k00 = _mm256_xor_si256( k13, mm256_ror128_32(
+     k00 = _mm256_xor_si256( k13, mm256_shuflr128_32(
                                  mm256_aesenc_2x128( k00, zero ) ) );

     if ( r == 0 )
@@ -75,7 +88,7 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )

     x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k00 ), zero );
     k01 = _mm256_xor_si256( k00,
-		     mm256_ror128_32( mm256_aesenc_2x128( k01, zero ) ) );
+		     mm256_shuflr128_32( mm256_aesenc_2x128( k01, zero ) ) );

     if ( r == 1 )
        k01 = _mm256_xor_si256( k01, _mm256_set_epi32(
@@ -84,25 +97,25 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )

     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
     k02 = _mm256_xor_si256( k01,
-		     mm256_ror128_32( mm256_aesenc_2x128( k02, zero ) ) );
+		     mm256_shuflr128_32( mm256_aesenc_2x128( k02, zero ) ) );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
     k03 = _mm256_xor_si256( k02,
-		     mm256_ror128_32( mm256_aesenc_2x128( k03, zero ) ) );
+		     mm256_shuflr128_32( mm256_aesenc_2x128( k03, zero ) ) );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );

     p3 = _mm256_xor_si256( p3, x );

     k10 = _mm256_xor_si256( k03,
-		     mm256_ror128_32( mm256_aesenc_2x128( k10, zero ) ) );
+		     mm256_shuflr128_32( mm256_aesenc_2x128( k10, zero ) ) );
     x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k10 ), zero );
     k11 = _mm256_xor_si256( k10,
-		     mm256_ror128_32( mm256_aesenc_2x128( k11, zero ) ) );
+		     mm256_shuflr128_32( mm256_aesenc_2x128( k11, zero ) ) );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
     k12 = _mm256_xor_si256( k11,
-		     mm256_ror128_32( mm256_aesenc_2x128( k12, zero ) ) );
+		     mm256_shuflr128_32( mm256_aesenc_2x128( k12, zero ) ) );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
     k13 = _mm256_xor_si256( k12,
-		     mm256_ror128_32( mm256_aesenc_2x128( k13, zero ) ) );
+		     mm256_shuflr128_32( mm256_aesenc_2x128( k13, zero ) ) );

     if ( r == 2 )
        k13 = _mm256_xor_si256( k13, _mm256_set_epi32(
@@ -138,31 +151,31 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )

     // round 3, 7, 11

-     k00 = _mm256_xor_si256( mm256_ror128_32(
+     k00 = _mm256_xor_si256( mm256_shuflr128_32(
                                     mm256_aesenc_2x128( k00, zero ) ), k13 );
     x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k00 ), zero );
-     k01 = _mm256_xor_si256( mm256_ror128_32(
+     k01 = _mm256_xor_si256( mm256_shuflr128_32(
                                     mm256_aesenc_2x128( k01, zero ) ), k00 );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
-     k02 = _mm256_xor_si256( mm256_ror128_32(
+     k02 = _mm256_xor_si256( mm256_shuflr128_32(
                                     mm256_aesenc_2x128( k02, zero ) ), k01 );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
-     k03 = _mm256_xor_si256( mm256_ror128_32(
+     k03 = _mm256_xor_si256( mm256_shuflr128_32(
                                     mm256_aesenc_2x128( k03, zero ) ), k02 );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );

     p1 = _mm256_xor_si256( p1, x );

-     k10 = _mm256_xor_si256( mm256_ror128_32(
+     k10 = _mm256_xor_si256( mm256_shuflr128_32(
                                     mm256_aesenc_2x128( k10, zero ) ), k03 );
     x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k10 ), zero );
-     k11 = _mm256_xor_si256( mm256_ror128_32(
+     k11 = _mm256_xor_si256( mm256_shuflr128_32(
                                     mm256_aesenc_2x128( k11, zero ) ), k10 );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );
-     k12 = _mm256_xor_si256( mm256_ror128_32(
+     k12 = _mm256_xor_si256( mm256_shuflr128_32(
                                     mm256_aesenc_2x128( k12, zero ) ), k11 );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
-     k13 = _mm256_xor_si256( mm256_ror128_32(
+     k13 = _mm256_xor_si256( mm256_shuflr128_32(
                                     mm256_aesenc_2x128( k13, zero ) ), k12 );
     x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );

@@ -196,35 +209,35 @@ c512_2way( shavite512_2way_context *ctx, const void *msg )

   // round 13

-   k00 = _mm256_xor_si256( mm256_ror128_32(
+   k00 = _mm256_xor_si256( mm256_shuflr128_32(
 			             mm256_aesenc_2x128( k00, zero ) ), k13  );
   x = mm256_aesenc_2x128( _mm256_xor_si256( p0, k00 ), zero );
-   k01 = _mm256_xor_si256( mm256_ror128_32(
+   k01 = _mm256_xor_si256( mm256_shuflr128_32(
 			             mm256_aesenc_2x128( k01, zero ) ), k00 );
   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k01 ), zero );
-   k02 = _mm256_xor_si256( mm256_ror128_32(
+   k02 = _mm256_xor_si256( mm256_shuflr128_32(
 			             mm256_aesenc_2x128( k02, zero ) ), k01 );
   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k02 ), zero );
-   k03 = _mm256_xor_si256( mm256_ror128_32(
+   k03 = _mm256_xor_si256( mm256_shuflr128_32(
 			             mm256_aesenc_2x128( k03, zero ) ), k02 );
   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k03 ), zero );

   p3 = _mm256_xor_si256( p3, x );

-   k10 = _mm256_xor_si256( mm256_ror128_32(
+   k10 = _mm256_xor_si256( mm256_shuflr128_32(
 			             mm256_aesenc_2x128( k10, zero ) ), k03 );
   x = mm256_aesenc_2x128( _mm256_xor_si256( p2, k10 ), zero );
-   k11 = _mm256_xor_si256( mm256_ror128_32(
+   k11 = _mm256_xor_si256( mm256_shuflr128_32(
 			             mm256_aesenc_2x128( k11, zero ) ), k10 );
   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k11 ), zero );

-   k12 = mm256_ror128_32( mm256_aesenc_2x128( k12, zero ) );
+   k12 = mm256_shuflr128_32( mm256_aesenc_2x128( k12, zero ) );
   k12 = _mm256_xor_si256( k12, _mm256_xor_si256( k11, _mm256_set_epi32(
 	       ~ctx->count2, ctx->count3, ctx->count0, ctx->count1,
 	       ~ctx->count2, ctx->count3, ctx->count0, ctx->count1 ) ) );

   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k12 ), zero );
-   k13 = _mm256_xor_si256( mm256_ror128_32(
+   k13 = _mm256_xor_si256( mm256_shuflr128_32(
 			             mm256_aesenc_2x128( k13, zero ) ), k12 );
   x = mm256_aesenc_2x128( _mm256_xor_si256( x, k13 ), zero );

@@ -304,7 +317,7 @@ void shavite512_2way_close( shavite512_2way_context *ctx, void *dst )
    uint32_t vp = ctx->ptr>>5;

    // Terminating byte then zero pad
-    casti_m256i( buf, vp++ ) = m256_const2_64( 0, 0x0000000000000080 );
+    casti_m256i( buf, vp++ ) = m256_const1_i128( 0x0000000000000080 );

    // Zero pad full vectors up to count
    for ( ; vp < 6; vp++ )      
@@ -384,13 +397,13 @@ void shavite512_2way_update_close( shavite512_2way_context *ctx, void *dst,

   if ( vp == 0 )    // empty buf, xevan.
   { 
-      casti_m256i( buf, 0 ) = m256_const2_64( 0, 0x0000000000000080 );
+      casti_m256i( buf, 0 ) = m256_const1_i128( 0x0000000000000080 );
      memset_zero_256( (__m256i*)buf + 1, 5 );
      ctx->count0 = ctx->count1 = ctx->count2 = ctx->count3 = 0;
   }
   else     // half full buf, everyone else.
   {
-    casti_m256i( buf, vp++ ) = m256_const2_64( 0, 0x0000000000000080 );
+    casti_m256i( buf, vp++ ) = m256_const1_i128( 0x0000000000000080 );
      memset_zero_256( (__m256i*)buf + vp, 6 - vp );
   }

@@ -474,13 +487,13 @@ void shavite512_2way_full( shavite512_2way_context *ctx, void *dst,

   if ( vp == 0 )    // empty buf, xevan.
   {
-      casti_m256i( buf, 0 ) = m256_const2_64( 0, 0x0000000000000080 );
+      casti_m256i( buf, 0 ) = m256_const1_i128( 0x0000000000000080 );
      memset_zero_256( (__m256i*)buf + 1, 5 );
      ctx->count0 = ctx->count1 = ctx->count2 = ctx->count3 = 0;
   }
   else     // half full buf, everyone else.
   {
-    casti_m256i( buf, vp++ ) = m256_const2_64( 0, 0x0000000000000080 );
+    casti_m256i( buf, vp++ ) = m256_const1_i128( 0x0000000000000080 );
      memset_zero_256( (__m256i*)buf + vp, 6 - vp );
   }

--- a/algo/shavite/shavite-hash-4way.c
+++ b/algo/shavite/shavite-hash-4way.c
@@ -12,8 +12,8 @@ static const uint32_t IV512[] =
 };

 #define mm512_ror2x512hi_1x32( a, b ) \
-   _mm512_mask_blend_epi32( 0x8888, mm512_ror128_32( a ), \
-                                    mm512_ror128_32( b ) )
+   _mm512_mask_blend_epi32( 0x8888, mm512_shuflr128_32( a ), \
+                                    mm512_shuflr128_32( b ) )

 static void
 c512_4way( shavite512_4way_context *ctx, const void *msg )
@@ -23,6 +23,8 @@ c512_4way( shavite512_4way_context *ctx, const void *msg )
   register __m512i K0, K1, K2, K3, K4, K5, K6, K7;
   __m512i *M = (__m512i*)msg;
   __m512i *H = (__m512i*)ctx->h;
+   const __m512i count = _mm512_set4_epi32( ctx->count3, ctx->count2,
+                                            ctx->count1, ctx->count0 );
   int r;

   P0 = H[0];
@@ -58,46 +60,46 @@ c512_4way( shavite512_4way_context *ctx, const void *msg )
   {
      // round 1, 5, 9

-     K0 = _mm512_xor_si512( K7, mm512_ror128_32(
+     K0 = _mm512_xor_si512( K7, mm512_shuflr128_32(
                                  _mm512_aesenc_epi128( K0, m512_zero ) ) );

     if ( r == 0 )
-        K0 = _mm512_xor_si512( K0, _mm512_set4_epi32( 
-		              ~ctx->count3, ctx->count2, ctx->count1, ctx->count0 ) );
+        K0 = _mm512_xor_si512( K0,
+                    _mm512_mask_xor_epi32( count, 0x8888, count, m512_neg1 ) );

     X = _mm512_aesenc_epi128( _mm512_xor_si512( P0, K0 ), m512_zero );
     K1 = _mm512_xor_si512( K0,
-		           mm512_ror128_32( _mm512_aesenc_epi128( K1, m512_zero ) ) );
+		           mm512_shuflr128_32( _mm512_aesenc_epi128( K1, m512_zero ) ) );

     if ( r == 1 )
-        K1 = _mm512_xor_si512( K1, _mm512_set4_epi32(
-	                 ~ctx->count0, ctx->count1, ctx->count2, ctx->count3 ) ); 
+        K1 = _mm512_xor_si512( K1, mm512_shuflr128_32(
+                 _mm512_mask_xor_epi32( count, 0x1111, count, m512_neg1 ) ) );

     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K1 ), m512_zero );
     K2 = _mm512_xor_si512( K1,
-		           mm512_ror128_32( _mm512_aesenc_epi128( K2, m512_zero ) ) );
+		           mm512_shuflr128_32( _mm512_aesenc_epi128( K2, m512_zero ) ) );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K2 ), m512_zero );
     K3 = _mm512_xor_si512( K2,
-		           mm512_ror128_32( _mm512_aesenc_epi128( K3, m512_zero ) ) );
+		           mm512_shuflr128_32( _mm512_aesenc_epi128( K3, m512_zero ) ) );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K3 ), m512_zero );

     P3 = _mm512_xor_si512( P3, X );

     K4 = _mm512_xor_si512( K3,
-		           mm512_ror128_32( _mm512_aesenc_epi128( K4, m512_zero ) ) );
+		           mm512_shuflr128_32( _mm512_aesenc_epi128( K4, m512_zero ) ) );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( P2, K4 ), m512_zero );
     K5 = _mm512_xor_si512( K4,
-		           mm512_ror128_32( _mm512_aesenc_epi128( K5, m512_zero ) ) );
+		           mm512_shuflr128_32( _mm512_aesenc_epi128( K5, m512_zero ) ) );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K5 ), m512_zero );
     K6 = _mm512_xor_si512( K5,
-		           mm512_ror128_32( _mm512_aesenc_epi128( K6, m512_zero ) ) );
+		           mm512_shuflr128_32( _mm512_aesenc_epi128( K6, m512_zero ) ) );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K6 ), m512_zero );
     K7 = _mm512_xor_si512( K6,
-		           mm512_ror128_32( _mm512_aesenc_epi128( K7, m512_zero ) ) );
+		           mm512_shuflr128_32( _mm512_aesenc_epi128( K7, m512_zero ) ) );

     if ( r == 2 )
-        K7 = _mm512_xor_si512( K7, _mm512_set4_epi32(
-                    ~ctx->count1, ctx->count0, ctx->count3, ctx->count2 ) );
+        K7 = _mm512_xor_si512( K7, mm512_swap128_64(
+                 _mm512_mask_xor_epi32( count, 0x2222, count, m512_neg1 ) ) );
 
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K7 ), m512_zero );
     P1 = _mm512_xor_si512( P1, X );
@@ -128,31 +130,31 @@ c512_4way( shavite512_4way_context *ctx, const void *msg )

     // round 3, 7, 11

-     K0 = _mm512_xor_si512( mm512_ror128_32(
+     K0 = _mm512_xor_si512( mm512_shuflr128_32(
                               _mm512_aesenc_epi128( K0, m512_zero ) ), K7 );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( P2, K0 ), m512_zero );
-     K1 = _mm512_xor_si512( mm512_ror128_32(
+     K1 = _mm512_xor_si512( mm512_shuflr128_32(
                               _mm512_aesenc_epi128( K1, m512_zero ) ), K0 );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K1 ), m512_zero );
-     K2 = _mm512_xor_si512( mm512_ror128_32(
+     K2 = _mm512_xor_si512( mm512_shuflr128_32(
                               _mm512_aesenc_epi128( K2, m512_zero ) ), K1 );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K2 ), m512_zero );
-     K3 = _mm512_xor_si512( mm512_ror128_32(
+     K3 = _mm512_xor_si512( mm512_shuflr128_32(
                               _mm512_aesenc_epi128( K3, m512_zero ) ), K2 );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K3 ), m512_zero );

     P1 = _mm512_xor_si512( P1, X );

-     K4 = _mm512_xor_si512( mm512_ror128_32(
+     K4 = _mm512_xor_si512( mm512_shuflr128_32(
                               _mm512_aesenc_epi128( K4, m512_zero ) ), K3 );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( P0, K4 ), m512_zero );
-     K5 = _mm512_xor_si512( mm512_ror128_32(
+     K5 = _mm512_xor_si512( mm512_shuflr128_32(
                               _mm512_aesenc_epi128( K5, m512_zero ) ), K4 );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K5 ), m512_zero );
-     K6 = _mm512_xor_si512( mm512_ror128_32(
+     K6 = _mm512_xor_si512( mm512_shuflr128_32(
                               _mm512_aesenc_epi128( K6, m512_zero ) ), K5 );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K6 ), m512_zero );
-     K7 = _mm512_xor_si512( mm512_ror128_32(
+     K7 = _mm512_xor_si512( mm512_shuflr128_32(
                               _mm512_aesenc_epi128( K7, m512_zero ) ), K6 );
     X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K7 ), m512_zero );

@@ -185,34 +187,34 @@ c512_4way( shavite512_4way_context *ctx, const void *msg )

   // round 13

-   K0 = _mm512_xor_si512( mm512_ror128_32(
+   K0 = _mm512_xor_si512( mm512_shuflr128_32(
 			             _mm512_aesenc_epi128( K0, m512_zero ) ), K7  );
   X = _mm512_aesenc_epi128( _mm512_xor_si512( P0, K0 ), m512_zero );
-   K1 = _mm512_xor_si512( mm512_ror128_32(
+   K1 = _mm512_xor_si512( mm512_shuflr128_32(
 			             _mm512_aesenc_epi128( K1, m512_zero ) ), K0 );
   X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K1 ), m512_zero );
-   K2 = _mm512_xor_si512( mm512_ror128_32(
+   K2 = _mm512_xor_si512( mm512_shuflr128_32(
 			             _mm512_aesenc_epi128( K2, m512_zero ) ), K1 );
   X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K2 ), m512_zero );
-   K3 = _mm512_xor_si512( mm512_ror128_32(
+   K3 = _mm512_xor_si512( mm512_shuflr128_32(
 			             _mm512_aesenc_epi128( K3, m512_zero ) ), K2 );
   X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K3 ), m512_zero );

   P3 = _mm512_xor_si512( P3, X );

-   K4 = _mm512_xor_si512( mm512_ror128_32(
+   K4 = _mm512_xor_si512( mm512_shuflr128_32(
 			             _mm512_aesenc_epi128( K4, m512_zero ) ), K3 );
   X = _mm512_aesenc_epi128( _mm512_xor_si512( P2, K4 ), m512_zero );
-   K5 = _mm512_xor_si512( mm512_ror128_32(
+   K5 = _mm512_xor_si512( mm512_shuflr128_32(
 			             _mm512_aesenc_epi128( K5, m512_zero ) ), K4 );
   X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K5 ), m512_zero );

-   K6 = mm512_ror128_32( _mm512_aesenc_epi128( K6, m512_zero ) );
+   K6 = mm512_shuflr128_32( _mm512_aesenc_epi128( K6, m512_zero ) );
   K6 = _mm512_xor_si512( K6, _mm512_xor_si512( K5, _mm512_set4_epi32(
 	       ~ctx->count2, ctx->count3, ctx->count0, ctx->count1 ) ) );

   X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K6 ), m512_zero );
-   K7= _mm512_xor_si512( mm512_ror128_32(
+   K7= _mm512_xor_si512( mm512_shuflr128_32(
 			             _mm512_aesenc_epi128( K7, m512_zero ) ), K6 );
   X = _mm512_aesenc_epi128( _mm512_xor_si512( X, K7 ), m512_zero );

@@ -292,7 +294,7 @@ void shavite512_4way_close( shavite512_4way_context *ctx, void *dst )
    uint32_t vp = ctx->ptr>>6;

    // Terminating byte then zero pad
-    casti_m512i( buf, vp++ ) = m512_const2_64( 0, 0x0000000000000080 );
+    casti_m512i( buf, vp++ ) = m512_const1_i128( 0x0000000000000080 );

    // Zero pad full vectors up to count
    for ( ; vp < 6; vp++ )      
@@ -372,13 +374,13 @@ void shavite512_4way_update_close( shavite512_4way_context *ctx, void *dst,

   if ( vp == 0 )    // empty buf, xevan.
   { 
-      casti_m512i( buf, 0 ) = m512_const2_64( 0, 0x0000000000000080 );
+      casti_m512i( buf, 0 ) = m512_const1_i128( 0x0000000000000080 );
      memset_zero_512( (__m512i*)buf + 1, 5 );
      ctx->count0 = ctx->count1 = ctx->count2 = ctx->count3 = 0;
   }
   else     // half full buf, everyone else.
   {
-    casti_m512i( buf, vp++ ) = m512_const2_64( 0, 0x0000000000000080 );
+    casti_m512i( buf, vp++ ) = m512_const1_i128( 0x0000000000000080 );
      memset_zero_512( (__m512i*)buf + vp, 6 - vp );
   }

@@ -463,13 +465,13 @@ void shavite512_4way_full( shavite512_4way_context *ctx, void *dst,

   if ( vp == 0 )    // empty buf, xevan.
   {
-      casti_m512i( buf, 0 ) = m512_const2_64( 0, 0x0000000000000080 );
+      casti_m512i( buf, 0 ) = m512_const1_i128( 0x0000000000000080 );
      memset_zero_512( (__m512i*)buf + 1, 5 );
      ctx->count0 = ctx->count1 = ctx->count2 = ctx->count3 = 0;
   }
   else     // half full buf, everyone else.
   {
-    casti_m512i( buf, vp++ ) = m512_const2_64( 0, 0x0000000000000080 );
+    casti_m512i( buf, vp++ ) = m512_const1_i128( 0x0000000000000080 );
      memset_zero_512( (__m512i*)buf + vp, 6 - vp );
   }

--- a/algo/shavite/sph-shavite-aesni.c
+++ b/algo/shavite/sph-shavite-aesni.c
@@ -74,15 +74,15 @@ static const sph_u32 IV512[] = {

 #endif

+/*
 #if defined(__AVX2__)
 // 2 way version of above
 // a[7:0] = { b[4], a[7], a[6], a[5], b[0], a[3], a[2], a[1] }
-
 #define mm256_ror2x256hi_1x32( a, b ) \
   _mm256_blend_epi32( mm256_ror256_1x32( a ), \
                       mm256_rol256_3x32( b ), 0x88 )
-
 #endif
+*/

 static void
 c512( sph_shavite_big_context *sc, const void *msg )
@@ -101,15 +101,6 @@ c512( sph_shavite_big_context *sc, const void *msg )

   // round

-//  working proof of concept   
-/*
-   __m512i K = m512_const1_128( m[0] );
-   __m512i X = _mm512_xor_si512( m512_const1_128( p1 ), K );
-   X = _mm512_aesenc_epi128( X, m512_zero );
-   k00 = _mm512_castsi512_si128( K );
-   x = _mm512_castsi512_si128( X );
-*/
-
   k00 = m[0];
   x = _mm_xor_si128( p1, k00 );
   x = _mm_aesenc_si128( x, zero );
@@ -144,7 +135,7 @@ c512( sph_shavite_big_context *sc, const void *msg )
   for ( r = 0; r < 3; r ++ )
   {
      // round 1, 5, 9
-      k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, zero ) );
+      k00 = mm128_shuflr_32( _mm_aesenc_si128( k00, zero ) );
      k00 = _mm_xor_si128( k00, k13 ); 

      if ( r == 0 )
@@ -153,7 +144,7 @@ c512( sph_shavite_big_context *sc, const void *msg )

      x = _mm_xor_si128( p0, k00 );
      x = _mm_aesenc_si128( x, zero );
-      k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, zero ) );
+      k01 = mm128_shuflr_32( _mm_aesenc_si128( k01, zero ) );
      k01 = _mm_xor_si128( k01, k00 );

      if ( r == 1 )
@@ -162,31 +153,31 @@ c512( sph_shavite_big_context *sc, const void *msg )

      x = _mm_xor_si128( x, k01 );
      x = _mm_aesenc_si128( x, zero );
-      k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, zero ) );
+      k02 = mm128_shuflr_32( _mm_aesenc_si128( k02, zero ) );
      k02 = _mm_xor_si128( k02, k01 );
      x = _mm_xor_si128( x, k02 );
      x = _mm_aesenc_si128( x, zero );
-      k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, zero ) );
+      k03 = mm128_shuflr_32( _mm_aesenc_si128( k03, zero ) );
      k03 = _mm_xor_si128( k03, k02 );
      x = _mm_xor_si128( x, k03 );
      x = _mm_aesenc_si128( x, zero );

      p3 = _mm_xor_si128( p3, x );

-      k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, zero ) );
+      k10 = mm128_shuflr_32( _mm_aesenc_si128( k10, zero ) );
      k10 = _mm_xor_si128( k10, k03 );

      x = _mm_xor_si128( p2, k10 );
      x = _mm_aesenc_si128( x, zero );
-      k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, zero ) );
+      k11 = mm128_shuflr_32( _mm_aesenc_si128( k11, zero ) );
      k11 = _mm_xor_si128( k11, k10 );
      x = _mm_xor_si128( x, k11 );
      x = _mm_aesenc_si128( x, zero );
-      k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, zero ) );
+      k12 = mm128_shuflr_32( _mm_aesenc_si128( k12, zero ) );
      k12 = _mm_xor_si128( k12, k11 );
      x = _mm_xor_si128( x, k12 );
      x = _mm_aesenc_si128( x, zero );
-      k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, zero ) );
+      k13 = mm128_shuflr_32( _mm_aesenc_si128( k13, zero ) );
      k13 = _mm_xor_si128( k13, k12 );

      if ( r == 2 )
@@ -231,38 +222,38 @@ c512( sph_shavite_big_context *sc, const void *msg )

      // round 3, 7, 11

-      k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, zero ) );
+      k00 = mm128_shuflr_32( _mm_aesenc_si128( k00, zero ) );
      k00 = _mm_xor_si128( k00, k13 );
      x = _mm_xor_si128( p2, k00 );
      x = _mm_aesenc_si128( x, zero );
-      k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, zero ) );
+      k01 = mm128_shuflr_32( _mm_aesenc_si128( k01, zero ) );
      k01 = _mm_xor_si128( k01, k00 );
      x = _mm_xor_si128( x, k01 );
      x = _mm_aesenc_si128( x, zero );
-      k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, zero ) );
+      k02 = mm128_shuflr_32( _mm_aesenc_si128( k02, zero ) );
      k02 = _mm_xor_si128( k02, k01 );
      x = _mm_xor_si128( x, k02 );
      x = _mm_aesenc_si128( x, zero );
-      k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, zero ) );
+      k03 = mm128_shuflr_32( _mm_aesenc_si128( k03, zero ) );
      k03 = _mm_xor_si128( k03, k02 );
      x = _mm_xor_si128( x, k03 );
      x = _mm_aesenc_si128( x, zero );

      p1 = _mm_xor_si128( p1, x );

-      k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, zero ) );
+      k10 = mm128_shuflr_32( _mm_aesenc_si128( k10, zero ) );
      k10 = _mm_xor_si128( k10, k03 );
      x = _mm_xor_si128( p0, k10 );
      x = _mm_aesenc_si128( x, zero );
-      k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, zero ) );
+      k11 = mm128_shuflr_32( _mm_aesenc_si128( k11, zero ) );
      k11 = _mm_xor_si128( k11, k10 );
      x = _mm_xor_si128( x, k11 );
      x = _mm_aesenc_si128( x, zero );
-      k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, zero ) );
+      k12 = mm128_shuflr_32( _mm_aesenc_si128( k12, zero ) );
      k12 = _mm_xor_si128( k12, k11 );
      x = _mm_xor_si128( x, k12 );
      x = _mm_aesenc_si128( x, zero );
-      k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, zero ) );
+      k13 = mm128_shuflr_32( _mm_aesenc_si128( k13, zero ) );
      k13 = _mm_xor_si128( k13, k12 );
      x = _mm_xor_si128( x, k13 );
      x = _mm_aesenc_si128( x, zero );
@@ -304,39 +295,39 @@ c512( sph_shavite_big_context *sc, const void *msg )

   // round 13

-   k00 = mm128_ror_1x32( _mm_aesenc_si128( k00, zero ) );
+   k00 = mm128_shuflr_32( _mm_aesenc_si128( k00, zero ) );
   k00 = _mm_xor_si128( k00, k13 );
   x = _mm_xor_si128( p0, k00 );
   x = _mm_aesenc_si128( x, zero );
-   k01 = mm128_ror_1x32( _mm_aesenc_si128( k01, zero ) ); 
+   k01 = mm128_shuflr_32( _mm_aesenc_si128( k01, zero ) ); 
   k01 = _mm_xor_si128( k01, k00 );
   x = _mm_xor_si128( x, k01 );
   x = _mm_aesenc_si128( x, zero );
-   k02 = mm128_ror_1x32( _mm_aesenc_si128( k02, zero ) );
+   k02 = mm128_shuflr_32( _mm_aesenc_si128( k02, zero ) );
   k02 = _mm_xor_si128( k02, k01 );
   x = _mm_xor_si128( x, k02 );
   x = _mm_aesenc_si128( x, zero );
-   k03 = mm128_ror_1x32( _mm_aesenc_si128( k03, zero ) );
+   k03 = mm128_shuflr_32( _mm_aesenc_si128( k03, zero ) );
   k03 = _mm_xor_si128( k03, k02 );
   x = _mm_xor_si128( x, k03 );
   x = _mm_aesenc_si128( x, zero );

   p3 = _mm_xor_si128( p3, x );

-   k10 = mm128_ror_1x32( _mm_aesenc_si128( k10, zero ) );
+   k10 = mm128_shuflr_32( _mm_aesenc_si128( k10, zero ) );
   k10 = _mm_xor_si128( k10, k03 );
   x = _mm_xor_si128( p2, k10 );
   x = _mm_aesenc_si128( x, zero );
-   k11 = mm128_ror_1x32( _mm_aesenc_si128( k11, zero ) );
+   k11 = mm128_shuflr_32( _mm_aesenc_si128( k11, zero ) );
   k11 = _mm_xor_si128( k11, k10 );
   x = _mm_xor_si128( x, k11 );
   x = _mm_aesenc_si128( x, zero );
-   k12 = mm128_ror_1x32( _mm_aesenc_si128( k12, zero ) );
+   k12 = mm128_shuflr_32( _mm_aesenc_si128( k12, zero ) );
   k12 = _mm_xor_si128( k12, _mm_xor_si128( k11, _mm_set_epi32(
               ~sc->count2, sc->count3, sc->count0, sc->count1 ) ) );
   x = _mm_xor_si128( x, k12 );
   x = _mm_aesenc_si128( x, zero );
-   k13 = mm128_ror_1x32( _mm_aesenc_si128( k13, zero ) );
+   k13 = mm128_shuflr_32( _mm_aesenc_si128( k13, zero ) );
   k13 = _mm_xor_si128( k13, k12 );
   x = _mm_xor_si128( x, k13 );
   x = _mm_aesenc_si128( x, zero );
--- a/algo/simd/simd-hash-2way.c
+++ b/algo/simd/simd-hash-2way.c
@@ -747,11 +747,6 @@ void rounds512_4way( uint32_t *state, const uint8_t *msg, uint16_t *fft )

  static const m512_v16 code[] = { c1_16_512(185), c1_16_512(233) };

-
-//  static const m512_v16 code[] = { c1_16(185), c1_16(233),
-//                                   c1_16(185), c1_16(233) };
-
-
  S0l = _mm512_xor_si512( S[0], M[0] );
  S0h = _mm512_xor_si512( S[1], M[1] );
  S1l = _mm512_xor_si512( S[2], M[2] );
@@ -764,11 +759,16 @@ void rounds512_4way( uint32_t *state, const uint8_t *msg, uint16_t *fft )
 // targetted, local macros don't need a unique name
 #define S(i) S##i

+#define F_0( B, C, D ) _mm512_ternarylogic_epi32( B, C, D, 0xca )
+#define F_1( B, C, D ) _mm512_ternarylogic_epi32( B, C, D, 0xe8 )  
+
+/*  
 #define F_0(B, C, D) \
   _mm512_xor_si512( _mm512_and_si512( _mm512_xor_si512( C,D ), B ), D )
 #define F_1(B, C, D) \
   _mm512_or_si512( _mm512_and_si512( D, C ),\
                    _mm512_and_si512( _mm512_or_si512( D,C ), B ) )
+*/

 #define Fl(a,b,c,fun) F_##fun (a##l,b##l,c##l)
 #define Fh(a,b,c,fun) F_##fun (a##h,b##h,c##h)
--- a/algo/simd/vector.c
+++ b/algo/simd/vector.c
@@ -6,10 +6,6 @@

 #define PRINT_SOME 0

-/* JDD all ocurrances of macro X in this file renamed to XX
- * due to name conflict
- */
-
 int SupportedLength(int hashbitlen) {
  if (hashbitlen <= 0 || hashbitlen > 512)
    return 0;
--- a/algo/skein/skein-4way.c
+++ b/algo/skein/skein-4way.c
@@ -2,14 +2,8 @@
 #include <string.h>
 #include <stdint.h>
 #include "skein-hash-4way.h"
-
-// 8 way is faster than SHA on Icelake
-// SHA is faster than 4 way on Ryzen
-//
-#if defined(__SHA__)
-  #include <openssl/sha.h>
-#endif
 #include "algo/sha/sha-hash-4way.h"
+#include "algo/sha/sha256-hash.h"

 #if defined (SKEIN_8WAY)

@@ -93,7 +87,6 @@ void skeinhash_4way( void *state, const void *input )
     uint32_t hash1[16] __attribute__ ((aligned (64)));
     uint32_t hash2[16] __attribute__ ((aligned (64)));
     uint32_t hash3[16] __attribute__ ((aligned (64)));
-     SHA256_CTX ctx_sha256;
 #else
     uint32_t vhash32[16*4] __attribute__ ((aligned (64)));
     sha256_4way_context ctx_sha256;
@@ -102,31 +95,23 @@ void skeinhash_4way( void *state, const void *input )
     skein512_4way_final16( &ctx_skein, vhash64, input + (64*4) );

 #if defined(__SHA__)      
+
     dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 512 );

-     SHA256_Init( &ctx_sha256 );
-     SHA256_Update( &ctx_sha256, (unsigned char*)hash0, 64 );
-     SHA256_Final( (unsigned char*)hash0, &ctx_sha256 );
-
-     SHA256_Init( &ctx_sha256 );
-     SHA256_Update( &ctx_sha256, (unsigned char*)hash1, 64 );
-     SHA256_Final( (unsigned char*)hash1, &ctx_sha256 );
-
-     SHA256_Init( &ctx_sha256 );
-     SHA256_Update( &ctx_sha256, (unsigned char*)hash2, 64 );
-     SHA256_Final( (unsigned char*)hash2, &ctx_sha256 );
-
-     SHA256_Init( &ctx_sha256 );
-     SHA256_Update( &ctx_sha256, (unsigned char*)hash3, 64 );
-     SHA256_Final( (unsigned char*)hash3, &ctx_sha256 );
-
+     sha256_full( hash0, hash0, 64 );
+     sha256_full( hash1, hash1, 64 );
+     sha256_full( hash2, hash2, 64 );
+     sha256_full( hash3, hash3, 64 );
+    
     intrlv_4x32( state, hash0, hash1, hash2, hash3, 256 );
-#else
-     rintrlv_4x64_4x32( vhash32, vhash64, 512 );

+#else
+
+     rintrlv_4x64_4x32( vhash32, vhash64, 512 );
     sha256_4way_init( &ctx_sha256 );
     sha256_4way_update( &ctx_sha256, vhash32, 64 );
     sha256_4way_close( &ctx_sha256, state );
+
 #endif
 }

--- a/algo/skein/skein-hash-4way.c
+++ b/algo/skein/skein-hash-4way.c
@@ -309,22 +309,16 @@ static const uint64_t IV512[] = {
      sc->bcount = bcount; \
   } while (0)
   
-// AVX2 all scalar vars are now vectors representing 4 nonces in parallel
-

 #if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

 #define TFBIG_KINIT_8WAY( k0, k1, k2, k3, k4, k5, k6, k7, k8, t0, t1, t2 ) \
 do { \
-  k8 = _mm512_xor_si512( _mm512_xor_si512( \
-                            _mm512_xor_si512( _mm512_xor_si512( k0, k1 ), \
-                                              _mm512_xor_si512( k2, k3 ) ), \
-                            _mm512_xor_si512( _mm512_xor_si512( k4, k5 ), \
-                                              _mm512_xor_si512( k6, k7 ) ) ), \
-                         m512_const1_64( 0x1BD11BDAA9FC1A22) ); \
+  k8 = mm512_xor3( mm512_xor3( k0, k1, k2 ), mm512_xor3( k3, k4, k5 ), \
+                   mm512_xor3( k6, k7, m512_const1_64( 0x1BD11BDAA9FC1A22) ));\
  t2 = t0 ^ t1; \
 } while (0)
-   
+
 #define TFBIG_ADDKEY_8WAY(w0, w1, w2, w3, w4, w5, w6, w7, k, t, s) \
 do { \
  w0 = _mm512_add_epi64( w0, SKBI(k,s,0) ); \
@@ -340,7 +334,6 @@ do { \
                                         m512_const1_64( s ) ) ); \
 } while (0)

-
 #define TFBIG_MIX_8WAY(x0, x1, rc) \
 do { \
     x0 = _mm512_add_epi64( x0, x1 ); \
--- a/algo/skein/skein.c
+++ b/algo/skein/skein.c
@@ -5,21 +5,18 @@
 #include <string.h>
 #include <stdint.h>
 #include "sph_skein.h"
-#include <openssl/sha.h>
+#include "algo/sha/sha256-hash.h"

 void skeinhash(void *state, const void *input)
 {
     uint32_t hash[16] __attribute__ ((aligned (64)));
     sph_skein512_context ctx_skein;
-     SHA256_CTX           ctx_sha256;

     sph_skein512_init( &ctx_skein );
     sph_skein512( &ctx_skein, input, 80 );
     sph_skein512_close( &ctx_skein, hash );

-     SHA256_Init( &ctx_sha256 );
-     SHA256_Update( &ctx_sha256, (unsigned char*)hash, 64 );
-     SHA256_Final( (unsigned char*) hash, &ctx_sha256 );
+     sha256_full( hash, hash, 64 );

     memcpy(state, hash, 32);
 }
@@ -27,8 +24,8 @@ void skeinhash(void *state, const void *input)
 int scanhash_skein( struct work *work, uint32_t max_nonce,
                    uint64_t *hashes_done, struct thr_info *mythr )
 {
-        uint32_t *pdata = work->data;
-        uint32_t *ptarget = work->target;
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
 	uint32_t hash64[8] __attribute__ ((aligned (64)));
 	uint32_t endiandata[20] __attribute__ ((aligned (64)));
 	const uint32_t Htarg = ptarget[7];
@@ -36,7 +33,7 @@ int scanhash_skein( struct work *work, uint32_t max_nonce,
 	uint32_t n = first_nonce;
   int thr_id = mythr->id;  // thr_id arg is deprecated

-        swab32_array( endiandata, pdata, 20 );
+   swab32_array( endiandata, pdata, 20 );

 	do {
 		be32enc(&endiandata[19], n); 
--- a/algo/swifftx/inttypes.h
+++ b/algo/swifftx/inttypes.h
@@ -18,16 +18,20 @@
 #ifndef __INTTYPES_H_
 #define __INTTYPES_H_

+#include <stdint.h>
+
 /* Use [u]intN_t if you need exactly N bits.
  XXX - doesn't handle the -mint8 option.  */

 typedef signed char swift_int8_t;
 typedef unsigned char swift_uint8_t;

- typedef int swift_int16_t;
+ typedef int32_t swift_int16_t;
+// typedef int swift_int16_t;
 typedef unsigned int swift_uint16_t;

- typedef long swift_int32_t;
+ typedef int32_t swift_int32_t;
+// typedef long swift_int32_t;
 typedef unsigned long swift_uint32_t;

 typedef long long swift_int64_t;
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Jay D Dee	47cc5dcff5	v3.18.1	2021-10-10 22:50:19 -04:00
Jay D Dee	2cd1507c2e	v3.7.4	2021-09-29 17:31:16 -04:00
Jay D Dee	9b905fccc8	v3.17.1	2021-07-26 15:01:37 -04:00
Jay D Dee	92b3733925	v3.17.0	2021-07-15 20:30:44 -04:00
Jay D Dee	19cc88d102	v3.16.5	2021-06-26 12:27:44 -04:00
Jay D Dee	a053690170	v3.16.4	2021-06-23 21:52:42 -04:00
Jay D Dee	3c5e8921b7	v3.16.3	2021-05-06 14:55:03 -04:00
Jay D Dee	f3333b0070	v3.16.2	2021-04-08 18:09:31 -04:00
Jay D Dee	902ec046dd	v3.16.1	2021-03-24 18:24:20 -04:00
Jay D Dee	d0b4941321	v3.16.0	2021-03-19 15:45:32 -04:00
Jay D Dee	40089428c5	v3.15.7	2021-03-08 22:44:44 -05:00
Jay D Dee	dc6b007a18	v3.15.6	2021-02-12 15:16:53 -05:00
Jay D Dee	06bfaa1249	v3.15.5	2020-12-21 13:25:33 -05:00
Jay D Dee	6566e99a13	v3.15.4	2020-12-15 13:15:02 -05:00
Jay D Dee	ccfccbadd5	v3.15.3	2020-12-10 18:23:49 -05:00
Jay D Dee	45ecd0de14	v3.15.2	2020-11-15 17:57:06 -05:00
Jay D Dee	4fa8fcea8b	v3.15.1	2020-11-09 13:19:05 -05:00